RSX not fast enough yet [Archive]

Pumpkin Head

06-04-2006, 10:55 PM

According to them, the core runs at 420MHZ atm, while the memory runs at 600MHZ. Final speeds are 550 for the core and 700 for the memory.

Quote:
SOME VERY INTERESTING tidbits have emerged about the PS3 GPU during my flight to Japan yesterday. It seems like the second gen dev kits were running nowhere close to full speed or spec. How close were they? Fairly.

Documents shown to us on the trip showed that there were two versions, labeled 'CEB' and 'DEH'. The earlier CEB has a chip called CBE (yes, I know) connected to main memory and to a south bridge. The SB is then connected to a chip labled 'NV4x' over a PCIe link. The upcoming version DEH has the CBE and main memory, but the NV4x/SB combo is replaced by the RSX. If you wanted to know what makes RSX more than a GPU, it looks like it is SB functionality.

The slightly more curious bit, encouraging and disturbing at the same time, is the current state of RSX. The disturbing part is that the slide I was shown had "Current DEH's aren't final spec or speed" in bold letters. Speed, OK, but not final spec at this point in time leaves precious little room for debugging before the console release. On a different note the current ones are running the RSX core at 420MHz with 550 expected for launch. Memory is set at 600MHz with 700 hoped for as final.

It looks like SoNVidia is most of the way there with RSX development, 80% of the clock, more for memory. One more spin, most likely another round of dev kits, and the code monkeys will know what they are facing.

Link: http://theinquirer.net/?article=32159

This actually coincides with information I have come across since E3. Apparently, the rack-mount kits will have full-spec hardware but the kits that are coming out now are not full-spec.

So far, everything we have seen is still early build gamewise and hardware-wise. This is one of the reasons why Sony was rather quiet at E3. They didnt want to be caught with their pants down.

At the same time, this is good news for the games. Assassin's Creed, FFXIII, and even Resistence still have a ways to go in development and they do not have full production cycles on final hardware. Yet, they look just as good as second gen 360 games.. This usually flies under radars, but the truth is that the final PS3 will be a real eye opener. Just wait til the TGS to see what the final hardware produces

yoshaw

06-04-2006, 10:59 PM

http://xs301.xs.to/xs301/06220/ohnozomg.gif

It's the inquirer so :whogives: Man these guys are good at making stories up. Don't listen to them. But if its true, more power to you(none to Inq)

masteratt

06-04-2006, 11:02 PM

They are [the inquirer] especially Anti-PS3.

Mate the best thing you can do is only disgust news officially coming from Sony's mouth.

The whole Media either in a subtle or a non-subtle way is anti-Sony now. Why? Well, who wants to read "Stop everything! Sony releases the best console once again." That's not exciting.
It's their job to create bullshit so people actually get involved in the drama that isn't even there.

xbdestroya

06-04-2006, 11:02 PM

Ok the 'thumbs up' throws me off, but whatever. ;)

Anyway some interesting info in that article and some Inquirer-esque lapses at the same time.

In the end I'm not bothered that the RSX isn't up to speed yet - keeps both an upgrade (and a downgrade) in play. As for the memory, well whatever, no issue there.

I did find the RSX = SouthBridge thing interesting though. We already knew it served for some of that functionality, but anything that adds novelty to the chip is ok with me.

Personally I think that article is nearly 100% factual though, so just to give it soem legitimacy here since we tend to be very anti-Inquirer on the whole.

CrumCon

06-04-2006, 11:08 PM

unless they could achive killzone E3 trailer nd killzone E3 trailer with 120mhz more speed for RSX.. i dont a fuk.

they just lied to us.. right in our face.

makeitlookreal

06-05-2006, 12:06 AM

Basically, these are more dribbles of information. What we really need is for Sony to be honest with us.

Obviously, their competitors already know these things. I am certain that MS and Nintendo know all there is to know about every kit so far. We are the ones in the dark.

For goodness sakes, what will it hurt for them to let us know what is going on with these kits!

Unless there is MORE going on that no one dares speak about.

Infernal

06-05-2006, 12:11 AM

For goodness sakes, what will it hurt for them to let us know what is going on with these kits!
I think the question is. What will it help them to let us know what is going on with these kits?

Pumpkin Head

06-05-2006, 01:19 AM

{Delta}

06-05-2006, 01:41 AM

For being a rumor site, many tidbits of real information come from them long before the mass news story everyone else has is released/

Regardless, it will not matter. When the RSX finally releases...We are going to have G80 and R600 silicon that surpases it...So I believe the heart it going to be in the games rather than hybrid G70 with a 128 bit memory bus pushing more polygons than another.

tech superiority is extreamly short lived.

gljvd

06-05-2006, 02:46 AM

well the older dev units were huge and had better cooling . So perhaps the smaller case is worse

venomv

06-05-2006, 03:17 AM

@gljvd The older dev kits were extremely loud from what I have heard, but it seems they have fixed that problem though.

pari

06-05-2006, 04:57 AM

--------------------------------------------------------------------------------
Sony still hasn't thrown the switch on final hardware because 65nm technology is going to roll out and drop production costs and hardware reliability.

That's kindof what I'm leaning towards now. I dont think they're making a lot of final kits yet because they want to get final devkits and final PS3s off the same assembly lines to save costs.

Again, this is just a suspicion, I do not have a source that coordinates this assumption

65nm technology would reduce the cost but not lower the hardware reliability. Interesting, I have not seen any paper on the RSX, might be I have not actively looked for it... I do not remember reading any of Nvidia stuff on the IEEE journal either... need to check. Normally Sony puts a paper about their design. PSP processor paper was interesting for the low power design techniques..

Pumpkin Head

06-05-2006, 05:47 AM

The only thing i do know that the final ps3 hardware is going to suprise us all.
In an interview held with Epic Games Mark Rein, he noted, that UT2007 is running on a very early PS3 hardware on unoptimised build at 50 FPS.
He said the final PS3 is going to surprise everyone, and it's a monster of a machine.
He added, that until now PC gamers had the upper hand in the graphics but now with the power of the PS3 it won't be the matter this time, and the fact that on the console you don't need to worry about the OS to take the power from the games is even better.

Sounds good to me, it's not coming from a biased source

Pumpkin Head

06-05-2006, 05:52 AM

I wish I knew what the RSX was going to be. Sony has been dead quiet on the RSX, which at this point, just doesnt make sense seeing as how the games are shaping up very well.

It may still very well have to do with the orgiinal design of the PS3 and the nature of other Cell based servers. I do not care to elaborate at this time, because it is mostly speculation.

All I can tell ya is to stay tuned and lets see what the mad scientists over at Sonys hardware labs are cooking up

LiquidEagle

06-05-2006, 06:01 AM

I don't understand. I thought the RSX taping out like back in March or April meant that they were 100% ready to start producing the thing? Why would they still be having issues?

makeitlookreal

06-05-2006, 06:34 AM

Lets get one point clear.

From what we know RIGHT NOW the RSX is basically an off the shelf PC part. If Sony had to spend lots of time and money working with NVIDIA to design the RSX I think the current information about the RSX is basically a cover for what is really under the hood.

If they went to all the trouble they claim they have then the RSX must be something much more than just a 7800 or 7900. Also, it must be more than what we know right now boosted by a 100mhz.

There must be something about this chip that sets it apart from the PC component. Yes, we know that because of the Flex IO the RSX can work well with the cell chip. But I think that has been pretty much common knowledge for a while now.

What else is going on? I sure wish Sony would tell us!

By the way, it might not make Sony one more cent if they told us the full specs of the RSX and were honest with us about the PS3. But we are the consumers who are supposed to aways be at a peaceful WAR against the producers for the best prices and quality. We have clout because we keep Sony in business. They don't have to tell us anything, but if we are to be good consumers we should use out clout to peacefully force them to tell us the truth. NOT by any illegal means or anything violent of course, but simply by withholding our dollars and not buying their products.

Seriously, I don't think we know hardly anything about the RSX yet, and Sony needs to tell us what we are going to be spending our hard earned dollars on. If not, some people might see high quality games like Fight Night for the 360 and consider purchasing it instead.

PS: I also really hope they manage to utilize the 65NM technology at launch.

Nerve-Damage

06-05-2006, 06:57 AM

WTF??? (http://www.theinq.com/?article=32171)

http://www.theinq.com/images/articles/PS3_memory_bandwidths.jpg

AFTER BREAKING THE news to me about PS3 RSX speeds earlier on the flight to Japan, my row-mate said 'if you think that's interesting, wait till you see this. Cell is hurting, badly'.

For those of you that believe in religions with karmic tendencies, scoops like this meant one of two things, the wings of the plane are about to fall off and I am going to die in a fiery ball, or worse yet, the movie selection will be worrisome. Cell memory access appears to be broken, RSX has half the triangle setup rate of the ATI chip in XBox360, and the true horror, Big Momma's House 2 and a Queen Latifa movie.

With the movie selection still making my brain throb from the glances I caught, I furiously took notes on what the source was saying. He started out saying that the RSX can only write about half as much vertex data as it can fetch, not an ideal situation by any stretch, but survivable.

Then came the horrible news, RSX appears to be limited to setting up 275 Million triangles/second, anemic compared to the 500+ million in XBox360. When asked about this apparent thumping dished out by MS, the reply from one notable ISV relations boffin was a terse 'What a Piece of Junk'. Talk about a steak in the heart.

Half the triangle setup capability in the PS3, could things get worse? Yes, far far worse, how about another disparity of three orders of magnitude? No, I am not joking, looking at Sony's own figures, Cell appears to be pretty badly broken.

For main memory, it looks like Cell has about 25GBps of main memory bandwidth, and RSX is about 15-20GBps. Achievable bandwidth is between about two thirds of that and nearly 100%, clearly the elves in the caves surrounding Rambus central did something right with XDR. That is the happy news.

For local memory, the measured vs theoretical bandwidth is missing, I wonder why? RSX is at a solid 22.4GBps for both read and write, good job there green team. Then comes the blue team with Cell. Local memory write is about 4GBps, 40% of the next slowest bandwidth there. Then comes the bomb from hell, the Cell local memory read bandwidth is a stunning 16MBps, note that is a capital M to connote Mega vs a capital G to connote Giga. This is a three order of magnitude oopsie, and it is an oopsie, as Sony put it "(no, this isn't a typo...)".

