oh so now Xenon has a TERAFLOP processor? (just CPU alone??)

Also the same guy worked on PipeRench, a co-processor for streaming multi-media acceleration. That was back in 1999. The link is a pdf describing the architechture.

http://www-2.cs.cmu.edu/~mihaib/research/isca99.pdf

 

It seems that you haven't heard that ONLY Sony has "bleeding edge research" and that's why they are giving this God's gift called...the cell just like they gave us THE EMOTION ENGINE!!No way will MS have a powerfull cpu in their console,the only powerfull console next gen. will be the ps3 cause of...the cell!!

 

I wonder what "Deadmeat" would say about this whole Teraflop hype now.
Actually I hope he'd come back and comment

 

http://eg04.inrialpes.fr/Programme/IndustrialSeminar/PPT/Trends_in_GPU_Evolution.pdf

 

now that is interesting. I've never heard abotu ash before.

 

If you look at some of DeanoC's previous (vague) comments about the chips that power Xenon, it sounds an awful lot like ASH, with its focus (in hardware and software) on making theoretical performance real-world performance.

 

I would say it was a slip of the tongue. Just like some people call their computers "CPUs". Even the PS3 teraflop goal/"rough target", I believe, was based on total system (back when a modified CELL was thought to be the GPU). I find it VERY ironic that MS is playing right into the hype WE (=geeks) made! How many people were expecting 1TFLOPs systems and told everyone they knew? Now in one swift move MS has delivered. Does not even matter if they are theoretical numbers, or for the entire system, all these people talking up 1TFLOPs chips--MS could not miss the opportunity to cash in on it. We are the victim of our own expectations.

And while I do not discredit that the machines may have a TFLOP of total system power (pretty hard to discredit) theoretically--i.e. who knows what the real world numbers are on EITHER system, does it really matter? TFLOPs are the Polygons/sec arguement from 2 gens past. More poly/s are not going to make better games. We are no longer at the point where we are comparing a 300poly model with x,y,z effect to a 500poly model with x effect.

But we are going to have to endure a load of marketing hype for the next 18mo or longer, and everyone is going to argue over it. All I want to know is WHO is making WHAT games for what system--and what features are on those systems that I will enjoy. Unlike some, I do not enjoy playing with TFLOPs 8)

Yet, back on the quote, the easiest explaination is to look at what was officially said at GDC, and that was that the system provides (most likely theoretically at that). The interview really looks like a slip of the tongue--unless they are counting everything as one processor since the GPU can write to the CPU cache.

Not that it matters. So far all we are hearing are theoretical numbers (which ironically tell us very little about the quality of the SOFTWARE we will have) and some of you are going to hate MS/Sony no matter what they say.

From this point on, EVERYTHING said from this point on is uber-Hype intended for the mass market. It will offend and alienate people on both sides on the fense, and those in the middle will laugh and say, "Uh, nice... so can we see the games yet?"

I just wish BF2 was a X2 launch title

 

I would say that most DeanoC "Jedi's comments" were about how general purpose seems to be the XeGPU, and how it was easy to use it in correlation of the general purpose XeCPU, in order to offload some, generaly considered as, CPU task to the GPU.

 

To,

Deadly Towers

This is taken from Nvidia's own PDFs from GDC. Specifically the PDFs pertaining to OGL and GPGPU solutions.

3Ghz Pentium 4 = 12Gflops (single precision) peak theoretical.

Geforce 5900 = 40Gflops (single precision) Observed performance.

Geforce 6800 = 53Gflops (single precision) Observed performance.

So deffinitely not a Tflops GPU by any stretch. (just a little FYI, The NV2A is not an 80Gflops part)

 

In the context of GPGPU, that makes sense, they're probably talking about programmable power. But the 1Tflop number NVidia apparently gave for the 6800 and the 1TFlop number MS are giving undoubtedly count ALL power.

 

Your numbers are the effective programmable FLOP count.
You're not considering the theoretical total FLOP count.

 

Do you really think that that is true? It appears that NO ONE not even developers have seen the real hardware outside of IBM and certainly dont know the performance pof any 'clean-sheet' design MS may have undertaken...

