Details trickle out on CELL processor...

[j^aws]

http://www.eet.com/semi/news/showArticle.jhtml?articleId=54200580

:P

[PeanutButterMunky]

The highly integrated Cell device has been billed as a beefy engine for Sony's Playstation 3, due to be demonstrated in May. But the architecture also addresses many other applications, including set-top boxes and mobile communications. Workstations fitted with the Cell architecture — a $2 billion endeavor — are already in the hands of game developers.That's good news =)

At root, the Cell architecture rests on two concepts: the "apulet," a bundle comprising a data object and the code necessary to perform an action upon it; and the "processing element," a hierarchical bundle of control and streaming processor resources that can execute any apulet at any time.

The apulets appear to be completely portable among the processing elements in a system, so that tasks can be doled out dynamically by assigning a waiting apulet to an available processing element. Scalability can be achieved by adding processing elements.That's what I was thinking ;)

Each processing element comprises a Power-architecture 64-bit RISC CPU, a highly sophisticated direct-memory access controller and up to eight identical streaming processors. The Power CPU, DMA engine and streaming processors all reside on a very fast local bus. And each processing element is connected to its neighbors in the cell by high-speed "highways." Designed by Rambus Inc. with a team from Stanford University, these highways — or parallel bundles of serial I/O links — operate at 6.4 GHz per link. One of the ISSCC papers describes the link characteristics, as well as the difficulties of developing high-speed analog transceiver circuits in SOI technology.Holy CRAP!!

But UNC's Zimmons has his doubts. "I believe that while theoretically having a large number of transistors enables teraflops-class performance, the PS3 [Playstation 3] will not be able to deliver this kind of power to the consumer," he wrote in response to an e-mail query from EE Times. "The PS3 memory is rumored to be able to transfer around 100 Gbytes/second, which would mean it could process new data at roughly 25 Gflops (at 32 bits) — far from the 1-Tflops number."Hmmm...I guess we'll just have to wait and see...

[Xeno]

Well apparently the 25Gflops calculation is untrue...who knows i'm no good with this tech stuff :oops:

http://www.beyond3d.com/forum/viewtopic.php?t=18554&sid=0026b93ff62b7ddb2d4ad560 20193576


Vince wrote:

Also, I wouldn't put much faith in Mr. Zimmon's comments, he's doing the same bandwith/flop math that Deadmeat used a year or so ago. External bandwith isn't an indication of calulation ability (ala contemporary GPUs).


Vince wrote:

Exactly, but what was funny is that your selection could have fooled most of us for Deadmeat. You picked out a comment from a PhD from UNC-Chapel Hill that used the exact same, falty, logic that Deadmeat did and we've covered here several times.

And he's talking about off-chip. On-die bandwith will be enormous if the SRAM is running at 4.8GHz and follows the patent. Seriously, haven't you seen his PDF? it's been around for a bit. That's likely why he was asked, not that it's something many on this board couldn't have made.



V3 wrote:

That's the rumoured about the PS3 off chip memory, which is rumoured to be those high frequency Rambus memory. Well not really rumoured anymore since Rambus made a statement about PS3 using them. Its just the configuration is unclear

[stanDarsh]

It can't be 25GFlops/s! Let's think about this logically for a minute. I doubt Sony, IBM and Toshiba are that silly. Why would a company spend over $2 Billion in R&D for a processor that can only produce 4x as many GFlops/s as its predecessor (25 compared to 6.2 GFlops)? The simple answer is you wouldnt. It doesn't make a lot of business sense. Why would you spend that amount of money for such minimal gains, when you could do what Microsoft did and just get IBM to design it, and pay them a one off fee for the design?

Oh well I guess we'll just have to wait til February to find out for sure! :wink:

[Xeno]

http://www.gfdata.de/gamefront-temp/sonycell2.pdf

maybe not 25 Gflops but what about... 16 teraflops :shock:

[ssssss]

http://www.gfdata.de/gamefront-temp/sonycell2.pdf

maybe not 25 Gflops but what about... 16 teraflops :shock:


Atleast wait for another Decade to put that much in a Console.

[Ferrismc]

How much Flops will Xbox2 do ? How much Flops did we expect from the PS3 so far ?

[stanDarsh]

I have my doubts about the authenticity of that pdf file. Kinda looks like a mockup from some fan site. I could be wrong, but it just looks amateur to me.

[PlayBoy]

I have my doubts about the authenticity of that pdf file. Kinda looks like a mockup from some fan site. I could be wrong, but it just looks amateur to me.

See the actual press releases from SCE 8) They mirrored them, perhaps?

http://www.scei.co.jp/corporate/release/pdf/041129ae.pdf
http://www.scei.co.jp/corporate/release/pdf/041129be.pdf

[The_One]

I have my doubts about the authenticity of that pdf file. Kinda looks like a mockup from some fan site. I could be wrong, but it just looks amateur to me.