If you can write at 250x the read speed, it makes Cell local memory just about useless. That means you do all your work out of main memory, and the whole point of local is, well, pointless. This can lead to contention issues for the main memory bus, and all sorts of nightmarish to debug performance problems. Basically, if this Sony presentation to PS3 devs shown to us is correct, it looks like PS3 will be hobbled in a serious way.

The next slide goes on to say "Don't read from local memory, but write to main memory with RSX(tm) and read it from there instead", and repeats the table numbers. This is very very bad. The number of times the presentation goes on to say that it is correct, and the lack of anything like "this will be fixed by production steppings, so take measures X, Y and Z" say to me that it is not a fixable snafu. Remember at E3 when I said that the PS3 demos there were object sparse? Any guesses why?

Someone screwed up so badly it looks like it will relegate the console to second place behind the 360. All the devs I talked to were lukewarm on the 360 architecture but universally negative on the PS3. Revelations like this go a long way to explain why you keep hearing about simmering problems from the Sony devs.

You end up with a console with half the triangle setup rate of the 360, a crippled CPU that is a bitch to program, and tools that are atrocious compared to the 360. To make matters worse, you have an arrogant set of execs telling us that twice the price is worth it for half the power, a year late. If it isn't already too late, Sony had better do something about this recto-cranial inversion or it may very well sink the console. µ

LaLiLuLeLo

06-05-2006, 07:02 AM

Um, that sounds like a load of crap (the article).
The article and the photo are what, conflicting?
I'm confused.

makeitlookreal

06-05-2006, 07:10 AM

Basically, that article sounds really, really bad for the PS3. Of course hopefully the information can be verified.... OOOPS! Sony does not wish to give us any specifications these days on the RSX and Cell!

So for all we know that article could be right on the money! Sony sure is not dying it or any others by posting specifications!

I don't know if that article is legitimate or not. Hopefully, someone here with a good knowledge of such issues will comment. But all of these issues could be solved once and for all if Sony would be honest with us and release the FULL SPECIFICATIONS of the Cell and RSX.

You gonna let this stand, Sony?

Nameless

06-05-2006, 07:12 AM

I'm having a hard time believing this article...
(Perhaps the info is valid, but presented in a misleading manner...)
Several 3rd party and of course 1st party developers do not feel hindered by the PS3 architecture and stated the machine will allow them creative freedom. (i.e Ubisoft, Square, Konami, etc.)If the Cell was this broken from a data transport perspective we would know at this point. Also, how are games so early in development looking so visually impressive... Only time will tell.

Nerve-Damage

06-05-2006, 07:25 AM

Acert93 clearing things up a bit... (http://www.beyond3d.com/forum/showthread.php?p=770828#post770828)

stanDarsh

06-05-2006, 07:29 AM

It's from the Inquirer, move along nothing to see here.

Nameless

06-05-2006, 07:39 AM

If you would like to read more about the cell and have some time to kill review this link:

http://en.wikipedia.org/wiki/Cell_(microprocessor)

I'm sure most of you have stumbled upon this info, but I added it just in case...

makeitlookreal

06-05-2006, 07:44 AM

yoshaw

06-05-2006, 07:46 AM

It seems like Inquirer is running short on money. No wonder a story like this would generate mega attention and in return lots of ads/money for the hits.

I highly suggest the mods to boycott(or delete) any links that are supplied on the forums for the website of Inquirer. This way, they won't get any hits that they so direly wish for with these half-assed made up sensationalist kind of stories against a franchise that has over 2 billion customers. It makes perfect sense for a tabloid site like Inquirer to stir things up to generate traffic from an endless audience from PS platform in particular.

Let us be the first ones to boycott their BS FUD spreading stories. Ridicule their stories by posting in a thread, sure np. But remove any links so they don't get any free hits, that's all I'm saying. How's that?

Nameless

06-05-2006, 07:50 AM

It seems like Inquirer is running short on money. No wonder a story like this would generate mega attention and in return lots of ads/money for the hits.

I highly suggest the mods to boycott(or delete) any links that are supplied on the forums for the website of Inquirer. This way, they won't get any hits that they so direly wish for with these half-assed made up sensationalist kind of stories against a franchise that has over 2 billion customers. It makes perfect sense for a tabloid site like Inquirer to stir things up to generate traffic from that audience in particular.

Let us be the first ones to boycott their BS FUD spreading stories. Ridicule their stories by posting in a thread, sure np. But remove any links so they don't get any free hits, that's all I'm saying. How's that?
Nah I disagree, news is news even if the source is questionable...
Usually most rumors contain a shred of truth...

I do believe this article is somewhat misleading, I would like to hear a developers response regarding the architecture. (Most of the thoughts lately from the development community has been overall positive for the PS3...)

Raijin

06-05-2006, 07:58 AM

This slide comes from the Devstation and at that time neither RSX and Cell were finished. So I suspect that the spec shown here are only from the unfinished components.

Voidler

06-05-2006, 08:11 AM

Anyone in a good enough position to be getting real info like this is also in a good enough position to not be making such article at the Inquirer

casualkiss

06-05-2006, 08:25 AM

On beyond, they are saying that the 16MB is not reference to the SPE's local store memory but to the RSX's GDDR3 memory and the work-around is to have the RSX fetch the info from GDDR3 and send it to the Cell when needed... I think.

Kabbage

06-05-2006, 11:53 AM

No, there is a lot to read about.

That Beyond 3D link did not really confirm or deny the 16MB/S bit of information, but no one said they thought it was not true.

If it is true, it sure does seem like the PS3's design is horrid. Not that developers can't work out it to get good results, but that someone left a major design flaw.

How do you figure....Acert wasnt trying to deny the 16MB rumor... he was saying there likely reason it was low.

Id really doubt they would overlook such a shortcoming... call it an architectual design/tradeoff rather than a flaw.

Kabbage

06-05-2006, 11:58 AM

Just as the price bitching starts to die down... looks as if INQ wants to give people more to compain about PS3 wise

P.S

The games look great even unfinished... I couldnt care less how fast the Cell reads/writes from the GDDR3 or how many triangles RSX sets up.

Since im not developing for it LOL.

When the INQ starts posting about bad broken games because of it... then let me know

overclocked

06-05-2006, 12:30 PM

A single slide can cause a lot of confusion esp when the people that knows nothing screams the loudest on the net.
So thumbs up for The inq and Charlie D for some good marketing and pagehits. :smoke:

Quite funny you can reverse it to something positive with his other statement about clocks that they have a complete different design now than when RSX taped out in jan/feb, the bigger/better chip is causing a downgrade in clockspeed now because of the time involved with G80 based RSX instead. :stirpot:

Kabbage

06-05-2006, 12:38 PM

Hell at first... according to early dev kit diagrams, I didnt think cell could write to the GDDR at all...OMG 0MB/s read and write :D

Kabbage

06-05-2006, 12:40 PM

jaxmkii

06-05-2006, 12:48 PM

For being a rumor site, many tidbits of real information come from them long before the mass news story everyone else has is released/
. unfortionatly for every one they get right they get 13 wrong

version

06-05-2006, 01:37 PM

if cell read/write to/from gddr, need to writeback every L1,L2 tilecaches on rsx,

Kabbage

06-05-2006, 02:16 PM

if cell read/write to/from gddr, need to writeback every L1,L2 tilecaches on rsx,

That kinda seems like an incomplete thought to me :Mrt:

cybergrue

06-05-2006, 02:49 PM

WTF??? (http://www.theinq.com/?article=32171)
http://www.theinq.com/images/articles/PS3_memory_bandwidths.jpg

Granted, the photo is not high enough quality to tell for sure, but notice the purple poka-dots over the top of the slide, and more importantly, notice how they end in a nice white box around the 16M (no this not a typo) cell of the table. The dots at the top side of the box look cut in half, yet continue on in the cell below it.

As I said, this piture is not high enough quality to tell for sure, but I am suspicious. Also, the 3 orders of magniture difference between read and write is simply unbelievable in of itself. Very suspicious.

Does anyone proper confirmation of this info (another pic from another angle? or something from another source?)

makeitlookreal

06-05-2006, 03:47 PM

I have been looking online for that slide and cannot find it.

Basically, if that slide is legitimate and the read speed is 16mb/s that is absolutely HORRIBLE and truly a disaster! Yeah, there might be ways to work around it and still get very good performance. But it is downright stupid to design a system and have such a HUGE flaw.

Remember, were not talking about only 16GB/S or even 1600MB/S or the very low 160MB/S.

We are talking about 16MB/S which is simply HORRENDOUS BEYOND MEASURE!

I think we need some confirmation on this and quickly.

If this slide was faked then whoever did so owes Sony a HUGE apology, because they are accusing Sony of designing a gaming console with the mother of all flaws.

Red_Eyes

06-05-2006, 03:50 PM

The thing is, other sites are not saying anything. How come only the Inquirer was at the meeting and no other journalists?

version

06-05-2006, 03:54 PM

yes tilecaches bad for cell, but the 1 terabyte/sec rendering speed good for rsx , or not?

Red_Eyes

06-05-2006, 03:59 PM

All I can say is this: "Don't read from local memory, but write to main memory with RSX(tm) and read it from there instead" because it is faster that way. And that's the whole point of that slide.

VG Aficionado

06-05-2006, 04:07 PM

In any case, which are the consequences of this new info if true? Because I certainly don't see how that will make PS3 half as powerful as 360 just like Inquirer's article states (from what we've seen so far).

Applefiend

06-05-2006, 04:08 PM

Why would you need to read video memory with the CPU anyway?

I understand why the GPU needs to read it, transparency, particle effects, etc, but the CPU? Why do you need to do such a thing?

I don't understand exactly how this makes Cell or PS3 crippled. When you do need to do such a thing... God knows why, you ask RSX to blast it into main memory for you. But really, why do that ever?

What I'd like confirmation of is the triangle rate, that sounds very bad. Would effectively mean PS3 games having half the frame rate or half the draw distance.

But coming from Inquirer... I think if there was a slide with "Half the triangle rate of 360". I'd be an upset puppy. But other heard bullshit from the Inquirer on the plane from japan... Who cares.

<*Uneasy grin*>

Nodieza

06-05-2006, 04:22 PM

"No, this isn't a typo..."

He's right, it's just complete BS, that wouldn't be professional at all to do that in a presentation, they are expecting everything you show to be truth, plus the triple dots afterwards makes it seems like Sony themselves are dissapointed with it or questioning it.

It's too fishy if you know what I mean.

