STI Industrial Capacity

[xbdestroya]

Well it's a slowish Friday night in the forums so I thought I'd just toss something out there. This is another one along the lines of the 'component costs' thread in that it seeks to create an image of what might be going on behind the scenes with regard to the PS3.

What inspired this one is this info from the Playstation Meeting:

...only 450 PS3 dev kits have been shipped worldwide. Sony plans to release an additional 200 units in August and 300 more in September, but most developers probably will have to wait until October, when the company will ramp up its production to 3,000 units per month...

So what I'm wondering, is what's happening in October?

Is East Fishkill on a minimal production schedule right now? Is Nagasaki coming online? Will October also mark the month they begin stockpiling Cell chips for PS3 production?

I know we cannot find the exact answers to these questions, but any thoughts and/or insights are appreciated!

We can take it to the larger picture also and discuss estimated transition times to 65nm and any knowledge of upcoming chip revisions.

[F089/H]

Damnit,I thought that Nagasaki Fab was supposed to be up and runnning at full production by the end of July? oh and good find x,your good for a grabbie!:1337:

[version]

rsx...

[Metal Sphere]

Hmm, version could be on to something. The RSX isn't done yet, and isn't expected to be done until late August. That would explain the low dev kit production, and the late introduction of the final kit.

[Darkon]

hmmm... IBM Fishkill factory is apparently also producing R500 chips

or at least according to the inquirer

source (http://www.theinquirer.net/?article=24539)

[xbdestroya]

hmmm... IBM Fishkill factory is apparently also producing R500 chips

or at least according to the inquirer

source (http://www.theinquirer.net/?article=24539)

Well that article was actually retracted later; never right to begin with though. R500 production is over at TSMC in Taiwan.

@Version: RSX is a good point, but would they really want to delay dev kit shipments just for that reason? Not to mention that the dev kits containing RSX won't be shipping until December and beyond, as per Sony's slides. In fact Sony has mentioned that RSX mass-production won't begin until December before as well. Which leads back to why October for mass shipments?

[rpgamer_2k5]

Damnit,I thought that Nagasaki Fab was supposed to be up and runnning at full production by the end of July? oh and good find x,your good for a grabbie!:1337:

It is up and running, the Nagasaki fab2 is able to manufacture chips with a 90nm process and is establishing 65nm production line in the lower level of Fab2. The Cell serial-production is being initiated by IBM since they carry the most expertise. Since the Cell will stay at 90nm, it will mean that production should be starting earlier than expected.

I wonder where the 45nm process production line will be established on Sony's side. It will be available at the joint-operation Oita fab.

[xbdestroya]

It is up and running, the Nagasaki fab2 is able to manufacture chips with a 90nm process and is establishing 65nm production line in the lower level of Fab2. The Cell serial-production is being initiated by IBM since they carry the most expertise. Since the Cell will stay at 90nm, it will mean that production should be starting earlier than expected.

I wonder where the 45nm process production line will be established on Sony's side. It will be available at the joint-operation Oita fab.


Ok ok wait a minute, does anyone have a link pointing to the fact that Sony's new Nagasaki fab line is up and running?

[Darkon]

here you go source (http://www.xbitlabs.com/news/cpu/display/20040524160323.html)

[version]

Hiroshige's Goto Weekly overseas news*
Publishing the schedule of SCEI and the PLAYSTATION 3 development kit


--------------------------------------------------------------------------------

- Offering the appraisal kit with 2 steps

PLAYSTATION 3 which is released for the first time in the country

The SONY computer entertainment (SCEI), holding the conference "PlayStation Meeting 2005 which" is entire balancing of accounts of 1 year of PlayStation. Among those, PLAYSTATION 3 (PS3) concerning, several new informations were made clear.

First, as for the schedule of PS3 it was reconfirmed, that it is sale of 2006 spring, it became clear as a play vent in 2006 February immediately before the that, "for PLAYSTATION Conference" to be held.

In addition, the Hisashi of Representative President and group CEO of SCEI 夛 well the wooden health person did explanation concerning the development tool and that road map of PS3. The reference machine of PS3 is offered at 2 stages.

SCEI at the beginning, with machine "Cell for appraisal of Cell processor Evaluation System", Bali day of Cell and the software stack did Shaun. This is for the company internal laboratory with the machine for generally known debugging, but you say that it offered to one for appraisal, vis-a-vis the cooperative vendor of extremely part. With 2.4GHz, loading operational frequency of the Cell processor which it loads XDR DRAM 256MB. The board of NVIDIA was placed to graphics.

