PS3 gpu 2x more powerful than X360 gpu?

Titanio said:
The 53bn dot products figure - is this for the CPU, or GPU, or combined? I believe's MS's 9bn figure was just for the CPU (?)

It makes some sense if they're counting bouth the CPU and GPU in the calculation and the GPU is an extension of the NV40 architecture.
 
nAo said:
What I'm trying to say to you is that 'that way' to calc shader ops has no factual meaning.
I agree. That's what I was saying.

I just thought I'd spin the numbers around for some entertainment, pointing out that NVidia's technique produces more impressive numbers. :devilish: It's fun to come up with some plausible numbers to take the shine off NVidia's wild claims.

If yours shader ops data are 'right' then explain to me why nv40 does 22 dot4 (16 ps + 6 vs) per cycle and g70 does 92 dot4 per cycle (according nvidia numbers), more than a four fold increase.
Hmm, the pixel shaders can do two Dot4's per clock, and a vertex shader should be doing Dot5s, shouldn't it?...

Anyway, my real interest in "shader op" specifications is to see if they can give any clues as to the architecture of the respective GPU. So far we're failing miserably :cry:

Moreover where did you get RSX 105.6 Gigaop/s? I haven't watched all the conference but IIRC nvidia quoted 136 shader ops per cycle * 550 Mhz = 74.8 Gigaop/s (whatever it means..)

I don't remember those figures, @36 minutes "100 billion Shader Ops / sec" appears on the slide, followed by "51 billion dot products per sec".

http://www.gamespot.com/live/stream...ath=2stream_sonypress_e305.asx&pid=927750

Jawed
 
At 25:50 the slide says "136 shader operations per cycle"... (Prolly includes vertex shader ops?)

I'll let someone else reconcile 136 shader ops per cycle with 105.6G ops/s. :)

Jawed
 
Jawed said:
Hmm, the pixel shaders can do two Dot4's per clock, and a vertex shader should be doing Dot5s, shouldn't it?...
about nv40: vertex shader -> one dot4 per clock cycle, pixel shader -> one dot4 per clock cycle (the second ALU can't do a dot4, just a vec4 mul, AFAIK, according GPU GEMS2 chapter 30..)
In fact I thought nvidia extended the second ALU in each pixel pipeline to perform a dot4 too, but also in this case numbers don't add up.
A 24 pixel pipes and 9 vertex shaders 'extended' NV40 would do 53 dot4 per clock cycle, there are needed other 40 (!!) dot4 products per cycle to reach RSX dot4 figure!
AS you can see even if it's just an improved nv40, it seems RSX, has undergone some non trivial changes.
I could be completely wrong though ;)
Maybe in their dot4 products per cycle number they're adding up TMUs work too :?

I don't remember those figures, @36 minutes "100 billion Shader Ops / sec" appears on the slide, followed by "51 billion dot products per sec".

http://www.gamespot.com/live/stream...ath=2stream_sonypress_e305.asx&pid=927750

Jawed
Yeah..but at minute 25th you can read:
Code:
Programmable shading processors: 136 shader operations per clock cycle
LOL :LOL:
 
rsx

12 vertexpipes -> 24ops
24 pixelpipes -> 96ops using nv40 design as reference
16 texture units -> 16ops
-------------------------------
136ops/cycle

or

8 vertexpipes -> 16ops
24 pixelpipes -> 96ops
24 texture units - 24ops
-----------------------------
136ops/cycle
 
Good idea tED!
But this would be the first time nvidia counts TMU ops too, they usually counts programmable operations.
 
Which would imply the major change in RSX/G70 is to separate the texture unit from the first shader ALU.

I suppose the second of those makes most sense.

The issue is that the vertex pipeline should also have texture units, though, shouldn't it?... Presumably there's a texture address to be calculated?...

Jawed
 
Well I predicted last year that the PS3 would have 2.4x more 'flops' than X360 on paper...so not far off if it's only 2x! :p

I actually think this paper 'flops' comparison is accurate. Why? It simple! 8)

Xbox = Microsoft + nVidia

They marketed the console with 'paper flops' to attain the most imaginitive way to substantiate flops.

X360 = Microsoft + ATI

---> They will try to do the sme IMHO....

PS3 = Sony + nVidia

---> They will try to do the sme IMHO....

So essentially, were comparing APPLES to APPLES! 8)


Total X360 ~ 1 TFLOPS

Total PS3 ~ 2 TFLOPS

Therefore PS3 twice as 'powerful' as X360! :p

QED

/runs
 
Joe DeFuria said:
Khronus said:
I thought MS said the ATI GPU had 4x MSAA not 2x? (my bold)

IIRC...

I believe it does support 4X AA, but the requirement is that all titles will support 2X. This tells me that 2X is basically "for free", and 4X will come with some performance hit.

So some titles may get 4X, but all will have 2X.
Where did you read about this 2x requirement?
 
3dcgi said:
Joe DeFuria said:
Khronus said:
I thought MS said the ATI GPU had 4x MSAA not 2x? (my bold)

IIRC...

I believe it does support 4X AA, but the requirement is that all titles will support 2X. This tells me that 2X is basically "for free", and 4X will come with some performance hit.

So some titles may get 4X, but all will have 2X.
Where did you read about this 2x requirement?

They specifically mentioned it during the introduction event. The exact words were "at least 2x AA".
 
One more question: anyone here thinks that R500 wouldnt surpasse very well a 6800, in 720p(+FPS), but would be harder (to surpasse) 1080i/p :?:

It isnt that the point, be the best at 720p (the ideal condition for XB360), but very IQ per pixel :?:

I mean having a image as good ( at least ) as RSX but with fewer pixels (lower rez) on the screen.

Because if they (RSX) are only puting 2X more pixels but the same Q, it only interest for the very few with 1080i/p tv. Unless they can make a better imagem at 720p( and no 1080 rez), but I dont think so.
I am expecting to have both the same IQ, but RSX @ higer-rez (that the reason for the almost 2 Tflops figure)
 
1080i and 720p aren't that far apart pixel wise. You must mean 1080p.

I think the jury is still out on who supports 720p better. The "eDRAM" has higher bandwidth, but its smaller, and HDR/AA modes might slow it down. Plus, there is shader rate vs fillrate. Many games are using shadow buffers and stencil shadows not, plus lots of render-to-texture, and that requires lots more fillrate.
 
Back
Top