EDIT:All this bad news yet Heavenly Sword still kicks so much butt and those devs love the PS3

The article itself is so blunt, fanboyish sounding, and shallow, no names "relations boffin"

Nobody would yell out 'what a piece of junk' why do we make threads with this crap, that's exaclty what that stupid website wants to happen.

These Inquirer articles are just annoying, they are worse than the Enquirer magazine "Osama has 5 alien babies who helped plot the attack of 9-11" WTF!!!!! This is so stupid....

cliffbo

06-05-2006, 04:26 PM

Applefiend

06-05-2006, 04:26 PM

I tell ya. I write OpenGL games for a living, not a very good one but...

There's one use I can think of for reading video memory into main memory.

Taking screenshots of games. That's it. Some of the gurus here will have to come up with others.

If you read the comments of the file you can tell what camera it was taken with btw. Time is 8:20pm 04/06/2006

makeitlookreal

06-05-2006, 04:27 PM

Well, both of these issues are very serious. If the Cell only has half the triangles then that is VERY bad news. Additionally, if the read rate is only 16MB/S that is HORRIBLE. I mean, that is just not low, but awful!

Seriously, the Inquierer better have backup for these statements, because they are very, very serious.

*IF* these are true, and we DO NOT KNOW, it means the PS3 is a very expensive 360 wannabe machine. I just hope we can get some verification of this.

Applefiend

06-05-2006, 04:29 PM

double post

cliffbo

06-05-2006, 04:30 PM

why the hell would they put 'no this isn't a typo' in a presentation like this. photoshop....

venomv

06-05-2006, 04:32 PM

MILR, look at what people are saying, there is almost no use for it so why should it be fast?

Viano

06-05-2006, 04:33 PM

photoshop loL....

Red_Eyes

06-05-2006, 04:36 PM

And the Inquirer is not being very professional too. Saying stuff like the PS3 has "a crippled CPU that is a bitch to program" for sounds very much like a fanboy.

VG Aficionado

06-05-2006, 04:49 PM

What I'd like confirmation of is the triangle rate, that sounds very bad. Would effectively mean PS3 games having half the frame rate or half the draw distance.

But coming from Inquirer... I think if there was a slide with "Half the triangle rate of 360". I'd be an upset puppy. But other heard bullshit from the Inquirer on the plane from japan... Who cares.I think they also used to say that 360's GPU had twice as many pixel pipelines as RSX, which implied performance was RSX x2. Too bad they forgot to mention unified pipelines are not as powerful as dedicated pipelines nor 360's GPU has 48 pipelines available for pixel shading all the time. Not to mention their statement is fundamentally flawed (did they assume that both GPUs work in the same conditions and in exact ways?).

Anyway, if there is so much truth about this info, I'm sure The Inquirer will not be the only source. What is more, it's extremely fishy that it's the only known one by now, and it turns out to be as bad or worse as their usual PS3 articles while "every developer" praises how 360 is twice as powerful and much easier to develop for.

xbdestroya

06-05-2006, 04:59 PM

Granted, the photo is not high enough quality to tell for sure, but notice the purple poka-dots over the top of the slide, and more importantly, notice how they end in a nice white box around the 16M (no this not a typo) cell of the table. The dots at the top side of the box look cut in half, yet continue on in the cell below it.

As I said, this piture is not high enough quality to tell for sure, but I am suspicious. Also, the 3 orders of magniture difference between read and write is simply unbelievable in of itself. Very suspicious.

Does anyone proper confirmation of this info (another pic from another angle? or something from another source?)

The picture is not a photoshop. But at the same time, the bandwidth concerns everyone is sounding alarms about... don't really matter for what we care about.

The photo is dated June 4th probably because Charlie (man but that guy sucks) took a picture with a camera of his traveling partners laptop screen.

These slides are confidential remember, it's just the Inquirers way of getting such information. That's why we all read them, even though their quality sucks - because they *do* break stories now and then.

Anyway but Charlie's all confused as to what he's seeing. This is really a non-issue here.

makeitlookreal

06-05-2006, 05:16 PM

Applefiend

06-05-2006, 05:18 PM

My googling isn't leaving me any wiser, just lots of guys on fanboy message boards who don't know the difference between a vertrex, a triangle, and a polygon. :)

But I like these two pages....

http://forums.xbox-scene.com/index.php?showtopic=231928
360 GPU Vertex rate : >.5 billion ish vertexes/sec

http://www.ps3source.net/content/view/17/33/
RSX Vertrex rate - 1.1 billion vertexes/sec

Even 1.1 divided by three is higher than the Inquirer nonsense :)

I smell bullshit. Big steaming piles of bullshit. Ain't worth me googlin' mate.

Yeah, maybe the banning of Inquirer articles would be a good idea. There's a huge industry in trash talking PS3 to get hits for web sites. I don't care for it personally.

xbdestroya

06-05-2006, 05:21 PM

If I remember correctly Sony told us the RSX and Cell could read data from each other's memory. This was a pretty important feature from what I remember, and they seemed to hype it. Now, the problem is if the 16MB/S is true you might as well say the Cell cannot read from the RSX, because the rate is so slow it would make the whole system shudder. Again, we are talking about 16MB/S.

Perhaps the PS3 will make great games without having the Cell read the RSX's memory, but the FACT of the matter is that it is supposed to do so, and at 16MB/S it might as well not even try!

RSX reading from XDR matters; Cell reading from the GDDR-3 hardly matters at all. I'd say it doesn't matter, save that who knows, maybe there's something it'd be useful for.

But the GPU is always going to be the more memory-starved component in this system.

makeitlookreal

06-05-2006, 05:39 PM

I just did some reading on FLASH memory and it seems that it is possible to read from flash memory much faster than a HDD. Perhaps this extra flash memory could be very useful in the PS3!

Could it be that the PS3 has such great streaming data because it can be cached not only on the HDD but also on the Flash memory?

Applefiend

06-05-2006, 05:43 PM

The flash is just where the operating system lives I think.

cliffbo

06-05-2006, 06:27 PM

RSX not fast enough? got one bullet in the chamber... goodbye cruel world...

PhYmon

06-05-2006, 06:44 PM

Im very confuse as well, could it be true? after seen the graphics and the amount of data that Cell can pull, but I dont know what to believe now, there is so many stories right now, that im getting tired to read crap and more crap..

cliffbo

06-05-2006, 06:57 PM

rog27

06-05-2006, 08:04 PM

Hahaha...I love it. All of the FUD. All of this is being taken out of context by people that haven't a clue what they are talking about. Here's my opinion on the matter:

People, as well as the Inquirer, are confusing the context of the phrase 'local memory' as portrayed on the slide in the Inquirer article. The Inquirer comes to this conclusion, which is hilariously misinterpreted:

If you can write at 250x the read speed, it makes Cell local memory just about useless. That means you do all your work out of main memory, and the whole point of local is, well, pointless.

'Local memory' in the context of the slide does not mean 'local store (LS)' that accompanies each SPE. 'Local memory' is referring to RSX's GDDR3 memory. The reason CELL would read anything from GDDR3 memory in the first place is beyond anyone's comprehension at this point (maybe something in the future) because it is faster for RSX to just write to XDR for CELL to read (if shuffling data back and forth between CELL & RSX). So, when the Inq article infers that the LS memory on the SPEs is useless...they haven't an f'ing clue what they are talking about. You should have stopped reading at this point. Inq = FUD.

And here's Tom with the weather....

cpiasminc

06-05-2006, 08:47 PM

'Local memory' in the context of the slide does not mean 'local store (LS)' that accompanies each SPE. 'Local memory' is referring to RSX's GDDR3 memory.
Which is an example of bad wording on the part of the slide authors -- it should actually be "video memory", as it's only "local" to RSX (and even that is a funny way of wording it). The thing is that reading from GDDR3 for CELL is so ass-slow because GDDR3 is always saturated, and RSX gets first "dibs" on everything. Meaning CELL is just constantly waiting most of the time. Same thing happens with writing, but the difference there is that when you're writing, you're typically not going into locations in memory space where RSX is actively making changes, while when you're reading, you're probably going to be somewhere where the GPU is writing, so you have to wait for it to finish. Hence, "teh suck."

In any case, this is pretty much suggesting that the fast way to get data into the GPU is to simply write out to main memory and point RSX to read from main memory. This makes a lot of sense given that the SPEs don't really have a direct access path out to the GPU push buffer -- they can at least writeout to main memory through a DMA list command. The fun stuff would be starting a write and initiating a render call around the same time, keeping the writes *just* ahead of the GPU.

section

06-05-2006, 08:57 PM

Mountain out of a molehill. When such slides are shown in a devfest like devstation and then pulled out of a proportion months after the actual happening in a sensationalistic site then all I can say again is "wow".

Yeah and thanks for the clarification XBD, rog and Cpiasminc.

rog27

06-05-2006, 10:42 PM

Which is an example of bad wording on the part of the slide authors -- it should actually be "video memory", as it's only "local" to RSX (and even that is a funny way of wording it).

Sony has a terrible time with their ENGRISH...

Many times I feel that someone techinical from Sony Japan (like a hardware engineer) writes summaries in Japanese and a PR monkey tries to directly translate them for the slides. This, in my opinion, is why things on Sony slides sometimes just don't sound right.

rog27

06-05-2006, 10:49 PM

cpiasminc, does the RSX have read/write access to Local Stores on the SPEs directly?

cpiasminc

06-05-2006, 11:31 PM

cpiasminc, does the RSX have read/write access to Local Stores on the SPEs directly?
I doubt it on the PS3, anyway. While there is a supposed environment setup within IBM slides some time back about the SPE local stores being included in the virtual memory space (in specific applications), I think most of the public talk from Sony has suggested that they won't do that.

woundingchaney

06-05-2006, 11:34 PM

I was wondering about heat issues.

Due to an internal power supply could there be a need to drop the speed of the RSX to ensure stability???

-Im not real familiar with the heat of BR or similar Nvidia cards but I believe the Cell runs reasonably cool-

Just a question.

Infernal

06-05-2006, 11:37 PM

Well I thought RSX ran reasonably cool with a 550mhz clock at 90nm compared to other nvidia cards at 110nm(?). Cell was shown to run pretty cool but I have no clue about blu-ray, shouldnt be too hot at 2x though. This thing weighs 5kg though so it could have an ice cube maker inside for all we know :stirpot: .

LaLiLuLeLo

06-05-2006, 11:40 PM

YOUR MOM'S NOT FAST ENOUGH YET. OHHHHHH....

Um, I have no idea why it wouldn't be finished at this point, but as long as it hits the initial spec they targetted, then it should be fine, right?