The appraisal machine which designates that SCEI calls "PS3 Evaluation System" next, offers to the customer as main purpose was developed. The type turn, "CEB-2030" due to this machine, the code name is made "Cytology". SCEI, vis-a-vis the software vendor with this machine works the fact that it is distributing, from this spring. You explain the summary of PS3 Evaluation System afterwards, but the basic specifications have become Cell 2.4GHz, XDR DRAM 512MB and GeForce 7800 GTX (G70) with.

When it is 2005 December, SCEI almost becomes equality constitution to the PS3 apparatus, it is the schedule which offers "PS3 Reference Tool". Cell 3.2GHz and RSX (Reality Synthesizer), with XDR DRAM 512MB, also BD drive is loaded. Form in present schedule is 2U size of rack-mounted type, but you say that it considers also vertical ranging.

SCEI is the schedule which consecutively keeps offering PS3 Evaluation System to November. Presently, as for PS3 Evaluation System 450 units are put out, but like under it keeps increasing the offer quantity in schedule. With that, you say that you answer the demand to the reference machine which it is intensive.

PS3 Evaluation System distribution schedule
August 200 units
September 300 units
October 3,000 units
November 3,000 units or more



Schedule to PS3 appearance The apparatus of PS3 Evaluation System PS3 Evaluation System distribution schedule



Machine "First for appraisal of Cell processor Cell Evaluation System" The appraisal machine "PS3 Evaluation System which" designates that it offers to the customer as main purpose Almost, it becomes the constitution which is equal to PS3 "PS3 Reference Tool"


- As for Cell and XDR DRAM 75% efficiency of PS3

With PS3 Evaluation System, PS3 last specification there is various differences.

First, the operational frequency of Cell has become 75% frequency of 2.4 GHz and product edition. In case of CPU, until Bali day Shaun ends, it is not the unusual thing where in the appraisal machine frequency is held down. Naturally, the last specification suitable efficiency of PS3 is the case that it does not come out, also method of thinking where is possible if, but as for this 75% efficiency you may suppose simply.

Main memory is connected with XDR DRAM, the XDR DRAM interface which is loaded onto Cell (XIO). This is not the full specifications. With PS3 Evaluation at June point in time System, also the data transfer rate of XDR DRAM was held down to 2.4Gbps. Because the transfer rate of XDR DRAM which PS3 loads is the schedule of 3.2Gbps, it means this 75% rate.

When the CPU clock is dropped, being linked, as also the transfer rate of XDR DRAM has fallen, it is visible. The fact that this has shown is the possibility the CPU core and XDR DRAM interface of Cell becoming the synchronous design. Same period there is an advantage the one which it could point the design becoming easy. Especially, in case of CPU- memory because it becomes the element whose レイテンシ is important, same period as for the advantage where it can point it is many.

Though, as for dropping the transfer rate of XDR DRAM, there is also a possibility of considering the yield rate of XDR DRAM which is new memory. At early stage perhaps that also XDR DRAM takes the 3.2Gbps item according to schedule it is difficult. When the core clock of the DRAM cell (Internal Column Frequency) you think, XDR DRAM 3.2Gbps is rather hard. XDR DRAM moves to 90nm process the time where the frequency of mass production of PS3 increases, but under present conditions also in the aspect of speed yield rate it is disadvantageous to be 100 - 110nm process. In addition, when we assume that with PS3 Evaluation System, RIMM (the memory module of Rambus) you use, perhaps amount of module, it influences also the fact that timing margin is eaten.

XDR DRAM 512MB it loads PS3 Evaluation System which is announced in the latest conference, in main memory. Because in PS3 specification it is XDR DRAM 256MB, the memory quantity reaches 2 times. As for this there is also a possibility of being increased with RIMM. At June point in time, as for PS3 Evaluation System when it is designed in such a way that you can use also RIMM it was explained. You say that is, because also verification of mass memory constitution does.

The interface of XDR DRAM pliability is high with コンフィギƒ…ラブル. It is x16 interface, but it is setting possible even in x8 and x4. As for XDR DRAM as for the data line point-to-point connection (between Cell and the DRAM tip/chip) is, but for example, x16 per channel 1 the DRAM tip/chip of 2 times can be connected by the fact that to x8 it sets modifies. RIMM of XDR DRAM although by the fact that this quality is utilized, being point-to-point connection, can connect 2RIMM to channel 1. By the way, with PS3 of last edition, as for XDR DRAM memory direct it is attached to the motherboard.


