PlayStation 3 GPU: NV40 and NV50 Hybrid with XDR DRAM...
- Thread starter pc999
- Start date
xbitlabs's translation is erroneous as you guess 
The original article (not "report" as xbitlabs suggested, but more of a column with educated speculations + some industry sources) by Hiroshige Goto says nothing like it's hybrid of NV4X and NV5x, but I can guess where xbitlabs had it mistaken.
I'd like to translate the first section of the article as acculately as possible:
So, basically he drags out his speculation from only 2 facts - the JHH comment and nVIDIA devoting some serious resources into the PS3. He says nothing like it "will use NVIDIA’s technologies found in the current NV40 generation of its own chips as well as numerous techniques developed for the next-generation part known under NV50 code-name." which xbitlabs suggested. According to his speculation, the PS3 media processor has very different architecture from the PC GPUs (NV4X and NV5X) and not based on the PC GPUs either. Only its development timeframe comes in between them as NV5X was delayed.
The continuing 3 sections of this article discuss the 3 things:
Section 2. Which part deals with geometry processing?
Goto speculates the media processor has geometry pipes too like the current nVIDIA architecture (Vertex Shader & Pixel Shader, not unified). It's better in performance per die area than a Cell-based GPU. It's logical to make 3D pipes in one-chip as in-chip processing capability can be pushed further than bus bandwidth, though in the case of the PS3 Redwood can mitigate this issue.
Section 3. SCEI heads for augmenting generic computing power
In the GDC 2002 the SCEI CTO explained that the developers before the PS2 wanted realtime graphics but after the PS2 they wanted performance for simulation. It means in the next generation the target is in higher CPU performance unlike the PS2 in which 75% of silicon was devoted to graphics processing. Programmable Shader in the media processor adds to the power of CPU in generic computing, too.
Section 4. Benefits for nVIDIA
SCEI has been taking the most aggressive architecture among the 3 console vendors and it means nVIDIA can gain the highest technical experience by the work with SCEI and can do some adventure not found in PC graphics development. nVIDIA can learn the implemetation of the Redwood and XDR-DRAM inrterface without risk which they have when they implemented it in PC GPU. Moreover, like the Xbox development brought nForce to nVIDIA, the collaboration with Sony will bring something back to nVIDIA.
The original article (not "report" as xbitlabs suggested, but more of a column with educated speculations + some industry sources) by Hiroshige Goto says nothing like it's hybrid of NV4X and NV5x, but I can guess where xbitlabs had it mistaken.
I'd like to translate the first section of the article as acculately as possible:
So, basically he drags out his speculation from only 2 facts - the JHH comment and nVIDIA devoting some serious resources into the PS3. He says nothing like it "will use NVIDIA’s technologies found in the current NV40 generation of its own chips as well as numerous techniques developed for the next-generation part known under NV50 code-name." which xbitlabs suggested. According to his speculation, the PS3 media processor has very different architecture from the PC GPUs (NV4X and NV5X) and not based on the PC GPUs either. Only its development timeframe comes in between them as NV5X was delayed.
The continuing 3 sections of this article discuss the 3 things:
Section 2. Which part deals with geometry processing?
Goto speculates the media processor has geometry pipes too like the current nVIDIA architecture (Vertex Shader & Pixel Shader, not unified). It's better in performance per die area than a Cell-based GPU. It's logical to make 3D pipes in one-chip as in-chip processing capability can be pushed further than bus bandwidth, though in the case of the PS3 Redwood can mitigate this issue.
Section 3. SCEI heads for augmenting generic computing power
In the GDC 2002 the SCEI CTO explained that the developers before the PS2 wanted realtime graphics but after the PS2 they wanted performance for simulation. It means in the next generation the target is in higher CPU performance unlike the PS2 in which 75% of silicon was devoted to graphics processing. Programmable Shader in the media processor adds to the power of CPU in generic computing, too.