Smokey

06-05-2006, 11:45 PM

This thing weighs 5kg though so it could have an ice cube maker inside for all we know
this is the reason its so heavy lol :)

LaLiLuLeLo

06-05-2006, 11:58 PM

It cooks breakfast. Just pop in Final Fantasy 13 or some other graphically intense game and the darn thing will make toast.
(I really want a console that makes toast, if you know my posting history here)

Grandia

06-06-2006, 12:19 AM

woundingchaney

06-06-2006, 12:21 AM

i dont really get tech stuff.
so the ps3 graphics are half as much as the x2? and the cpu is crippled? so the ps3 isnt 2x more powerful than the x2 like sony said? more sony bs being exposed i guess. i guess it explains why ps3 games arent on par with x2 games tho.
Thats not fair and derogatory. The truth is it is still being speculated which system will be more "powerful". I personally believe they will be very comparable with each having strong points.

rpgamer_2k5

06-06-2006, 12:22 AM

The SPEs could be used for vertex ops too. Remember it's 800m vertices per second for the SPE. :)

woundingchaney

06-06-2006, 12:23 AM

The SPEs could be used for vertex ops too. Remember it's 800m vertices per second for the SPE. :)
Not again, oh no not me

:grouphug:

Infernal

06-06-2006, 12:26 AM

i dont really get tech stuff.
so the ps3 graphics are half as much as the x2? and the cpu is crippled? so the ps3 isnt 2x more powerful than the x2 like sony said? more sony bs being exposed i guess. i guess it explains why ps3 games arent on par with x2 games tho.
Well neither system is no where near twice the power of the other. Both have their advantages (360 has AA and PS3 has fillrate) and both are going to have comparable graphics. Its the other things like physics, sound, and AI that I expect PS3 to primarily be better at.
The SPEs could be used for vertex ops too. Remember it's 800m vertices per second for the SPE.
Thought it was 200m.

yoshaw

06-06-2006, 12:34 AM

venomv

06-06-2006, 12:38 AM

Well I'm not techminded either.........but people that are have pretty much said don't worry about it, as it doesn't make any foreseeable differnce.

digital neXus

06-06-2006, 12:41 AM

My Flash Memory goes faster than 16MB/s :)

cpiasminc

06-06-2006, 03:03 AM

The SPEs could be used for vertex ops too. Remember it's 800m vertices per second for the SPE.
If the setup rate is limited, the setup rate is limited. It's like saying that if the 48 shader pipes of Xenos were all dedicated to vertex processing, you could get out 6 billion verts per second... sure, but the setup engine only pumps out 1 vertex per cycle, so it doesn't matter how much you can actually process.

makeitlookreal

06-06-2006, 03:25 AM

CPIASMINC,

So are you saying that the SPE's really won't be able to help the RSX very much with vertex data? If so, could you go into a little more detail to explain the concepts involved for those of us who are not quite as technically literate.

Thank you very much.

cpiasminc

06-06-2006, 04:42 AM

No, I'm saying that, sure the SPEs could help, but the limitation in question is not a matter of processing power but simply how frequently a triangle is fetched for rasterization. If you hit that limit, you've hit that limit no matter how much more you could theoretically do.

A vertex routine on an SPE might be able to do a 200 million verts per second by itself, but what good is it if, say, you're hooked up to a GeForce2 MX, meaning not all 200 million can even be processed fast enough to keep up with the SPE?

That said, 275 Million triangles is not a limit you're likely to hit. You're probably going to hit limits elsewhere long before that limit of under 4-9 million tris per frame becomes a problem.

Red_Eyes

06-06-2006, 05:08 AM

i dont really get tech stuff.
so the ps3 graphics are half as much as the x2? and the cpu is crippled? so the ps3 isnt 2x more powerful than the x2 like sony said? more sony bs being exposed i guess. i guess it explains why ps3 games arent on par with x2 games tho.

Yeah, the 360 can do more triangles than the PS3. But that's only IF the 360 only do triangles and nothing else, no texture, just pure triangles. If the 360 starting doing textures too (and why would a game be made of just pure triangles and nothing else) then it would never reach it's peak. On the other hand, the PS3 has it's own dedicated pipeline to do the textures, triangels, etc, seperately. So when the PS3 starts doing triangles and textures, it'll still reach it's peak in triangles. The 360 on the other hand, has share pipelines. So if it's being use for textures, then it can't do as much triangles. In the end, the triangles doesn't matter anymore. It's not how much triangles you can do alone, but it's how much triangles you can do while doing textures and other cool effects. And PS3 wins in that area.

And to clarify the cpu cripppled issue. The PS3 has two pool of memory. One for Cell to use. And one for the video card to use. Cell can read and write from and to it's own memory pool very fast. The video card can read and write to and from it's own memory pool very fast too. Cell can write to the video card's memory pool very fast. The video card can write to cell's memory pool very fast. The only time it slow down is when Cell is trying to read from the video card's pool of memory. But than again, Cell will never need to read from the video card's pool of memory anyway, so it doesn't matter if that's slow or not. All Cell is concern about is writing to the video card's pool of memory, not reading from there. And right now, writing to there is very fast. So, for the last time, it doesn't matter!

Nerve-Damage

06-06-2006, 06:30 AM

Yeah, the 360 can do more triangles than the PS3.

Three things;
• Nothing official or final has come from Sony about the RSX raw triangle count or performance.
• Making final assumptions based off Inquire news is a big “No-No” :skull:
• And basing RSX solely off the PC-ish variation of G70/G80 technology is also begging for trouble.

yoshaw

06-06-2006, 07:02 AM

Three things;
• Nothing official or final has come from Sony about the RSX raw triangle count or performance.
• Making final assumptions based off Inquire news is a big “No-No” :skull:
• And basing RSX solely off the PC-ish variation of G70/G80 technology is also begging for trouble.

Couldn't have said it any better mate!

makeitlookreal

06-06-2006, 11:45 AM

After spending a good bit of time looking around on the internet I do not see any other reference to the 275m polygon setup limit.

I did see a reference that NVIDIA claimed the RSX could push 1.1 Billion vertices. If you divide that by three you get 366 million triangles. I hope this is not an indication that the RSX could theoretically produce more triangles than it is actually allowed to setup.

Cpiasminc, do you think that the RSX will even need help with vertex data from the RSX if all of it's vertex units are chugging away at full power? Basically, I guess what I should ask is if the RSX has the power to produce 275 million polygons per second in the first place.

Of course those polygons need to be shaded and manipulated in various ways, but since the PS3 does not use a unified shader architecture are there still barriers that prevent it from pushing as many as it's non-shader pipelines can allow?

I guess I am just a little confused about this.

Chrome

06-06-2006, 12:15 PM

A word of wisdom nugget from someone smart I'm sure, "Never underestimate an enemy you don't fully understand" :D

Grandia

06-06-2006, 12:25 PM

Nerve-Damage

06-06-2006, 12:43 PM

but didnt sony say it was a nv47 which is a g70?

Not to my knowledge has Sony officially said RSX is based off G70 technology. I know NVIDIA has stated that the RSX GPU is based off there latest graphics technology (which would be G70). However, that doesn’t mean RSX is solely G70 innards. You can belong to the same family (NV G70/G80 IP) and contain certain traits of all the members. However that doesn’t stop the RSX from outperforming them all (err X-Men :laugh: ).

Nameless

06-06-2006, 05:19 PM

Considering we have not seen final specs on the RSX it could go in the direction of the G80 architecture, highly unlikely but possible...

If this occurs the PS3 would have the most advanced GPU for about half a year after launch and that would be an impressive feat...(I don't count SLI setups or quad cards PC setups, it's not a fair fight because you have multiple cards. One on One is a fair fight baby...)

At this point I could care less if it's the G70 or G80 since it's a closed environment the power will be put to good use regardless... Peace

rog27

06-06-2006, 07:11 PM

You all need to remember that sheer number of polygons is not the only thing necessary to reach the holy grail of visual glory: That cinematic CGI look. In fact, about 10+ years ago, CGI was pushing less polygons per frame than some games are now. But for some reason, games still don't look or feel as natural as CGI. The reasons:

-choppy animation systems lacking proper transistions
-lack of realistic physics
-lack of layers of meaningful interaction with environment, characters, etc.
-lack of realistic lighting
-predictable patterns of occurence on-screen (lack of simulated reality)

^^^all of this even though the graphical representations of objects/entities themselves (models, environments, etc.) are approaching good enough. Most of these can't be solved with a better GPU. It's going to take a CPU with a magnitude difference in performance and the ability to run in parallel many different programs/simulations to solve this problem. I believe that is why Sony chose to invest so much money in the CELL in the first place.

Sephiroth_VII

06-06-2006, 07:20 PM

Not to my knowledge has Sony officially said RSX is based off G70 technology. I know NVIDIA has stated that the RSX GPU is based off there latest graphics technology (which would be G70). However, that doesn’t mean RSX is solely G70 innards. You can belong to the same family (NV G70/G80 IP) and contain certain traits of all the members. However that doesn’t stop the RSX from outperforming them all (err X-Men :laugh: ).
We know that it's based off the NV47 (http://www.watch.impress.co.jp/game/docs/20060329/3dps309.jpg), and that pretty much sums up our knowledge of RSX.:(

cpiasminc

06-06-2006, 08:09 PM

Yeah, the 360 can do more triangles than the PS3. But that's only IF the 360 only do triangles and nothing else, no texture, just pure triangles. If the 360 starting doing textures too (and why would a game be made of just pure triangles and nothing else) then it would never reach it's peak.
With modern GPUs, texturing isn't really an issue as they're all made for multitexturing up to a point. The basic limitation is this -- the setup engines of both units is limited to a certain vertex feed rate completely irrespective of computational units or texturing units or anything else.

If there WAS no such limitation at all -- say you're not doing any pixel shading at all, so every available shader unit is strictly doing vertex processing. RSX has 8 vertex shader pipes, and since the least you can do with a vertex is transform it, that's 4 clock cycles. Assuming perfect multithreading, 8 pipes/4 cycles = 2 vertices per clock. 2 vertices per clock * 550 MHz = 1.1 billion verts/sec.

Similarly, with Xenos, you've got 48 pipes, which, with no pixel shading, would all be dedicated to vertex processing. 48/4 cycles = 12 vertices per clock = 6 billion verts/sec.

Neither of them can really hit this because there's a completely separate bottleneck that limits them to setting up only a specific number of verts per second. And in any case, like you said, there will be more to do per vertex than simply transforming them. If they have textures, you at least need to copy over texture coordinates and so on.