PS3 System and PS3 Evaluation System
As for PDF editionthis


- Under present conditions graphics of PCI Express x4 connection

With PS3 Evaluation System, "GeForce 7800 GTX (G70)" for PC it is on-board in graphics. The media processor "RSX which is loaded onto PS3 (Reality Synthesizer)" it is substitution. G70 is presumed that with the almost same Shader constitution as RSX, also architecture inside Shader resembles rather. Because of that, around the graphics, developing the software in the G70 base, it probably will not that much cause problem. シェーダプログラム it is the expectation which can use completely same ones.

However, as for G70 because operational frequency is lower than RSX, there is a performance it falls. But, the fact that difference is larger than performance inside GPU is interface.

With PS3, Cell and RSX, the parallel interface "FlexIO which Rambus was developed (Redwood: The redwood)"with 35GB/sec (it descends and 20GB/sec/ rising 15GB/sec) is connected with wide band. But, because it is not mounted, Cell you cannot connect FlexIO to G70 which is the PCI Express x16 tip/chip directly.

Then, with PS3 Evaluation System, as for G70 it is connected with southern bridge tip/chip and PCI Express. With PS3 Evaluation of June point in time System, with PCI Express x4 it was connected. Because as for this, as for the southern bridge tip/chip which is loaded onto PS3 Evaluation System, the I/O chip which is loaded onto the Cell work station of IBM basically they are same ones. Because of that, PCI Express of peripheral I/O for the server x4 is mounted on the same tip/chip. With the southern bridge tip/chip of last edition as for PCI Express it is the schedule which goes out, but. Under present conditions as for G70 it is connected there.

Because of that, under present conditions either native PCI Express of G70 x16 is not utilized. Cell and the southern bridge tip/chip on the specifications have meant to be connected with FlexIO of 5GB/sec. When we assume, that even with PS3 Evaluation System it is similar, first zone becomes more markedly than apparatus specification thin here. Furthermore, the southern bridge tip/chip and between G70 being to be PCI Express x4 with 2GB/sec, here are thin. Cell -> When it compares with the zone of GPU, PS3 Evaluation System only zone of 1 of 20 parts of PS3 means not to have.

According to SCEI, with PS3 Evaluation System, you say that also memory of graphic side is increased to GDDR3 512MB. In the apparatus of PS3, as for video memory it is the schedule which becomes GDDR3 256MB. Because you have taken memory of video side more largely than PS3, when the bus has been less crowded, transferring the material to video memory side, you say that also the correspondence that is achieved, the buffer it does. But, as for the cooperation of Cell and RSX only of PS3 which utilizes wide bus zone, with PS3 Evaluation System appraisal is difficult.

By the way, if as for the GDDR3 interface of RSX vis-a-vis being 128bit width, as for G70 being to be 256bit width, the 512Mbit DRAM tip/chip of the same x32 the memory airlift reaches 2 times.

- As for merit of PS3 architecture the cooperation of Cell and RSX

Major feature of PS3 graphics is the cooperation of Cell and RSX. The architecture of RSX itself is G70 and resemblance, but host interface differs from G70 for PC completely. In case of G70 with PCI Express x16 8GB/sec (type direction 4GB/sec) with it is connected by chip set, G70 cannot do direct access to either main memory. RSX 35GB/sec (descends vis-a-vis that and 20GB/sec and rising 15GB/sec) with Cell is connected directly, can designate also rendering directly as main memory of Cell side.

When method of using GPU to which this difference is large, because of that differs from PC architecture completely is possible you explain SCEI. First, because the bus is wide, forming the geometry object of enormous number with Cell, method of using that is possible transfers that apex data steadily. It can make also easy conversely from RSX side once to reset the data to Cell.

"Can the Cell processor, use in pre- processing and post processing. For example, tessellation (plane surface division) and it seems like dot filling. Doing physical processing on Cell, motion and collision it calculates, that it can do also trance form to do apex array in the base, "that David B. of NVIDIA Kirk (the デビッド ･ B ･ kirk) the person (Chief Scientist) it explains.