Just cause we know what SONY has done and are wowed by it doesnt mean A) that their success is not reproducible or that B) their is best solution for achieving that level of performance.

Being skeptical is fine but.... we really dont know what MS/IBM has created.

 

Is it me or are they incorporating programmble floating-point-operations-per-second AND hardwired fixed-point-operations-per-second into this infalted 1TFLOPS figure???

If it's the latter, then it's not even,

F-L--O-P-S

it's

F-L-O-P-S + F-P-O-P-S

and they might aswell include integer maths in there aswell as integer-operations-per-second,

I-O-P-S

So is the total a sum of FLOPS, FPOPS, IOPS and what size bits, 32bit, 16bit or 8bit?

They might aswell call it a total system operations-per-second so that when they say 1TeraFLOPS, they mean,

Total system operations per second ~ FLOPS+FPOPS+IOPS

and

1TeraFLOPS >>> 1TeraOPS

Bah!...It's all misleading BS!

 

Mabye some parts of the xenon gpu are quad pumped ? Like in the p4 ? So some things might be running two or four times faster ?

Mabye there is a physics cpu in there and we just don't know about it and that plus the main cpu can be stretched into the 1tflop zone ?

Mabye they are using fast 14 or whatever and hit 9 ghz

 

do people really believe that a 300-400 dollar gaming machine is going to have that much power I just cant see it

numbers people are talking about would put them in the supercomputer range

 

no and i've been saying it for a very long time. You wont find a single part inside one of these machines that are capable of 1tflop . I highy doubt even the whole system added up will hit 1tflop for any of the next gen systems unless one of these companys is willing to loose 500$ on each system sold

 

Maybe MS is capitolizing on the internet forum frenzy Sony diehards created with the idea of 1TFLOPs consoles by claiming a total theoretical system performance number no one will be able to test anyway, but it looks good on paper, *might* be justifyable, but most important is the first to stake the claim "1TFLOP console".

Yeah, for the next 18mo we are going to have to hear about Xenon with the press quip, "that according to MS's Allard exceeds 1TFLOPs of total targeted system performance".

Ms not only gets to play "Me too" but also gets to say "First there!" I see this as a small swipe at Sony's thunder. MS did not even release a number, so if Sony says "We can do 1.5 TFLOPs" the "over 1TFLOPs" does not look bad. This is as bad as politics. Spin, respin, everyone argueing why everyone else lies--and we STILL do not even have the facts. How many processors are in X2? We have no clue, but we sure like to argue who is right and wrong! Oh yeah, next gen is in full swing now!!

Anyone got popcorn?

 

CAP (Complexity Adaptive Processing) seems to be along the lines of ASH.

http://www.ccs.rochester.edu/projects/cap/


IBM has worked on IPCMOS for a good number of years now, so I guess it might be possible for a IPCMOS fabbed CPU. Back in 2000 on a .18 nm node they got a max clock speed of 4.5 GHZ although that wasn't a complete cpu design.

http://www.electronicsweekly.com/Article18998.htm

Pradip Bose comes across as being involved in many of the asynchronous research, shares at least one patent with Michael K. Gschwind who has been linked to both CELL and the Xenon project.

IBM patent

 

Addressing my previous post...

I understand entirely that that is programmable maximum. But I thought we were speaking in that context (we were on the Xenon with 1tflops CPU performance schtick... that would indicate programmable floating point performance, eh?). If not, if we're just speaking of equivalent performance of a Tflops CPU via the specialised nature of the GPU then yeah, I can see where a comparison would be derived.

Later

Sunscar

PS: Jaws, yeah, that's precisely it. Most of the FP numbers that get thrown around for marketing purposes anymore are indeed an agregate of atleast Fl-ops and Fp-ops. That's where the 80Gflops number comes from for the NV2A (or XGPU or whichever name it's called).

 

Xenon is not 1Tflop + of processing power. It is Microsofts newest marketing speak "MSFLOPS" which is an inclusion of fixed function processing.

Microsoft knows it will not beat Sony at the hardware game so they are tring to create: 1. That there system is ultrapowerful and 2. Confuse any real power advanatge between itself and PS3.