Cell can write to the video card's memory pool very fast. The video card can write to cell's memory pool very fast.
I wouldn't really say that. If you notice the chart, the throughput of Cell writing to VRAM is pretty slow compared to the throughput of RSX reading from the XDR main memory pool. The throughput is near the theoretical limits of PCI-Express x16.

After spending a good bit of time looking around on the internet I do not see any other reference to the 275m polygon setup limit.
Yeah, Inquirer doesn't make any mention of where it even came from... they just say "so and so said 275 million". Nonetheless, the figure at least sounds valid as 275 Million would be equivalent to the 550 MHz RSX setting up 1 triangle for rasterization every other cycle.

I did see a reference that NVIDIA claimed the RSX could push 1.1 Billion vertices. If you divide that by three you get 366 million triangles. I hope this is not an indication that the RSX could theoretically produce more triangles than it is actually allowed to setup.
Why would that be a bad thing? If it has more power than it really needs, that just means a little bit of extra breathing room to do more work per vertex. And 1.1 billion vertices can theoretically be 1.1 billion triangles since tris can share vertices. The typical worst-case scenario happens to be particles, where you have 2 verts per tri (every particle is a set of 4 vertices giving you two triangles). But of course, sharing of vertices is also dependent on how big the post-transform cache is, so you might end up reprocessing some verts anyway.

Cpiasminc, do you think that the RSX will even need help with vertex data from the RSX if all of it's vertex units are chugging away at full power? Basically, I guess what I should ask is if the RSX has the power to produce 275 million polygons per second in the first place.
Sure RSX has the power, but there are other benefits to doing things in software. For instance, if I wanted to extrude shadow volumes, I could potentially have loads and loads of scissor planes, which can be handled much more thoroughly in software since the GPU only has small register space. And it also means sending along a tiny vertex format to the GPU when you're done.

It can also save bandwidth, for instance, if you skin characters in software since you won't have to send several matrices, and you can ship the character off in a more compact vertex format.

Smokey

06-06-2006, 08:21 PM

crap i dont know if cps post is Good, Bad or Ugly lol :)

Arnaud_M

06-06-2006, 08:38 PM

But for some reason, games still don't look or feel as natural as CGI. The reasons:

-choppy animation systems lacking proper transistions
-lack of realistic physics
-lack of layers of meaningful interaction with environment, characters, etc.
-lack of realistic lighting
-predictable patterns of occurence on-screen (lack of simulated reality)

^^^all of this even though the graphical representations of objects/entities themselves (models, environments, etc.) are approaching good enough. Most of these can't be solved with a better GPU. It's going to take a CPU with a magnitude difference in performance and the ability to run in parallel many different programs/simulations to solve this problem. I believe that is why Sony chose to invest so much money in the CELL in the first place.

I completly share your opinion. Moreover, I am certain future games will evolve in the direction you cite (more realism and entertainment value by more realistic interactions with the world, physics and the like) and indeed we already see new middleware emerging (such as the Endorphin product for characters animation) to support this approach. Sony is on the right track with Cell I think, but dividends won t probably pay off until a few years. But they will :-)

Arnaud

Pumpkin Head

06-06-2006, 08:48 PM

Quote:
Update: The misunderstanding evident in the linked story relates to the distinction between local video memory and local system memory. The slow read speed under discussion is indicative of the feature's lack of utility. This is even reflected on the slide's statement: "No, this isn't a typo ..." A contact at Sony confirmed this telling me, "Again I cannot imagine a situation where you have any SPU reading from the RSX local memory." Nothing to see here, folks.

http://portable.joystiq.com/2006/06/05/rumor-ps3-hardware-slow-and-broken/

I know, it's just as shocking to me as it is to all of you. Imagine that, IBM doesn't design their processors on the back of the short bus after all.

xbdestroya

06-06-2006, 09:25 PM

Cpi great useage scenarios there.

overclocked

06-06-2006, 10:03 PM

Just one thing i noticed, what the hell was it with 420Mhz, cause it was 450Mhz thats sure, wonder if those devs worked with a regular G71-GT card, thats the only thing thats a little strange.

solidus

06-06-2006, 10:39 PM

Just one thing i noticed, what the hell was it with 420Mhz, cause it was 450Mhz thats sure, wonder if those devs worked with a regular G71-GT card, thats the only thing thats a little strange.
I find it strange that first they say that the clockspeed is only around 80% of what it should be, but than they claim that the RSX has a triangle set-up of 275 million, which doesn't make sense unless they're talking about the final clockspeed (550 MHz).

Either way, I'm still curious about the claim that the RSX can do 1 triangle every other cycle...

makeitlookreal

06-07-2006, 12:01 AM

This is what I have came to after reading Cpiasmics posts and considering everything.

The RSX is capable of 1.1 Billion vertices per second and limited to 275 million triangles (the second number has not been 100% confirmed however).

Basically, at only 275 million triangles per second the PS3 produces less than the XBox 360. However, the PS3 makes up for this lack of triangles by having the ability to do *more* to the triangles by shading them and performing other processes.

Even though the PS3 has a maximum rate of 275 million triangles the news is not really bad at all, because the PS3 can do MORE to those triangles than the 360.

Additionally, the more it is doing to each triangle the less triangles can be produced. I believe this is one situation of several where the SPE's of the Cell can be utilized to help produce more vertex data.

Some people are saying (and it seems logical) that having massive shading power is actually to some degree more important than having massive vertex power. If this is the case then the balance that Sony has created is a good thing if we want the best possible graphics from the system.

It really doesn't matter if you push tremendous ammounts of triangles if you cannot shade them and make them look the way you need them too!

woundingchaney

06-07-2006, 12:24 AM

cpiasminc

06-07-2006, 01:18 AM

Even though the PS3 has a maximum rate of 275 million triangles the news is not really bad at all, because the PS3 can do MORE to those triangles than the 360.
Mmmm... depends what you mean by "do more to the triangle", as PS3's power is largely in its pixel shader units. If you were referring to how much work can be done per vertex, then 360 has that edge as it has more shader pipes that can dynamically be allocated towards vertex processing. Certainly, per-pixel, RSX has some power.

When I spoke of doing more work per vertex, I was referring to the overall ratio of peak theoretical throughput of the vertex units vs. the setup engine's limits. I wasn't really comparing any two GPUs with that. Since this ratio is greater on Xenos, Xenos can do more per vertex. Of course, in practice, you tend to hit more on the pixel side of things, and that's the stuff that will cost you in the end anyway.

PS3 doesn't really have as much vertex processing power as 360 theoretically could have, but since its main weapon is its pixel processing capabilities, that's where it will shine.

Take for instance, images like the following --
http://www.acm.uiuc.edu/siggraph/images/eoh_projects/aEmber1.jpg
http://www.flipcode.com/iotd/05-23-2001.jpg

The engine that rendered those is very, very vertex heavy, and does next to nothing per pixel beyond alpha blending (and actually, when I worked on it, the term "GPU" didn't exist let alone vertex/pixel shaders 8-)).

Now on the other hand, you could take snaps like those from just about any UE3-based game or something like Lair, which all do a lot of per-pixel work, which is where RSX would get to show its mettle.

In practice, there will be other limits you'll hit long before you hit limitations like 275 million triangles. For instance, think of 275 million tris, each of them occupying, on average, 50 pixels (which means, on average, a tri is 10 pixels wide and 10 pixels tall)... 275 * 50 = 13.75 GPixels/second... which is more than the fillrate of RSX and Xenos combined. You'll hit a fillrate limit long before you hit a setup rate limit.

makeitlookreal

06-07-2006, 01:21 AM

Basically, most of the interviews with devs seem to indicate that PS3 has massive shading power more so than vertex power if that answers your question.

Grandia

06-07-2006, 02:07 AM

solidus

06-07-2006, 02:24 AM

so the rsx geometry rate is much lower than the 7800 or 7900 so it cant be on par with either one of them. really dumb sony should have just put a 7800 with no cutbacks instead of hack it down to whatever the rsx is.
Actually, I don't recall Nvidia nor Sony say anything concerning the RSX' triangle set-up. Everyone was assuming it would be the same as 7800GTX (1.1B vertices/s), although it also seems that all NV4x/G70-based GPU's have the same set-up (1tri/2 cycles).

I'm not sure about the latter part though...

makeitlookreal

06-07-2006, 02:55 AM

Once again, I believe Sony is hiding some major aspects of the RSX. Do we really think they wasted millions of dollars and all the thousands of man hours they bragged about if the RSX is little more than a 7800 with a few minor modifications?

My guess is before launch we will discover that the RSX is probably based on the 7800, but has other features we don't know about which will make it at least to some degree more powerful.

venomv

06-07-2006, 04:20 AM

I think a well modified 7800 is what we are getting, if the differnce were slight it wouldn't make sense to make a custom GPU.

makeitlookreal

06-07-2006, 04:26 AM

Venum,

If it is truly well modified why would they pick a chip such as the 7800 just to downgrade it?

My guess is that the final RSX will be very different from the 7800. I think so far they have only released specs similiar to the 7800 to distract us from a surprise announcment in the coming months. Seriously, anything else just does not make sense! Instead of spending lots of money to modify down a powerful chip they could have just included a less powerful chip and saved much of the time and effort.

Probably the specs so far are just those of the 7800. What they add to it will increase those specs.

venomv

06-07-2006, 04:43 AM

We are thinking along the same lines, it may not be light-years beyond the 7800, but I think we should expect something better, or at least have some strange quirks (even I don't know what they may be, unless Nerve Damage hit the mark at some point).

Clixx

06-07-2006, 07:46 AM

How, if at all, does Cell factor in ? A while ago there were articles alleging that
Cell outperforms NVIDIA GPUs in vertex shading by a factor of 5. Can't it be used
to boost RSX's vertex performance if needed ?

overclocked

06-07-2006, 11:00 AM

I find it strange that first they say that the clockspeed is only around 80% of what it should be, but than they claim that the RSX has a triangle set-up of 275 million, which doesn't make sense unless they're talking about the final clockspeed (550 MHz).

Either way, I'm still curious about the claim that the RSX can do 1 triangle every other cycle...