As for SCEI, Cell does the part whose degree of abstraction is high as basic thought, is small (apex and the pixel unit? ) You call processing that the allotment that is supposed, GPU does. As for this being rational, for example, when metamorphosis of geometry is done in the CPU side, problem does not occur with the hitting decision which becomes important with the game. When it metamorphoses on GPU side, if the data is not reset to CPU side, because original object and form differs, the gap occurs with hitting decision. In case of PS3, also metamorphosis is possible on Cell side and, assuming, that it metamorphosed with GPU, it can make relatively easy to reset the data to CPU side.

With former architecture, CPU either one of GPU becoming the bottleneck, unless it can process, it could not do the thing above that. Vis-a-vis that, in case of PS3 architecture, when GPU the neck is to Cell, Cell side is the neck, you say that is possible the fact that load is made to allot to GPU. For example, making graphics process more on Cell side depending upon the software, it improves expression, because or we would like to cut that conversely, Cell side more in physical processing, you explain graphics that the adjustment that becomes easy, it can be entrusted to GPU. In other words, it is the case that, flexible between two programmable processors, CPU and GPU balance adjustment becomes possible.

In case of former PC architecture, because the pipe of CPU and GPU is restricted, the number of geometry objects being limited, improved the technology how much you show to rich in the range of. Vis-a-vis that, in case of the game console like PlayStation 2, conversely polygon it tried to be able to make in large quantities, but after that expression power about PC was not high. In case of PS3 architecture, that both being possible, it is the case that furthermore it can take balance softly.

But, in case of PS3 Evaluation System which presently is offered, from restriction with respect to architecture, you cannot verify the balancing such as that. This is difficult point, but if you say opposite, software demonstration with the present system means not to utilize the latent characteristic of PS3 to full yet. In the actually PS3 system, there is a possibility of being possible the thing above demonstration of present condition.

Though, if you say with just bus zone, when with 21.6GB/sec, you compare also between CPU-GPU of Xbox 360 with PC, it is wide markedly. In the next generation machine, as for CPU-GPU connection of wide band, is not the merit just of PS3.

With PS2 the simple loader started from the firmware, loaded OS and the library from the disk. Vis-a-vis that, with PS3 the hyper visor first starts from firm. The hyper visor "VMM (Virtual Machine Manager)" it is one kind of software, but is not on OS and completely the machine you imagine convert under OS layer. In case of PS3, case you use independently for game play Cell OS, the hyper visor starts by all means, OS of former definition (guest OS) stands up on that. OS is the image which becomes 2 layer constitution which also hyper visor layer includes. As for this basic OS class you say that even with PS3 Evaluation System it is similar.

[Z]

I think they can manage early devkits with an SLI GPU setup. I am thinking Cell isn't that friendly to mass fab at this stage. my guess is Nagasaki needs a little more time to get things going. after that, it is mass time! that is why i think their is a significant jump from 300 to +3000. don't forget that Nagasaki will also be handling RSX production.

[Viper]

It would seemt hat the next few months are more for testing and increasing the efficiency capabilities before ramping up production.

Perhaps they are doing it to coincide with RSX production capabilites.

[xbdestroya]

here you go source (http://www.xbitlabs.com/news/cpu/display/20040524160323.html)

*thumbs up*

So true! I'm going to have to research to get all my Sony fab interests in order. Nagasaki... Toshiba joint fab... capacity agreement at East Fishkill... gotta read up, 90nm vs 65 and other aspects.

[rpgamer_2k5]

Ok ok wait a minute, does anyone have a link pointing to the fact that Sony's new Nagasaki fab line is up and running?

It looks like Version beat me to it. :D

The 65nm line is what Sony was trying to get online but that alot more time was needed. The 90nm line is already up and the PSP CPU is using this process. However getting the Cell manufactured at Nagadaki isn't an easy task even with the infastructure ready. It looks like Sony is going to be doing a technology transfer from the Fishkill/Oita to Nagasaki since they don't seem to have the expertise to do it on their own.

[makeitlookreal]

Hello Everyone,

This article is very facinating and leaves me with a few questions...

Does this mean that it is possible Sony is waiting until the 65nm fabs are up and running to start mass producing the PS3 cell chip and/or RSX?

It seems to me that if they were able to fab RIGHT NOW at 90nm they would be doing so in mass. However, it seems that they are waiting. Apparently, they can fab now, but are waiting.

Could they be waiting for the 65nm process?

If so, that would be a devestating blow against Microsoft.

Can you imagine a potentially eight SPE cell chip or at the very least perhaps a 4.0 cell chip?

