Support NeoGAF

[doncale]

that wasnt known before?

from what I've read there are going to be 2 tflop Cell workstations and 16 tflop cell workstations. where does this put Playstation3? 1 tflop or less than 1 tflop? I guess that much is not known.

regardless of the above question, what is now the known "Cell-structure" ? meaning what is in a Cell now? before it was 1 Processing Element that had a CPU core (PU) and upto 8 APUs. has this changed?


ok now, within the 1st generation and 2nd generation of Cell processors, what is the most that can be crammed into a single chip/die ?

It'll be interesting to find out how many chips are needed in Cell workstations to reach 2 tflops and 16 tflops.


it seems Playstation3 is going to out power Xenon by some amount, but how much is the question. I think it's going to be less than the power difference between DC and PS2. but i'm not sure.


excuse my slowness, ive been sick all week and not paying attention to gaming and technology very much.

<sips coffee and waits for Pana to reply>

 

[gofreak]

where does this put Playstation3? 1 tflop or less than 1 tflop? I guess that much is not known.

True, we still don't know this. I think it'll be likely less than 1TFlop, but I wouldn't be disappointed with that.

regardless of the above question, what is now the known "Cell-structure" ? meaning what is in a Cell now? before it was 1 Processing Element that had a CPU core (PU) and upto 8 APUs. has this changed?

No, it hasn't changed. Although the number of APUs was always variable, so it may not be 8 in the final design. 8 is the "preferred implementation" as per the original patents, but it could possibly be more or less. We know from Monday's news that the PEs have been clocked up to 4.6Ghz.

ok now, within the 1st generation and 2nd generation of Cell processors, what is the most that can be crammed into a single chip/die ?

Unknown, I think (?)

The significant info from monday was the revelation that they've broken 4.6Ghz, albeit with a fair bit of heat. The original patent's flop specifications (32GFlops per APU) appeared to depend on the PE being clocked at 4Ghz, so it's good news to see that they seem to be tracking that.

Also, we'll hear more from Feb 6 - 10 on the Cell architecture generally (4/5 papers on Cell are being presented at the ISSCC then), and according to Gamespot, PS3 specifics should be forthcoming in March (hopefully early March).

 

[vireland]

True, we still don't know this. I think it'll be likely less than 1TFlop, but I wouldn't be disappointed with that.

I wish I could remember where I saw it (EEtimes?), but some engineer was doing calculations based on what was known recently and came up with a sub-1tflop number, and not on the high side. If we got 1tflop out of it, I would be quite thrilled, but, as always, the software is really the thing that matters. Tflops is a pretty abstract measure of a machine.

 

[gofreak]

I wish I could remember where I saw it (EEtimes?), but some engineer was doing calculations based on what was known recently and came up with a sub-1tflop number, and not on the high side. If we got 1tflop out of it, I would be quite thrilled, but, as always, the software is really the thing that matters. Tflops is a pretty abstract measure of a machine.

That, I believe, was Paul Zimmons. His calculation was just a tad pessimistic. He was basing it on off-chip communication speed (i.e. speed between the PEs and main memory), meaning that for his figure to hold, the Processor would have to be loading instructions from main memory before execution. In reality this isn't the case - the processors have cache memory, and would only have to hit main memory if the instruction or data needed wasn't in cache. With a good memory management policy, this can be minimised.

If you take the raw, peak numbers, a 4Ghz, 8-APU PE yields 256GFlops.

 

[doncale]

thanks for answering my q(s) as best as you could.

this seems to be encouraging news overall.


if they're reaching 4.6 or 4.8 GHz now, with very much expected high amounts of heat, i would imagine a stable 4.0 or 4.0+ GHz clockspeed would be very possible on 65 nm chips in consumer devices (i.e. PS3) in early to mid 2006 with not overally difficult heat issues.

it also seems that at least 600+ gflops for PS3 CPU is almost a lock at this point, making it roughly 100x incease in fp performance over PS2 CPU.

of course i could be very wrong, the above is just my simple observations of the new info coming out.

 

[xexex]

the low figure was just 25 gflops iirc, which has to be incredibly off the mark. the Xbox2 CPU will beat 25 gflops from what ive read.

I think we can count on a figure in the hundreds of gflops.


edit: http://www.eetimes.com/story/showArticle.jhtml?articleId=54200580&pgno=2

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."

 

[McFly]

That engineer posted on B3D about that low number and he told us that it was only a minimum calculation of what's possible based on the known "specs" and he admited that it could be much more, but without more informations he can't say how much more (if more), but that maybe 1TF is not impossible.

It's somewhere in that big thread at B3D, somewhere around page four.

Fredi

Edit: gofreak was a bit faster then I ... just a bit ... a three posts bit.

 

[gofreak]

the low figure was just 25 gflops iirc, which has to be incredibly off the mark. the Xbox2 CPU will beat 25 gflops from what ive read.

I think we can count on a figure in the hundreds of gflops.


edit: http://www.eetimes.com/story/showArticle.jhtml?articleId=54200580&pgno=2

Given the context he admits he made the calculation in, it makes sense, but that's not a real-world context..

A lot of variables will determine how well the overall system keeps the processors firing.