I can only answer for the last part with any knowledge. The former is that it could be when i think of it the first real kit in a smaller factor and heat/stability should be at top and when the last kit arrives its final. :shrug:

/also i bet theres only a handful in a big team that cares or even bothers whats in the box. :angelgrin

About the last part then from what i read in developers paper from nVidia nVidia, i never seen they mention tringle setup. For NV4x hardware its short shaders and many instead of few long shaders, use FP16normalize when you can etc. They always list "Do not do that" cause it will limit or break down performance of *this* part of the pipeline but i never seen triangle setup engine as one of these factors.

The ATI pappers were more funny to read! :hugegrin:

But it seems reasonable.
I think many got/gets confused when vertex shader counts drew away massively because of increased VS units and MHz on top of that.
Well just a idiot thoughts. :crazy:

How, if at all, does Cell factor in ? A while ago there were articles alleging that
Cell outperforms NVIDIA GPUs in vertex shading by a factor of 5. Can't it be used
to boost RSX's vertex performance if needed ?

Because its software, and take any numbers with a big grain of salt. In Warhawk(my favorite title hehe) the developers are doing that with the water with Cell, the spu does compute waves/their height and so on, dont know exactly how it works or what they are calculating per se. Then they send the info to the PPE which then sends the final frame to be rendered by RSX. Just go in and take a look at the Warhawk homepage.

But i think they are just playing around now at first, the "real" good things takes longer/will come later..

Sebastiano

06-07-2006, 03:19 PM

Guys
I found this article technewsworld.com about Nvidia's latest GPU (GeForce 7950 GX2).
there is a short comment in it about the PS3 GPU
not much... but something. So I figure I'll share with you

Graphics silicon giant Nvidia doubled up on graphics processors for its latest high-end graphics processing card, touting the new GeForce 7950 GX2 as the "world's fastest."

The new GeForce 7 series cards draw on two graphics processing units (GPUs), dramatically increasing single video card performance and ushering two-card, four-GPU "extreme" high-definition resolutions, smooth frame rates and stunning visuals, the company said.

Nvidia spokesperson Brian Burke told TechNewsWorld the new GeForce 7950 cards were priced between about US$599 and $649, and were immediately available to the public.

Performance Times Two
Nvidia touted the coupling of two GeForce 7950 GPUs on a single graphics card , described in some reports and discussions online as a technical truth, but perhaps a stretch since the unit consists of two cards mounted together.

Nvidia was nevertheless praised for its design, which packs more graphics punch into nearly the same space as a single video card, uses a single PCI Express slot, and delivers cutting edge graphics.

"The GeForce 7950 GX2 is a great card for the gamer that's looking for the absolute best performance in a single graphics card solution," said Nvidia General Manager of Desktop GPUs Ujesh Desai.

Desktop and Widescreen
Although the new 7950 may be using some of the same technology that powers the Nvidia graphics of Sony's (NYSE: SNE) next-generation gaming console PlayStation 3, the new high-end desktop graphics card is not the same technology as the PS3 graphics silicon, according to Burke.

"This is a desktop GPU part, it's not part of what we're working on with Sony on PS3," he said.

In addition to the twin GPUs running on one card, the 7950 GX2 also features 48 pixel pipes, 16 vertical shaders, more bandwidth and memory, and DVI support for widescreen flat panel displays and resolution as great as 2,560 x 1,600.

Nvidia also stressed the support for its Quad Nvidia SLI technology, which will allow Nvidia system builders to combine two 7950 GX2 cards for even more graphics horsepower the higher resolutions.

Integration Pros and Cons
At the Computex tradeshow in Taipei, Taiwan this week, Nvidia also showcased its nForce graphics processing technology support for Intel (Nasdaq: INTC) Core 2 Duo processors, which the two companies said would result in faster, feature-rich motherboards for enthusiasts.

The advantage of the technology is that the Core functionality is integrated into the Nvidia motherboards, which remain nicely packaged, small chipsets, Gartner Research Vice President Martin Reynolds told TechNewsWorld.

The only issue with the processor integration -- which may be ideal for desktops, notebooks and even servers -- is that an all-Nvidia chipset may be missing out on some of Intel's other technologies or capabilities, while Intel's chipset technology may be missing Nvidia features or functionality.

Source (http://www.technewsworld.com/rsstory/50939.html)

Sephiroth_VII

06-07-2006, 03:46 PM

If this is true, it'll be awesome!! the 7950 is a great GPU. Nice find!!

Nameless

06-07-2006, 04:07 PM

Still waiting on the G80 architecture before I make an investment...
This post has nothing to do with the PS3, but interesting read...

venomv

06-07-2006, 06:16 PM

That wasn't much, just basically saying the RSX isn't the 7950.

cpiasminc

06-07-2006, 07:01 PM

My guess is that the final RSX will be very different from the 7800. I think so far they have only released specs similiar to the 7800 to distract us from a surprise announcment in the coming months.
The only real reason that the specs sound like a 7800 is because we haven't heard many of them. The 7800, 7600, 7900, 7300 are all built on the same base architecture. And RSX fits somewhere in this group.

Just note that things like the transistor count nearly matching cannot be taken as anything more than a coincidence -- it is already a given that RSX wouldn't have a PCI-E bus controller and instead have a FlexIO controller, and that it more than likely won't have any sort of hardware video decoding (though it has miscellaneous other stuff that a typical GeForce wouldn't have)... The fact that it still matches up is chance.

How, if at all, does Cell factor in ? A while ago there were articles alleging that Cell outperforms NVIDIA GPUs in vertex shading by a factor of 5. Can't it be used to boost RSX's vertex performance if needed ?
I believe it's been said a few times before that the limitation in question has nothing to do with the computational power of anything. No matter how many vertices Cell or RSX or both can compute, that has nothing to do whatsoever with the setup rate. That's a fixed limit.

rog27

06-07-2006, 07:18 PM

The fixed triangle setup rate is orders of magnitude greater than the practical number they can use relative to the number of pixels these machines can push...both machines will be fillrate limited ever before they reach that theoreticall 275 mill or 500 mill triangles number. There are many reasons that many polys (small sized triangles) shouldn't be pushed in the first place:

-too complex a level of detail for the CPU to manipulate accurately (precision it too high...need double floating point precision for some of that)
-too much resources spent on cleaning up jagged-edged mess that would result from all of the aliasing
-post-processing/framebuffer/multi-pass effects would clober system resources.

For efficiency purposes...you scale the abilities of your GPU to work with your what your CPU can handle.

overclocked

06-07-2006, 07:22 PM

The only real reason that the specs sound like a 7800 is because we haven't heard many of them. The 7800, 7600, 7900, 7300 are all built on the same base architecture. And RSX fits somewhere in this group.

Just note that things like the transistor count nearly matching cannot be taken as anything more than a coincidence -- it is already a given that RSX wouldn't have a PCI-E bus controller and instead have a FlexIO controller, and that it more than likely won't have any sort of hardware video decoding (though it has miscellaneous other stuff that a typical GeForce wouldn't have)... The fact that it still matches up is chance.

Whats been "said/known" public is well basically NV47-based, 8VS,24 2D Tex lockups, the speed and 8ROPs if memory serves me right cpi?
And of course out with some PC stuff and in with some console stuff like you said.

cpiasminc

06-07-2006, 07:32 PM

Whats been "said/known" public is well basically NV47-based, 8VS,24 2D Tex lockups, the speed and 8ROPs if memory serves me right cpi?
Plus a few others like bus speed, bandwidth across the FlexIO, max instruction throughput, and so on... though the 8 ROPs, transistor count, and a few others are all largely guesses among the community.

A few other things are not yet announced, though planned, but those aren't really meaningful to the rendering capabilities anyway (but again, they're there because it's a console).

makeitlookreal

06-07-2006, 08:41 PM

I cannot wait until Sony comes fourth and tells us the truth. I will be so happy I will stand up, jump for joy, and dance in circles. On that day I will be full of joy and glee! Joy and glee! JOY AND GLEE!!!!!!!!!!!

JOY AND GLEE I TELL YOU!!!! JJJOOOYYY AND GGGLLLEEEEEE!!!!!!!!!

:banana: :queer: :clapping: :banana:

But until then every time I hear a rumor and all it would take is the truth from Sony to set it strait I am about to pull all my remaining hair out! We need the truth!

:swear: :guns: :swear:

CPIASMINC, I just want to say thank you for at least sharing what you are allowed to with us. I may become frustrated quite often because we don't have all the information, but I realize you don't have to share one tiny bit of information with us but you tell us all you possibly can, answer our questions, tolerate our lack of technical understanding, and at the same time treat us with respect! Thank you so much CPIASMINC!

You DA MAN!

:clapping: :banana: :clapping: :banana:

Nameless

06-07-2006, 08:56 PM

MILR, you are losing it... ;-)
I do agree that CPI is always respectful and informative with his responses... (the developer stories were priceless...)

CPI good to have you in the forums.:praise:

LinpinWangyFoot

06-07-2006, 09:55 PM

i'm so sick of this sort of thing. we have no idea how powerful the RSX is. Sony have signed another deal with them so clearly its not that bad

rpgamer_2k5

06-08-2006, 05:12 PM

With the information we have, we can conclude:

One, the RSX seems like a part inbetween the 7600 and 7800 with the 128-bit bus. The setup rate inquirer claims is actually around the same if not a bit higher than the GT. And that would be possible if the core clockspeed is 550Mhz allowing it to match the GT.

Two, on paper it may seem like the Xenos is indeed much superior than the RSX but in reality the 500 triangle setup rate seems is only possible if all of the Xenos pipes are all tasked for vertex ops. Therefore if half of the pipes were processing vertex shaders than the setup rate would likely drop by half (matching the speculated setup rate of the RSX) simply because the pipes are homogenous in design/performance (hence unified). And note that the Geforce 7800 vertex pipes constitute only 25% of the total pipes, so if we see a similar setup for the Xenos, the RSX will have a higher theoretical triangle setup rate.

Three, even if Xenos uses half of the pipes for vertex ops with the eDRAM possibly undertakings tasks that fragment ('pixel') pipes traditionally do -- who knows, anything is possible --- efficiency would probably decide the 'better' part if Xenos can use 50% of the pipes for vertex ops. And for that reason we need to note that the Xenos with 48 pipes can churn out 500m triangles/sec versus 275m triangles/sec with only 8 VS units. And for that reason it is fair to say that RSX vertex pipes would probably fall under a higher percentile in reaching the max triangle setup rate compared to the Xenos. But anyways, the percentage for both won't be really high.