Also, do you think that since the dev kits have one gigabyte of RAM that the final PS3 might have this too?

Personally, if the new PS3 just had a gig of RAM and a 4.0 cell chip with all eight SPE's I would be very happy.

I think the news would destroy the XBox's launch.

[rpgamer_2k5]

4.0Ghz Clock Speed for Cell
1 GB of Total RAM

If that happens, I wouldn't be able to express my happiness in words. However fate is not feeding the greedy ones like us so the chances of it happening is quite low.

The SDKs usually have higher memory becuase of the debugging and the other diagnostic tools. The reason why we might be seeing any progress with the Nagasaki Fab2 is because this will be Sony's first time manufacturing such a CPU. Even the EE of the PS2 was initially manufactured at Toshiba's Fab while the GS was manufactured at Nagasaki. Sony will be taking part in a very ambitious project which has huge risks but even bigger rewards. Now if Sony could activate the 65nm line (doubtful) then it will give Sony a big edge.

[tazz3]

450 kits is not alot.
how will this make games in a short time for the PS3.
the only thing i see is the PS3 launching in the usa
next may. there is not enough time for a march launch

[Gounmckuber]

The japanese will buy it anyway. They can get a cheap blu ray player and they know that ps3 will bring the gaming goods later on

[makeitlookreal]

I really do feel that Sony is going to boost the specs of the PS3. Too many of them are basically the same (in appearance at least) to the XBox 360.

Both the XBox and PS3 are clocked at 3.2ghz for the CPU and have 512 MB of RAM.

Now, I know that the ghz of the two systems are totally different. A cell chip is different than the xenon.

But the two things that I believe would be potentially easiest and obvioius to boost are the cell and ram.

Even if just the 90nm fab is utilized I am almost certain there will be some small bump in ghz. Perhaps only to 3.6ghz, but something to make the PS3 a little faster (if only in the eyes of customers) than the XBox.

However, if they do decide to wait a little longer and use the 65nm fab then probably a 4.0ghz and eight SPE cell chip is an obvious upgrade.

And about the RAM... I have heard that if anything could be a bottleneck on the PS3 it could be the RAM. I have been reading about how to capture high resolution textures and so fourth you must have a lot of RAM. With RAM prices always going down and the mass bulk of RAM Sony will be buying for the PS3 my guess is that they are going to boost the RAM.

Even if they just gave the cell 256 it would be nice for the GPU to have 512. At least this way the system would have a marked boost over the XBox.

What are your thoughts on this? Personally, my philosophy about the PS3 is that even if Sony has to postpone it's release for a period of time having the 65nm fab, more RAM (due to lowered RAM costs), and other upgrades (such as more work on the RSX) would be worth it.

Additionally, it would allow developers to have more time making first gen PS3 games with their dev kits for a fantastic launch of the system.

[version]

4.5 GHZ cell
4.5 GHZ xdr and redwood
550 rsx
and 256MB xdr for cell +512 MB xdr for rsx

and iam very happy :)

[Gounmckuber]

Hmm...My hopes for ps3 are:256 bit bus, 768mb for rsx, 8th spe enabled, 4.5 ghz Cell, 800 mhz Rsx and 512mb for Cell :tada: :splitspin :lick:

[Viper]

I hope you can afford to spare some flow for the rest of us.

[Domination]

I really do feel that Sony is going to boost the specs of the PS3. Too many of them are basically the same (in appearance at least) to the XBox 360.

Both the XBox and PS3 are clocked at 3.2ghz for the CPU and have 512 MB of RAM.

Now, I know that the ghz of the two systems are totally different. A cell chip is different than the xenon.

But the two things that I believe would be potentially easiest and obvioius to boost are the cell and ram.

Even if just the 90nm fab is utilized I am almost certain there will be some small bump in ghz. Perhaps only to 3.6ghz, but something to make the PS3 a little faster (if only in the eyes of customers) than the XBox.

However, if they do decide to wait a little longer and use the 65nm fab then probably a 4.0ghz and eight SPE cell chip is an obvious upgrade.

And about the RAM... I have heard that if anything could be a bottleneck on the PS3 it could be the RAM. I have been reading about how to capture high resolution textures and so fourth you must have a lot of RAM. With RAM prices always going down and the mass bulk of RAM Sony will be buying for the PS3 my guess is that they are going to boost the RAM.

Even if they just gave the cell 256 it would be nice for the GPU to have 512. At least this way the system would have a marked boost over the XBox.