Section 4. Benefits for nVIDIA
SCEI has been taking the most aggressive architecture among the 3 console vendors and it means nVIDIA can gain the highest technical experience by the work with SCEI and can do some adventure not found in PC graphics development. nVIDIA can learn the implemetation of the Redwood and XDR-DRAM inrterface without risk which they have when they implemented it in PC GPU. Moreover, like the Xbox development brought nForce to nVIDIA, the collaboration with Sony will bring something back to nVIDIA.
DaveBaumann said:
If Sony/Toshiba engineers try to implement some fixed/3D-dedicated features in silicon and then it turns out nVIDIA can do it with smaller die area, I think SCEI choose nVIDIA and still dictates the route. Why does Sony have to license needless parts of the GPU?
kyetech said:
Yep, that was also my idea of the NV2A. Just a NV20 with an additional vertex shader for improved polygon performance. That's it.
That's all fine and well, but when you already have a CPU part with ~250-300GFlops, tacking on extra vertex hardware on the GPU makes little sense, regardless of the area efficiency.One said:
Particularly when you consider cumulative performance of the competing hardware is in the range of 300GFlops.
Without sounding like a broken record, I'll re-iterate this post from a SCE employee under NDA,
http://www.beyond3d.com/forum/viewtopic.php?p=427308#427308
http://www.beyond3d.com/forum/viewtopic.php?p=427308#427308
AndrewM said:london-boy said:
It had to be you, didnt it
It would be reasonable if people actually speculated on something plausible, not fanciful.
Just saying...
kyetech said:
According to a few of the dev's around here NV2A has double pumped Z and stencil, a feature that was introduced on the PC with NV30, not NV25. There is also this age old Allard quote (originally sourced from a Gamespot interview that appears not to be online anymore) that said "The second thing is the part that we have in the box is not just a super clocked NV20 but it actually draws some of the geometry features out of the NV30.", never heard what was drawn out of NV30 though (and IMO that probably actually relates to the Z and Stencil features more than geometry capabilities).
one said:
Given the composition of PS2 I think Sony have "fixed function" 3D hardware pretty well understood! This is the easy end of the pipeline and not really where NVIDIA are spending the majority of their R&D (although there will still be improvements to come).
Jaws said:
DaveBaumann said:
That's true and would satisfy those comments but why omit the major part,
"totally different architecture that is not based on an existing architecture from Nvidia."
...of that quote!
...and bearing in mind the 2006 release timeframe (likely released after the R500 rumoured Xe GPU), why would it be based on NV40 architecture and not the NV50 architecture? Which would satisfy those omissions!
Jaws said:That's true and would satisfy those comments but why omit the major part,
"totally different architecture that is not based on an existing architecture from Nvidia."
...of that quote!
Because I thought that was already a given seeing JHH has already come out and flat out said that its based on their next generation architecture, which is not a "currently existing" architecture.
...and bearing in mind the 2006 release timeframe (likely released after the R500 rumoured Xe GPU), why would it be based on NV40 architecture and not the NV50 architecture? Which would satisfy those omissions!
Did I suggest that it was NV40 based?
london-boy said:
Duh.
Still, have you stopped to think exactly how much 20Gpix/s is?1080P res is just over 2Mpix. That's nearly 20.000 screen fills per second, or more than 160 fills per frame at 60 frames/sec. It's still 40 full-frame fills/frame with 4x supersampling. Can you realistically think of anything you want to do that needs those kind of fillrate figures?
And remember, 20Gpix is counting LOW, as NV40 clocks faster than 400MHz in some variants.3xNV40 performance is completely unrealistic. It's even more silly than expecting 1Tflop from PS3's CPU, because it's actually doable to make a chip with that performance level, as prototype cells clock at 4+ GHz already and there are viable uses for that level of performance in games for a variety of purposes, while you have no use whatsoever for pixel fills in the range of 20 thousand thousand thousand pixels/s on output devices that can realistically only display a fraction of that.
It's simply a prepostrous idea! Period.
one said:
About nVidea CTO of curse he say that there way is better.
On the other side GPUs can already do a lot of work
http://www.eet.com/in_focus/silicon...html?articleID=55300904&_requestid=802507
Similar threads
- Deleted member 13524
- Architecture and Products
- Deleted member 13524
- Mobile Devices and SoCs