Now if the RSX is equal to the Geforce 7800GTX @ 90nm with a clockspeed of 550Mhz then the vertex shader clockspeed would be a bit higher anywhere from 570-600Mhz. Heck even if it is 550Mhz, it would have a triangle setup rate in between the 7900GT (470Mhz VS) and GTX (700Mhz). At worse we're talking about over 360m trianges/sec which would mean the Xenos would require ~70% of the pipes designated for vertex shaders. And if the end the specifications of that heavy, 5 kg PS3 has packages that can churn out close to 2B vertices/sec (>600 triangles/sec) of the 7950GX2 then the Xenos is owned even if all 48 pipes are used. ^_^

Things seem better, when brain is used. :)

cpiasminc

06-08-2006, 08:34 PM

Two, on paper it may seem like the Xenos is indeed much superior than the RSX but in reality the 500 triangle setup rate seems is only possible if all of the Xenos pipes are all tasked for vertex ops.
How do you figure? If you assume that all 48 pipes are dedicated to vertex processing, and the simplest possible vertex processing loop is simply to transform the vertex (4 dot products), that's 48/4 = 12 vertices per cycle. Even in a worst case scenario of 3 verts/tri, that's still 2 billion tris per second at 500 MHz.

The reason the quote from MS said that it can achieve that throughput with "non-trivial shaders" is because if every pipe is dedicated to vertex processing, to get you down to 1 vertex per cycle (which is the setup limit), you need a vertex shader of 48 cycles (48/48 = 1, after all :))

In no GPU ever created does the power of the shader units have anything to do with the setup engine. It is a separate unit entirely that simply has the job of fetching vertices out of the post-VS cache and issuing a triangle for rasterization. It doesn't care how quickly vertices get into the cache, except to know that the vertices it needs are actually present. The only thing that limits the setup engine is the clock speed of the GPU.

Of course, if the vertex shaders you're running are so long that the shader units end up keeping the setup engine waiting, that's another matter, but as it so happens, that is infinitely more likely in any given game than other conceivable cases. All the "extremes" we keep bringing up are pretty contrived, when you get down to it. In practice, the fact that Xenos can process more vertices is mainly a result of the fact that the total number of shader pipes is that much greater (48 vs. 32). RSX is made for pixel shading.

Therefore if half of the pipes were processing vertex shaders than the setup rate would likely drop by half (matching the speculated setup rate of the RSX) simply because the pipes are homogenous in design/performance (hence unified).
No, it wouldn't... it would just mean that the limit of the number of cycles per vertex shader to keep you *just* matching the setup rate has gone down by half. The setup rate is still the same. I believe there is a condition under which the setup rate on Xenos DOES go down to half, but even though I don't remember what it is, I'm pretty sure it's an esoteric combination of settings that isn't very practical (again, nothing to do with the processing power of the GPU).

The same is true of RSX with its 8 vertex pipes. It's not that it needs 8 pipes to setup x number of vertices per second, it's that it can compute many more, but to get the same vertex throughput as the setup rate means no more than 16 cycles per vertex shader.

In either case, there are other limiting factors far worse than the setup rate that prevent it from ever reaching those limits in practice.

Three, even if Xenos uses half of the pipes for vertex ops with the eDRAM possibly undertakings tasks that fragment ('pixel') pipes traditionally do
The eDRAM doesn't really do anything that a pixel shader pipe might otherwise do AFAICT, but it does take over some of the tasks that would otherwise be done by ROPs.

jako

06-08-2006, 08:49 PM

It seems like everyone missed this sentence :
"This is a desktop GPU part, it's not part of what we're working on with Sony on PS3,"

I've read that NVidia delivered the final ship (RSX) in march, so how can they still be working on it. I really doubt they are upgrading the RSX, now that they will soon begin mass production of the PS3.
Does anyone has an explanation ?

Applefiend

06-08-2006, 10:24 PM

One of the things that interests me is we were always told we couldn't have EDRAM because PS3 was driving 2 1080p screens, right.

Well... It ain't doing that no more... I hope RSX is getting sent back to nVidia for some last minute speed boosts.

cpiasminc

06-08-2006, 10:40 PM

I've read that NVidia delivered the final ship (RSX) in march, so how can they still be working on it.
AFAIK, March was the time of a design tapeout. The manufacturing can still expose some surprises and since nVidia is the original IP source, they can be called on in spite of the fact that production is entirely Sony's domain.

Also, they are working on a further shrink of RSX for later revisions of the console in any case.

One of the things that interests me is we were always told we couldn't have EDRAM because PS3 was driving 2 1080p screens, right.

Well... It ain't doing that no more...
Whether it's one or two 1080p screens, eDRAM is still expensive. 1 1080p screen with 2x MSAA is still about 24 MB worth of eDRAM (assuming no esoteric tiling nonsense). And RSX still has a way higher transistor count than Xenos, eDRAM included. It's costly enough as it is.

I hope RSX is getting sent back to nVidia for some last minute speed boosts.
Speed is mainly a manufacturing thing as it's simply a question of yields which bin down to some clock rate successfully, and that's Sony's domain, so it's primarily Sony's decision to make whether it will get any faster. Power consumption is going to be a concern, of course.

makeitlookreal

06-08-2006, 10:45 PM

It sounds like to me NVIDIA finished their work a while back and then handed the chip to Sony for them to do their part. I doubt NVIDIA is doing any work on the RSX right now. You know, a while back Sony had some pretty neat ideas with the visualizer GPU. Since that time they could have up with additional ideas they may have or may not have wanted to utilize. Perhaps now they have decided to add a few features to the RSX themselves that we don't even know about.

According to Nerve Damage's source the RSX is going to be based on the 7800 (or that line of chips by NVIDIA) but Sony will add a few extra sub-processors to it.

Personally, Sony went all out with the Cell. It is a fantastic CPU. I have a feeling the went ALL OUT on the RSX as well. After all their talk of the RSX being such a custom part with huge ammounts of labor and now how it seems they are taking additional time I think it is more than a slightly modified 7800.

rpgamer_2k5

06-09-2006, 12:16 AM

Thanks for the reply, cpiasminc

In practice, the fact that Xenos can process more vertices is mainly a result of the fact that the total number of shader pipes is that much greater (48 vs. 32). RSX is made for pixel shading. And say if the RSX is akin to the 7950GX2, wouldn't that allow the RSX to have a setup rate equal to the Xenos. Seeing that the PS3 is quite large and has a 5kg weight, the dual GPU should be in. $_$

The same is true of RSX with its 8 vertex pipes. It's not that it needs 8 pipes to setup x number of vertices per second, it's that it can compute many more, but to get the same vertex throughput as the setup rate means no more than 16 cycles per vertex shader. That would mean the SPEs capable of 800m vertices/second each or 5.6B vertices per second, even though 1 SPE is reserved to the OS and a good number would be used for non-graphics purposes. Would this mean that the theoretical maximum of the PS3 is higher than the Xenos (6.4-7B vs. 6B)? I'm aware that these figures cannot be reached, but it's definitely interesting knowing. :)

Now my primary wish is to get the vertex per cycle to 1, while also having a GX2-like configuration. The damn thing is $600US, 5Kg, they need to get the specifications up. Even if it's 1 triangle/2 cycle, a GX2 GPU would have a triangle setup rate equal to the Xenos. And lets remember, Kutaraji did say that the RSX is not a derivative of the PC desktops. Even though it does sound off, it must mean something. And seeing that the price, controller were totally unexpected, maybe it is possible that the RSX specification released are just a part of a ruse. :)

Infernal

06-09-2006, 12:40 AM

I think its safe to assume that the RSX is only slightly more powerful than a 7800GTX. I mean they used the 7800GTX in the beta kits, it should be safe to say that they didn't downgrade from beta to final hardware and the kits should have represented near final power, so nobody should be expecting anything drastically more powerful either. All of these crazy hopes and theories are good and all, but its time to look at the facts.

cpiasminc

06-09-2006, 01:41 AM

That would mean the SPEs capable of 800m vertices/second each or 5.6B vertices per second, even though 1 SPE is reserved to the OS and a good number would be used for non-graphics purposes. Would this mean that the theoretical maximum of the PS3 is higher than the Xenos (6.4-7B vs. 6B)? I'm aware that these figures cannot be reached, but it's definitely interesting knowing.
800m * 6 SPEs + 1 VMX = 5.6 Billion + 1.1 Billion for RSX = 6.7 billion.
If you want to say 7 SPEs, then okay... 7.5 Billion.

Though you'd need two threads on the PPE to get 800 million out of it, but assuming that happens....

But if you're going to start including CPU power, then X360 has 3 VMX units at 3.2 GHz, which means 2.4 billion + 6 billion on Xenos = 8.4 billion.

Now my primary wish is to get the vertex per cycle to 1, while also having a GX2-like configuration.
I'd like to see the attribute read rate go up to at least 2 per cycle. Preferably 8 (get 1 per cycle per vertex shader). Yeah, I know, that takes a miraculous memory bus (8 * 128 bits * 550 MHz = 70.4 GB/sec [and that's how much needs to be available for a vertex attribute read]), but so does everything else.

rog27

06-09-2006, 05:34 AM

In practice, the fact that Xenos can process more vertices is mainly a result of the fact that the total number of shader pipes is that much greater (48 vs. 32). RSX is made for pixel shading.

Wouldn't it be more fair to consider RSX's total number of ALUs in comparison to Xenos's total number of ALUs. If you were doing that way it would be RSX (8 Vertex Shader ALUs + 48 Pixel Shader ALUs) vs Xenos (48 Vertex/Pixel Shader ALUs)?

sif

06-09-2006, 12:01 PM

A few other things are not yet announced, though planned, but those aren't really meaningful to the rendering capabilities anyway (but again, they're there because it's a console).

And RSX still has a way higher transistor count than Xenos, eDRAM included. It's costly enough as it is.

OK cpi, I am a little confused by these two comments. A 7900GTX has around ~280 million transistors and you are saying that RSX has much more than the ~325 million transistors of Xenos. Taking out Purevideo (as you have suggested before) would widen the gap further to at least 70 million transistors, and you say RSX has many more transistors than this. So what is it apart from HDCP, that would be in a console and not a standard Geforce, whilst not being meaningful to the rendering capabilities? You are probably not allowed to answer this, but it is at least a hint that RSX=!G71.

EDIT: Grammar. Also how many transistors is the PS2's GS?

overclocked

06-09-2006, 03:02 PM

G71 is 278million/transistors, Xenos combined is between = lowest 332million and highest 337million transistors, but lets say between 330-340million transistors for simplicity cause i doubt they have counted them by hand.:hugegrin:

version

06-09-2006, 03:08 PM

800m * 6 SPEs + 1 VMX = 5.6 Billion + 1.1 Billion for RSX = 6.7 billion.
If you want to say 7 SPEs, then okay... 7.5 Billion.