See the actual press releases from SCE 8) They mirrored them, perhaps?

http://www.scei.co.jp/corporate/release/pdf/041129ae.pdf
http://www.scei.co.jp/corporate/release/pdf/041129be.pdf16TFLOp/s is the performance for the CELL Workstation, is it not?
How much Flops will Xbox2 do ? How much Flops did we expect from the PS3 so far ? Xenon... according to deadmeat: 32GFLOps/s. However, he gave PS3's CELL a mere 128GFLOp/s for his estimate, so don't pay deadmeat any attention. Other then that, there's no official info on the Xenon's performace

Atleast wait for another Decade to put that much in a Console. 16TFLOp/s is for the workstations, so don't worry, we won't see that much number crunching processing power in the PS3 itself ;).

[j^aws]

Oh shit...there's stuff posted here too! :lol:...How confusing! :D

Edit: GScube had like '16' blades so to speak of PS2s...If the Cell workstation is the same with 16 Blades for 16TFlops, then perhaps PS3 can be 'one' blade'?

[GUNDAMSEED]

hhmm 2005 i can hardly wait . hopefully some more stuff will come since all thoses things. seem a bit funny in a way .

[cpiasminc]

"The PS3 memory is rumored to be able to transfer around 100 Gbytes/second, which would mean it could process new data at roughly 25 Gflops (at 32 bits) — far from the 1-Tflops number."
Hmmm... if Rambus says that its Yellowstone memory connections run at an effective speed of 6.4 GHz (really, 6.4 GT/s, you'd need massive current to wiggle voltages across the motherboard at 6.4 GHz), that means that 100 GBytes/sec transfer rate means only a 16-bit memory bus (102.4). I suppose that would cut down on pins and wire traces, but I think that something massively parallel would benefit from more bandwidth... a lot more. A 32-bit memory bus is still very reasonable within the tiny confines of a console motherboard. Unless they're intending to make something really Gamecube-sized -- that is, considering the fact that the CPU and GPU would have their own independent memory ports.

Either way, it's not like memory bandwidth is a very significant measure of performance unless you're really choked or you're doing something that is very data-dependent. Especially when you have something that is designed for repetitive tasks. Even if it was a major problem, I think this is the sort of thing where the eDRAM would come in handy. A lot of data can be buffered within eDRAM for fast access, and you could get much higher effective throughput since you can use a lot of data over again for multiple calculations. The eDRAM is effectively cheap cache in that respect. Besides which, instruction lengths are varied anyway. I have a feeling from some of the packet format stuff that Cell wouldn't have any majorly large immediate instructions and would instead reference main memory, eDRAM, or local cache for its data. So the instruction words would probably be pretty darn small.

[iamlucid]

Here is another link with some information about the cell from CNN Money.

http://money.cnn.com/2004/11/29/technology/cell.reut/index.htm?cnn=yes

I can't wait till May!!!

[cpiasminc]

Xenon... according to deadmeat: 32GFLOps/s. However, he gave PS3's CELL a mere 128GFLOp/s for his estimate, so don't pay deadmeat any attention. Other then that, there's no official info on the Xenon's performace
If anything, I'd think he's underestimating both consoles with a figure like that. Assuming that Xenon's CPU cores are more or less PowerPC-derivatives with all the usual OOOE and everything, each core's performance with scalar ops would generally be Celeron/Sempron level performance (maybe higher with the addition of 2-way SMT per core). It's just with Altivec-optimized code that it really gets fast.

The thing is that most likely the instruction set would include packed FMACs or FMADDs so it, too could get up to 8 FLOPs in one cycle. 32 GFLOPs divided by 3 cores divided again by 8 FLOPs per cycle means only a clock speed of 1.33 GHz. Even allowing for the possibility that there are no packed FMADDs/FMACs and so you can only get 4 FLOPs/cycle max, that's still only a 2.66 GHz clock. Either sounds kind of low. The block diagram suggested "3.5+" on 3 cores, which would yield at least 42/84 GFLOPs.

BTW, Deadmeat argued 64 GFLOPs for CELL, IIRC -- 1 GHz, 1 PE. Which again, is underestimating. Especially when all your pipelines are in-order.

[megadrive]

I've lost interest in my house and won't be back up til later this week, hopefully. im posting from library..

25 GFLOPs is just completely rediculas, obviously.

the GSCube 16 (16 EEs) was rated at over 90 GFLOPs and I think PS3 will surpass that significantly.

128-256 GFLOPs seems to be the minimum for the PS3 CPU system alone, not counting the GPU.

I think PS3 could well have alot less performance than even the prototype Cell workstation, though, since the workstation will have the advantage of having many seperate processors. PS3 can only have a few seperate dies.

a Cell workstation should get 1-2 TFLOPs and then eventually 16 TFLOPs.

the PS3 should have a few hundred GFLOPs at least which is alot more than 25 GFLOPs.

that's about all i can say, going by what i've read today.

[Matt]

Anything Cell-related goes here:

http://www.psinext.com/forums/viewtopic.php?t=2872