Behold, CELL specs!!!! (I think it is..)

[Johnny Awesome]

1x8 is a pretty safe bet and still VERY powerful.

 

[Megadrive1988]

only if each of the eight APUs has 16 FPUs and 16 integer units

 

[Megadrive1988]

remember all the old reports of PS3 having 4 to 16 Cells ? now its only gonna have 1 Cell? wtf that would be awful imho. 256 Gflops peak. probably less than 100 Gflops sustained. plus the Nvidia GPU. and 128-256 MB of external RAM. say 8-16 MB of CPU eDRAM and maybe no GPU eDRAM. worst case. i would be disapointed as hell. that would mean less of a leap over PS2, than PS2 was over PS1. probably. powerful yes. reaching Sony's goals, no. that would mean we'd have to wait for PS4, just to get to what PS3 was going to be, plus maybe a little more. why so pessimistic?

I'm hoping for 2-4 Processing Elements. 16-32 APUs.


don't forget the preferred embodiment in the patent showed 48 APUs.


bah. it all comes down to the GPU in the end anyway, for us graphics whores 8)

 

[jvd]

you'd be unhappy with say 50-75gflops sustained in a console ?

Thats what i'm expecting .

 

[Almasy]

you'd be unhappy with say 50-75gflops sustained in a console ?

Thats what i'm expecting .

Sony wanted 1 TFlops, if they only had 50-75 gflops of sustained performance, perhaps very similar to Xenon´s performance (which would make you a very happy dude, I suppose), then I´d also be VERY dissapointed as well.

So much money invested to only come up with one PU that performs as well as a CPU that costed a LOT less to come up with? Who wouldn´t be dissapinted in that?

 

[Mythos]

CPU (Cell)/GPU(NV) = 1 Tflop.

 

[Wunderchu]

CPU (Cell)/GPU(NV) = 1 Tflop.

I agree .... theoretical performance of CELL + NVIDIA GPU has high likelihood of being at 1 TFLOPS or higher, IMO

 

[psurge]

To be honest, given a memory bandwidth of 50 GB/s, I'm not sure it makes any sense to have more than 1 PU. Assuming that 50 GB/s is dedicated to exclusive use of a single Cell with ~256 GFlops peak, you have enough bandwidth to load/store a single 32 bit float from/to external memory once every 20 instructions...

The Xenon CPU, even if only capable of ~100 GFlops, would have the same problem (possibly worse due to cache thrashing effects). Maybe this is why the patent mentions procedural geometry creation - this is an operation taking a small amount of input data, performing lots of computation, and producing large (relative to input) amounts of output which is transferred directly from L2 to GPU.

 

[randycat99]

So what's the word on clockrate now? I saw Tomshardware post a blurb on PS3 specs, in reference to an Inquirer article. Does 4.6 Ghz sound possible?

 

[Acert93]

Sony fans are really back and forth on the 1 TFLOP deal :| I thought it was just a marketing statement early on and never repeated since? Unless Sony has stated their goal is 1 TFLOP for the PS3 CELL then I wont even expect that. No point getting dissappointed in something never promised!

The CELL looks to be crazy powerful. 256 GFLOPs is a LOT (as is ~100 GLOPs sustainable). In comparison someone in B3D said a top of the line AMD/Intel chip did ~30. I could not find hard numbers, but did find some sites that measured the FP performance in the server market, and it looks like Xenon's and Opterons are in the 4-10 GFLOPs range.

So if a single CELL in PS3 hits ~100GFLOPs you are looking at a huge performance boost over the PCs (which is the platform for high end gaming at the moment).

To put this in perspective: The Xbox has a modified PIII Coppermine @ 733MHz. Current top end PCs are about 5x as powerful as that, and the CELL looks to be what, 15x?

One thing that was stated in the IGN mailbag (which was pretty silly in some areas, like the GF6 comment... hello, they just had an interview that made it clear it is the next gen GF chip still in development... oh well, they usually do a good job) that is true: Because of cross platforming, a lot of the differences in the systems' perfromances (if they substantially exist) will be nullified by the fact that cross platform development and porting is so important to the industry right now. The Nvidia deal guarantees that their games will be their graphic system will produce just as nice visuals (assuming Sony is not on the cheap) if not better and that the graphics system will be quite a bit more excessible.