Though you'd need two threads on the PPE to get 800 million out of it, but assuming that happens....

But if you're going to start including CPU power, then X360 has 3 VMX units at 3.2 GHz, which means 2.4 billion + 6 billion on Xenos = 8.4 billion.

I'd like to see the attribute read rate go up to at least 2 per cycle. Preferably 8 (get 1 per cycle per vertex shader). Yeah, I know, that takes a miraculous memory bus (8 * 128 bits * 550 MHz = 70.4 GB/sec [and that's how much needs to be available for a vertex attribute read]), but so does everything else.

hehe :D

sif

06-09-2006, 03:09 PM

Thanks Overclocked! :cheers:

Ok so that is at least 335-278+25(for purevideo) = 82 million transistors extra (over a G71), half of which could be a GS (around 40 million transistors), 10.5 million for an EE, so what is the rest? Especially considering its much more than 340 million transistors according to cpiasminc:

And RSX still has a way higher transistor count than Xenos, eDRAM included. It's costly enough as it is.

I'll avoid anything sensationalist like OMg Rsx is teh 795o gx2!!1! and leave the speculation to everyone else!

Also, version: do you actually know anything under NDA and that was some kind of hint from cpiasminc, or are you just playing with us?

makeitlookreal

06-09-2006, 04:14 PM

EDIT: Ok - just say no to teh L337. -xbd

I had a very long, detailed, and funny post here that was supposed to be humourous but was deleted. I am VERY angry, VERY upset, and very MAD that it was deleted. But I want to thank XBD for restoring at least... well... none of it... darn... I was going to edit it and put some normal discussion in it to explain it better, but now it's gone.... let me go as XBox Destroyer....

Could you please put back the text of the post?

Crossbar

06-09-2006, 04:52 PM

And RSX still has a way higher transistor count than Xenos, eDRAM included.
Are you sure about this? I've read that the GPU manufacturers often only count the transistors in the logic circuitry, excluding the transistors used in memory cells (caches etc.). Probably because the transistors in memory cells can be packed much more dense and therefore make use of little die space.

The transistor numbers split mentioned between the two dies in Xenos does not correspond very well to the actual sizes of the dies which can be seen in pictures of the Xenos chip package (see anandtech).

makeitlookreal

06-09-2006, 05:03 PM

Sif,

Just what are you wanting banned? Could you please be more specific?

makeitlookreal

06-09-2006, 05:15 PM

Thank you very, very much XBox Destroyer. I will edit the post with something understandable. Again, thank you VERY much for restoring the post! I sincerely appreciate it!

xbdestroya

06-09-2006, 05:17 PM

Thank you very, very much XBox Destroyer. I will edit the post with something understandable. Again, thank you VERY much for restoring the post! I sincerely appreciate it!

LOL, MILR no big deal. ;)

I'm sorry I deleted it in the first place! Didn't know it meant so much to you. :)

makeitlookreal

06-09-2006, 05:21 PM

I'm angry that it was deleted and very upset, but not mad at you at all... it just frustrated me.. I am sorry if I am a little unnerved right now. I have spent hours and more hours searching for MGS4 and HS direct feed 720 clips online and cannot find them... it is driving me to madness...

That post was an effort to try and provide a little light humor and of course lower my blood pressure...

Could you please restore the entire post if you don't mind? I will edit it and add some appropriate text, but I don't think i could ever create that same post again from scratch...

Yes, it did mean a lot to me just because it was a stress relief valve...

Again, I am angry, mad, and furious but NOT at you! :-)

Your a great guy XBD, thanks for all yo udo.

xbdestroya

06-09-2006, 05:33 PM

MILR, I'm very sorry but the way the caching works in this browser, and the editing in general workd, I can't restore it. :(

xbdestroya

06-09-2006, 05:44 PM

MILR it's just not possible, at least to my knowledge.

If anyone has MILR's post though - the original - in their browser history, please feel free to PM me with the contents and I will restore it.

makeitlookreal

06-09-2006, 07:09 PM

The following is an attempt at humor and although some of my true feelings are mixed in many comments are exaggerated for effect. I personally think the PS3 is going to be a great gaming console, and think we have some surprises ahead of us. I hope you like the following. My first was BETTER and SHORTER, and this time I just could not find the creative spark needed to make something truly funny. If anyone has the original please post it.

RSxer Willz RuleZ: By BIRL (Better It Look REAL!)

(My not real but funny alter ego!)

I willz getts bak on da topix of da RSXes thee GPUer of the PS3 gamerz conSOUL. Zony Entertainzments put no fool in da PS3 for da GPUer. They have da mothaa uv all chipperz in the PS3. It'z a tru beastie ands is muchum more dan what yallz give da dog kredit for me broothers! For da Sake of mArIO HALT the talkz betreenz yall bout a mHz bOOst or DwonGrader. Mooch more is hapening mi pals!

Mans, pleeze stop with da deLETING uv me postz! Me and da man knowns far and widez as NEEVE DAMAGE "R" trYing to BRINGS "U" the TROOTH! Yallz do no wanna hearz that the PS THREE haz gottum a multiKORE GpU and dat U fakie "nOOBers" were allz jus DENIERS! Zony did notum zpend mmmiiilliioooNNZZ on the PS3 to ENclude a AWsoom Celler of a Proocesser to leaf out a jazzy GPUer! Haves you foLKs heard of the vISUaliZER man? Or are ye DEEF after allz this TIME? Nows we know da truth why ye POSTERS online all de time! Ye DEEF to the ViSUAlizer talk which pOWNs the 7800 pooh pooh chew toy! Woof Woof! My home dogger FIDO heere woun't tooch dat sorry silcone wannabe with hiz snout!

I smellz some pooh in dats yall shood know bETTER dan to think Zony woulds go soft on da GPUer. Are all ye underz those NDA spellz? Dem things are so suckie! Mi thinks yallz are not really nOObs after all but are in on da KONSPIRACY to keep da SuSPence moving till da launch happenz. But folkz it'z too late! LOL! ROLF! Becuzz NEEVE DAMAGE seen da goods and let his Feengers do da talking on deeze forums. He pOWNs yall KONSPIRATORZ becuzz he ExPOZED it allz from HIS sourcez! Zony RSX is da bomb and willz radicalize your lives with da more Bites for Memorize for TextTUREs and sub-proceZZors which will leaf the 360 in da dust!

My broothers and SiSters rememberz whats I zay here todayz. "B" cause I am GOING to SURPRIZE yallz at Launch by MAKINGZ "A" nother post to showz yallz dat NEEVE DAMAGE waz da MAN!

Zony will oWNZ dis generation and cause da 360 MASSIVE "D"-mage my homieze!

cpiasminc

06-09-2006, 07:49 PM

Wouldn't it be more fair to consider RSX's total number of ALUs in comparison to Xenos's total number of ALUs. If you were doing that way it would be RSX (8 Vertex Shader ALUs + 48 Pixel Shader ALUs) vs Xenos (48 Vertex/Pixel Shader ALUs)?
Sure if you were talking about raw processing power (and indeed RSX's is greater), but I was talking specifically about vertex processing. Even in some of the various "lockdown" modes, Xenos would end up effectively dedicating more ALUs to vertices.

So what is it apart from HDCP, that would be in a console and not a standard Geforce, whilst not being meaningful to the rendering capabilities? You are probably not allowed to answer this, but it is at least a hint that RSX=!G71.
Let's just say that some of the rumors are in fact true. I'll leave it at that. Well, that and that certain process technology changes demand some ancillary circuitry to combat problems inherent with scaling down a design.

Are you sure about this? I've read that the GPU manufacturers often only count the transistors in the logic circuitry, excluding the transistors used in memory cells (caches etc.). Probably because the transistors in memory cells can be packed much more dense and therefore make use of little die space.
Yeah, but for various testing purposes, you do need total figures, so either way they need to be able to know both logic transistors and totals (and it's not just transistors, either). There are also several ways to estimate based on what goes into a GPU and the die size... I know several people who are better informed on the electrical engineering side of things than I am whose judgments are usually in agreement on such estimates.

The transistor numbers split mentioned between the two dies in Xenos does not correspond very well to the actual sizes of the dies which can be seen in pictures of the Xenos chip package (see anandtech).
Yeah, but like you said, things like caches are more dense than other parts of logic, but in turn, caches don't scale down so well with process shrinks. Also DRAMs are not really superdense in terms of transistor layout (nothing is more per-transistor dense than SRAM, AFAIK). Okay, it may have been unfair to say "WAY" more, but whatever.

Also, version: do you actually know anything under NDA and that was some kind of hint from cpiasminc, or are you just playing with us?
I think he was just grinning because he agreed. Not because there was any specific truth to it. What I posted was more of an "if only" kind of thing. At least that's my guess. I can never really tell for sure what version means when he posts.

sif

06-09-2006, 08:09 PM

Thanks for the reply cpiasminc, insightful as always! I'm obviously not expecting a reply to this, but I wonder which rumours are true?

Just what are you wanting banned? Could you please be more specific?

Lol no, I don't remeber exactly what I had put, what I was joking about was I could not tell if all the pressure of waiting for news had taken its toll and had caused you to lose it and start writing teh l337 or if you were just kidding.

makeitlookreal

06-09-2006, 09:07 PM

Sif,

The pressure of waiting for news has caused me to go slightly crazy. Right now, I spend most of my time reading on here, on other forums, and searching the net for info. What frustrates me the most is that so many questions could be so easily answered by Sony, but instead we get a tribble of vague bits of information that don't tell us anything for certain. Every time a new bit of information comes out there are a thousand posts about it all over the net. It is spun one way and spun another way. I go searching and read a few dozen articles about the same thing.

All I want is for Sony to give us the full specs of the PS3 including the RSX without holding back. Obviously, Microsoft and Nintendo know what they are up too. They have billions of dollars and lots of resources at their disposal. Also, don't forget all the connections they must have. I don't think that revealing the specs would give Sony's competitors any significant advantage over them at this point.

It's getting closer and closer to launch but instead of MORE information we are getting less "solid" information and an ever increasing number of rumors. What I want is more solid information and less rumors so we can get past the entire hardware angle.

I am facinated with the RSX and want to know what exactly is under it's hood so to speak. But the way things are going now my brain will be deep fried before we ever find out....

I am loosing it a little, and I totally admit that.