What are your thoughts on this? Personally, my philosophy about the PS3 is that even if Sony has to postpone it's release for a period of time having the 65nm fab, more RAM (due to lowered RAM costs), and other upgrades (such as more work on the RSX) would be worth it.

Additionally, it would allow developers to have more time making first gen PS3 games with their dev kits for a fantastic launch of the system.

Even if the Cell was shrunken down to the 65nm, it still wouldn't have 8 functioning SPEs. It will still be seven unless Sony originally designed the processor with nine or ten SPEs in the begining. People seem to think Sony suddenly decided to disable an SPE when they realized the cost it would take to put everything in the console. That's false. The Cell was never designed to use 8 SPEs in the first place. That is only what people assumed to be the case. But anyway, when Sony first designed the Cell, they knew it'll be difficult to build a perfect processor without a single defect, so an extra SPE was added incase one of the other seven SPEs turned up defected during manufacting. Plus, the Cell isn't the only processor to follow this trend neither.

Furthermore, I don't see what the big deal is anyway on whether it has 7 or 8 SPEs. If what you are already seeing is great, what does it matter? I doubt you'll be thinking of the amount of SPEs in the unit when playing a preferred title of yours. It only seems important because it was mentioned. Now if you had actually seen the 8th SPE working in its full glory before it was disable, then I'd see your point. But right now, you're walking in a dark room with no idea as to how much of a leap an extra SPE would have granted you. You forget, people were calling Gundam and Killzone an FMV, and that's only those two, never mind the others.

[xbdestroya]

4.5 GHZ cell
4.5 GHZ xdr and redwood
550 rsx
and 256MB xdr for cell +512 MB xdr for rsx

and iam very happy :)


I like your style Version, you always take it that extra step beyond. :spiny:

[xbdestroya]

Hello Everyone,

This article is very facinating and leaves me with a few questions...

Does this mean that it is possible Sony is waiting until the 65nm fabs are up and running to start mass producing the PS3 cell chip and/or RSX?

It seems to me that if they were able to fab RIGHT NOW at 90nm they would be doing so in mass. However, it seems that they are waiting. Apparently, they can fab now, but are waiting.

Could they be waiting for the 65nm process?

If so, that would be a devestating blow against Microsoft.

Can you imagine a potentially eight SPE cell chip or at the very least perhaps a 4.0 cell chip?

Also, do you think that since the dev kits have one gigabyte of RAM that the final PS3 might have this too?

Personally, if the new PS3 just had a gig of RAM and a 4.0 cell chip with all eight SPE's I would be very happy.

I think the news would destroy the XBox's launch.


I'm wondering about this same sort of thing, fab wise. It seems completely impossible that they would wait for 65nm, or have it soon enough to meet their launch window; yet at the same time I can't undertsand what the hold-up is on the present 90nm process in getting these dev kits out.

When I have a chance I'm going to compile everything out there known about these fabs and their production schedules.

In the meantime, best be forgetting about that gig of RAM Makeitlookreal! :alien:

[makeitlookreal]

First of all, the cell chip was indeed designed to have all eight SPE's working.

Now, I am not an expert, but this is what I understand about the FAB process...

Some chips will have eight SPE's working, but these will be few. Others will have one defect and will have seven SPE's. Others might only have six working SPE's.

THEY WERE DESIGNED TO HAVE EIGHT SPES!

The reason that the PS3 is going to have seven is because to only use the eight SPE chips would be waste.

They can save the eight SPE chips for something even more expensive. Such as servers, etc.

I however hope that if their yeilds increase that they will include the eight's SPE.

[makeitlookreal]

Also, I do not see why the PS3 might not have one gig of RAM. RAM prices are going down all the time.

[xbdestroya]

Also, I do not see why the PS3 might not have one gig of RAM. RAM prices are going down all the time.

Yes I agree, but they go down to sort of a floor in the pricing, not just down down down forever. PC-133 SDRAM is going to be costing you more than PC-3200 DDR, for example; and now we're reaching a cross-over point where DDR-2 is becoming less expensive than DDR.

GDDR-3 and XDR are much more expensive than these, but certainly Sony has taken into account their inevitible price drops into their future pricing/costs considerations.

Anyway it just comes down to my not thinking a gig of RAM is in any way likely; if they throw in a gig of it, you can make sure that I'll be there backing you up when you lay claim to having been the first to call it. :smoke:

[Schmeh]

Also, I do not see why the PS3 might not have one gig of RAM. RAM prices are going down all the time.