So CELL, which is magnitutes more powerful than current top end PCs, and a next gen GF7 chip is pretty powerful. Memory may be a concern, but to be "dissappointed" to me would be silly! Dissappointed because the P3 CPU outpaces everything in the market and has a top of the line GPU AND BR?

And this is from someone who is more excited personally about the Xenon.

Also, one aspect of CELL is important NOT to miss: It looks like it will be in everything, and really sets Sony up well for the PS4. I fully expect PS4 to have CELL, and LOTS of them. Imagine Agent Smith in the Matrix. So PS4 should be PS3 compatible and will be insanely powerful AND developers will already be familiar with the basic concepts. The CELL really looks like a long term plan with some great immediate benefits and even MORE longterm benefits.

So there are still a lot of reasons to be excited about PS3

 

[pahcman]

Boring link that over praise CELL. What said have not be said before? May be we know the writer?

The good thing about topic is the ever elusive MrSingh has again blurb out more numbers!
1:8 is good. 256Gflops + NV5-PS is powerful. What are flops rating of rumored Xenon CPU, 80? RX500 may be powerful, but to make up the shortfall? What will be multi core Pentiums A64s flops in 2005/6, what are they now? Of course flops ratings as our experts have said, are ego numbers.
Given small time window, costs and same gen GPUs, not much to expect between PS3, Xenon and Rev.

Dont read too much in self intepretated numberss and you be all rights. Sony may feel the heat and have 2-3-4 PEs, in end more powerful dont mean more success.

 

[DudeMiester]

One thing I think is particulary interesting is the possibliy of add-on unit/cards that operate using the Cell interface. These would directly and transparently intergrate into the computing environment. You could have hardware that contained a group of specialized cells for certain types of processing that just drop right in and can be used. It doesn't have to be just a video or sound card, and it doesn't need fancy drivers, it just works! For example, games could be bundled with a set of specalised Cell processors it uses, packaged not unlike a memory card, you just stick it in and go. If you want to use some of these hardware algorisms yourself, no problem, just use it. It would make hardware very intercompatible and very interesting.

 

[jvd]

you'd be unhappy with say 50-75gflops sustained in a console ?

Thats what i'm expecting .

Sony wanted 1 TFlops, if they only had 50-75 gflops of sustained performance, perhaps very similar to Xenon´s performance (which would make you a very happy dude, I suppose), then I´d also be VERY dissapointed as well.

So much money invested to only come up with one PU that performs as well as a CPU that costed a LOT less to come up with? Who wouldn´t be dissapinted in that?

Sony spent alot not only on the cpu design but also on the fabs .


This is a 300$ console even if they budget it at 500$ they still have money constraints and if they put int o large of a chip that runs to hot and draws to much power they will have alot of probems .


THe console has to a cost to power ratio and I dont' believe the 1tflop is going to be it . That is more of sony hype .



Yes i'd be happy with 75gflops sustained. That is more than i think the xenon chips are going to supply

 

[BriefcasemanX]

I'm going to laugh it if ends up being only 256Gflops when a year or so ago we had people making claims as high as 2tflops as feasible for ps3.

256gflops is still awesome.

 

[one]

So what's the word on clockrate now? I saw Tomshardware post a blurb on PS3 specs, in reference to an Inquirer article. Does 4.6 Ghz sound possible?

Possible?

http://www.siliconvalley.com/mld/siliconvalley/10323259.htm

In a disclosure to the International Solid State Circuits Conference, IBM said it has made Cell chips that run at 4.6 gigahertz and operate at 1.3 volts.

That speed is notable in part because Intel canceled its 4-gigahertz Pentium 4 chip this year. IBM earlier in the week would say only that a rack of Cell chips could compute 16 teraflops, or trillions of precision math operations, a second. The world's fastest supercomputer currently deployed in the commercial market is NEC's Earth Simulator, which computes 36 teraflops.

``They will clearly challenge Intel in the living room in several ways,'' said Jon Peddie, an analyst at Jon Peddie Research in Tiburon. ``They have so much performance, but it remains to be seen if they can hit low costs for consumer markets.''

The Cell chip will be able to communicate with other chips at a speed of 6.4 gigabits per second. Jim Kahle, IBM fellow and director for Cell in Austin, said the International Solid State Circuits Conference requires that companies have working chips when they submit papers for the event, which runs Feb. 6 to 10 in San Francisco. Those working sample chips aren't necessarily going to be the same as the final version, Kahle noted.