The reason is twofold. The first and biggest reason why the PS3 won't have 1GB of RAM (because you said RAM and not VRAM, I assume you mean main memory and not total.) is price. Yes the cost of DDR is falling all the time, XDR is a new technology and is owned by one company, Rambus. Since XDR is not used in that many instances yet, the price is going to be relatively high. The second reason not to have 1GB of ram is neccessity. Granted, developers could find a way to use that much memory, it is not necessary at this point in time.

[Z]

can anyone tell me the reason Sony split the memory in two? is there an advantage over one unified memory-ala X2? I am sure there is a good reason for this, but I can't put my fingure on it.

[Schmeh]

can anyone tell me the reason Sony split the memory in two? is there an advantage over one unified memory-ala X2? I am sure there is a good reason for this, but I can't put my fingure on it.

There are advantages both ways. For unified (Xbox 360), if the GPU needs more than 256MB, it has direct access to more, it doesn't have to go through the CPU's memory controller. The catch is that there is only one memory controller in this kind of setup and in the case of the 360 it is on the GPU, meaning that both the GPU and the CPU will have to go through this everytime they need memory. Combine a triple core CPU and a high performance GPU and you need alot of bandwidth to feed both of these simultaneously. The 360 has 22.4GB/s from the memory to the controller on the GPU, now keep in mind that a 7800GTX has 38.4GB/s of bandwidth for it, and you can see that there may be some issues with the amount of bandwith.

[Domination]

First of all, the cell chip was indeed designed to have all eight SPE's working.

Now, I am not an expert, but this is what I understand about the FAB process...

Some chips will have eight SPE's working, but these will be few. Others will have one defect and will have seven SPE's. Others might only have six working SPE's.

THEY WERE DESIGNED TO HAVE EIGHT SPES!

The reason that the PS3 is going to have seven is because to only use the eight SPE chips would be waste.

They can save the eight SPE chips for something even more expensive. Such as servers, etc.

I however hope that if their yeilds increase that they will include the eight's SPE.

You completely miss my point. We are talking about the PS3, not some expensive server. If this Cell is strunken down to the 65nm, it's still not going to contain 8 functing SPEs because it was never in their architecturing limits in the begining. It would be the same as me saying 2 or 3 SPEs being defected and Sony still putting the processor in the PS3 console. If something is over or beneath the limit in which they previously set to manufacture the processor for the console, it will not be used.

[makeitlookreal]

Domination,

If you had read the interviews with Ken (don't know how to spell his last name) of Sony you would see how the only reason they are going to use seven SPE cell chips in the PS3 is because doing so is more economical. They would like to use an eight cell SPE chip (actually, all cell chips are eight SPE but the truth is that any chip could randomly have a defect in any of the eight SPE's) but if they did then they would only be able to utilize a small percentage of the total chips produced.

If they shrank down to 65nm and the yields of cell chips with eight working SPE's went up (just making up numbers here) from 30% to lets say 60% then obviously they would probably use eight SPE chips in the PS3.

The PS3 was NOT designed just to utilize one SPE, four SPE, eight SPE, or seven SPE.

The PS3 could utilize any standard of cell chip they desire. For goodness sakes, if yields are really bad we might see a six SPE chip! But to you that would be IMPOSSIBLE because it was "designed" to use a seven SPE chip.

It was NOT designed to use any certain number of SPE units.

If the yields are good enough there is no reason that an eight SPE chip could not be utilized.

Do you really think that Sony would purposely make their console less powerful than it could be if the yields turned out to be so good that it would not cost them anything more?

......

Also, to a previous poster... when I say a gig of total RAM what I really meant was 512 XDR and 512 VRAM.

However, even a simple boost to ONLY the VRAM would be sufficent enough to bottleneck proof the RAM issue.

Right now 256mb of RAM is "enough" to play just about any computer game and a console can do more with less RAM.

But a few years from now 512MB of VRAM would become VERY useful.

[rpgamer_2k5]

Hmm..what you are saying is correct. Right now I don't want to employ too much optimism regarding the manufacturing process. The 90nm fab process is available, the slow down could have 2 explanations:
1) Since Sony has not manufactured such an advanced microprocessor, they cannot start manufacturing the chip as soon as IBM. (More Likely)
2) Sony is almost completing the construction of the clean room in the first floor which will be employing the 65nm fabrication process. (Less Likely)

The reason I don't see (2) happening is because Sony is not a manufacturing pro in this market. We have seen how poorly they did fabbing the EE+GS solution with the 90nm process. I would lower your expectation on that one but if it is possible then it is only better for the PS3. One thing we do know is that Sony and Toshiba has manufactured a 32Mbit eDRAM module under the CMOS5 (65nm) process. Now here comes the speculation; Toshiba has licensed the Synopsys PSM technology which helps in increasing the yield and maintain the performance of chips manufactured under the CMOS5 design process.

Some question arise:

1) Does this mean Sony will also do the same?

2) Will Toshiba's fab OTTS or joint-Sony/Toshiba cleanroom ('virtual' fab space) cater the Cell demand first?

3) If this occurs, what type of jump in performance will see if the Cell and RSX are manufactured under the 65-nm design process?

Take a look at this:
http://img.cmpnet.com/eet/news/03/september/FOS/one_world_yh_2.gif

Toshiba is also pouring alot of money into the 65-nm design process at their own fab and they are aiming for a 2005 operationalization like Sony. Moreover IBM's Fishkill fab will be employing the 65-nm design process in a similar time frame. This might explain why Sony is investing heavily into OTTS and Fishkill but with IBM and Toshiba using most of the line for their own products (ie. Cell CPU for Servers and HDTVs respectively).

[Domination]

Domination,

If you had read the interviews with Ken (don't know how to spell his last name) of Sony you would see how the only reason they are going to use seven SPE cell chips in the PS3 is because doing so is more economical. They would like to use an eight cell SPE chip (actually, all cell chips are eight SPE but the truth is that any chip could randomly have a defect in any of the eight SPE's) but if they did then they would only be able to utilize a small percentage of the total chips produced.

If they shrank down to 65nm and the yields of cell chips with eight working SPE's went up (just making up numbers here) from 30% to lets say 60% then obviously they would probably use eight SPE chips in the PS3.

The PS3 was NOT designed just to utilize one SPE, four SPE, eight SPE, or seven SPE.

The PS3 could utilize any standard of cell chip they desire. For goodness sakes, if yields are really bad we might see a six SPE chip! But to you that would be IMPOSSIBLE because it was "designed" to use a seven SPE chip.

It was NOT designed to use any certain number of SPE units.

If the yields are good enough there is no reason that an eight SPE chip could not be utilized.

Do you really think that Sony would purposely make their console less powerful than it could be if the yields turned out to be so good that it would not cost them anything more?

......

Also, to a previous poster... when I say a gig of total RAM what I really meant was 512 XDR and 512 VRAM.

However, even a simple boost to ONLY the VRAM would be sufficent enough to bottleneck proof the RAM issue.

Right now 256mb of RAM is "enough" to play just about any computer game and a console can do more with less RAM.

But a few years from now 512MB of VRAM would become VERY useful.

I'm not sure what Sony plans are for the RSX, but it seems to be following this trend as well. And it isn't just the Cell or the RSX that follows this. I forget who designed the processor, but one of the Sony represenitives compared the Cell to this processor as far as disabling certain parts of the chip for possible future defects go. I'm sure CP can shine a little more light on this issue. But for now, I will say that you are making it seem as if Sony just disabled an SPE in the heat of the moment to cut cost, when infact that was never the case. It has little to do with whether the PS3 can handle an 8th SPE or not. It had more to do with the base inwhich they were targeting.

Now could Sony have included a ninth SPE so that they can have eight functioning SPEs in the console? I think if that was ever the case, they probably would have done it. But seeing how it wasn't done, most likely it was due to it just not being possible to get anything other than eight SPEs to actually work, with the eighth one just plain pushing the limits of the architecture in something as massively spreaded as a console.

[xbdestroya]

Rpgamer, definitely some good insights. I agree that it's probably the case that Sony is still in the stages of ramping up their own fab for mass production, and may just be taking the attitude now of "we'll get 65nm when we do" in order to go ahead with 90nm production. Afterall we already know they intend to produce the RSX at 90nm at Nagasaki, so obviously they're not waiting for the 65nm line to come online.

As for the Synopsys PSM deal I wasn't aware - good to know.

Sony is indeed investing a lot into the joint fab with Toshiba and IBM's East Fishkill. I know at least with Fishkill, part of that agreement is that Sony has rights to a portion of the capacity once 65nm comes on - probably the same situation with Toshiba, but I don't know for certain.