... and it's in the 90nm process. If they can put 4 PE in the 65nm process, they may clock it down (4Ghz) to match it with the preferred embodiment and reduce heat, though.

It was 6.2Gflops (theoretical, before someone points it out) in PS2 in 2000, and when you apply Moore's law it should be 64 times increased in 2006, therefore the number < 400Gflops for PS3 is unacceptable!

1 PE is very unlikely as in 2004 an SCE developer told multi-core would be a must-have in the next-gen. It can hide memory-access latency since one core can work while another core is waiting for memory access.

 

[Inane_Dork]

1 PE is very unlikely as in 2004 an SCE developer told multi-core would be a must-have in the next-gen. It can hide memory-access latency since one core can work while another core is waiting for memory access.

Um... does that really speed anything up? Now instead of 1 memory starved processor you have 2. One CPU is already capable of handling the load. It's the memory that's the bottleneck.

If anything, stuff like HyperThreading would help you overcome memory latency. Multicore would only possibly help if you multiply your cache size along with the processor count.

 

[zidane1strife]

So CELL, which is magnitutes more powerful than current top end PCs, and a next gen GF7 chip is pretty powerful. Memory may be a concern, but to be "dissappointed" to me would be silly! Dissappointed because the P3 CPU outpaces everything in the market and has a top of the line GPU AND BR?

It's just that a ps3 with a BE will have more tech-parity longevity than one with a single PU, and this system is being fabbed with a cutting edge process unlike the last sony console(250nm) and its launching relatively later than its predecessor did and more resources have been spent. It's thus unexpected for it to provide a smaller jump than the previous h/w(done with a non-cutting edge process and with less R&D, and shorter preceding console cycle), only fab. problems or the like could cause something like that, IMHO.

 

[one]

1 PE is very unlikely as in 2004 an SCE developer told multi-core would be a must-have in the next-gen. It can hide memory-access latency since one core can work while another core is waiting for memory access.

Um... does that really speed anything up? Now instead of 1 memory starved processor you have 2. One CPU is already capable of handling the load. It's the memory that's the bottleneck.

If anything, stuff like HyperThreading would help you overcome memory latency. Multicore would only possibly help if you multiply your cache size along with the processor count.

Yeah, SMT is what helps to hide latency more easily But I'd like to note that the same SCE developer remarked a large cache is useful to hide latency, too, though how it can relate to the Cell in the PS3 is unclear.

 

[Megadrive1988]

of course, with a single Processor Element (1 PU + 8 S|APUs) PS3, there would almost be no need for 65nm. the 90nm process would be more than good enough. at least that is what I would tend to think. so that makes me glad that the 65nm process is the target process. to put more than one PE on a single die.

but it seems my dream of an 8-PE ~ 72 processor CPU
(8 PUs + 64 S|APUs) is most likely dead


or, another less likely but interesting strategy. release a single PE~256 Gflop PS3 by the end of 2005 to combat Xbox2 immediately. Sony continues work on Cell with IBM. working towards releasing a multi Cell-2/PE-2 multi-Tflop PS4 (say 12-36 Tflops) by the end of this decade on smaller than 45 nm process, to combat Xbox3, instead of waiting 5-7 years to release PS4 as would otherwise be, if PS3 came out in 2006 with 1 Tflop. it still leaves a 4 year console cycle which would probably still be ok for Sony. Sony has stated that they wouldn't want a 2-3 year cycle. but 4 years wouldnt be that bad. that would also help to keep console games more upto date compared to PC games. just a far out wild thought :|

 

[one]

of course, with a single Processor Element (1 PU + 8 S|APUs) PS3, there would almost be no need for 65nm. the 90nm process would be more than good enough. at least that is what I would tend to think. so that makes me glad that the 65nm process is the target process. to put more than one PE on a single die.

but it seems my dream of an 8-PE ~ 72 processor CPU
(8 PUs + 64 S|APUs) is most likely dead

Well, if they manage to put 64MB eDRAM on a Cell, 1 PE is not unlikely. But so far there's no sign of eDRAM on a CPU, while the GPU can have it.

Anyway, in 2007 the 45nm process will go online, then they may decide to start with an even crazier die size than the initial GS, and with a noisier fan