FS ATI interview on R500 + Block Diagram
- Thread starter Acert93
- Start date
mckmas8808
Legend
And Tim to me isn't that the point. When devs make games aren't they going to use it Cell to help with some calculations while the RSX renders somethings? To me they are saying in that slide is "Our PS3 will be able to produce 100 billion shader ops/sec." And isn't that what really matters? I mean thats what the game will use right?
________
MODEL C TEN
________
MODEL C TEN
Last edited by a moderator:
mckmas8808 said:
The Xbox 360 patents mention offloading some of the vertex processing load to the xCPUs. I do not believe that is unheard of. Actually with the R500 unified shaders that would we advantageous for games that require a lot of pixel shading power and minimal vertex shading because there would be less wasted horse power (of course PS3 may have so much you can just waste some and still get a superior result, we do not know yet).
The question I have is about the Spiderman type demo that was done on CELL. Specifically the flexiblity of CELL interactivity in the rendering pipeline. With WGF 2.0 we are looking at GPUs passing vertex and pixel shader data back and forth which would open up a lot of flexibility. Is CELL able to do this with RSX (i.e. latency between the two, and is RSX designed to pass/receive information like that), or is the CELLs involvement mostly pre- and post- rendering effects like vertex and lighting manipulation before and postrendering frame buffer rendering.
The pre- and post- manipulation is not necessarily new (e.g. Doom 3's shadows are processor bound), what I want to know is the INSANE level of detail seen demos like the the Spiderman demo if CELL realistically can contribute to such processes in realtime WITH the GPU.
Has anyone read the 360 GPU article on Extremetech.com: http://www.extremetech.com/article2/0,1558,1818139,00.asp. They claim the GPU can do 120 billion shader ops per second. Could they have gotten that figure by counting the logic units in the eDram?
XavierS said:
The war of shader ops is never going to end!
Next Sony/NV will say, "Opps we forgot to count the opps from bluetooth interface, that makes it 130"
Next MS/ATI will mention "We forgot the shader ops the video chip offers, that makes it 140"
Then Sony will say, "We missed the shader ops from the wifi chip, it is now 150"
Then MS will say, "we miss the shader opps from our power cord!..."
http://www.extremetech.com/article2/0,1558,1818139,00.asp
The 48 ALUs are divided into three SIMD groups of 16. When it reaches the final shader pipe, each of the 16 ALUs has the ability to write out two samples to the 10MB of EDRAM. Thus, the chip is capable of writing out a maximum of 32 samples per clock. At 500MHz, that means a peak fill rate of 16 gigasamples. Each of the ALUs can perform 5 floating-point shader operations. Thus, the peak computational power of the shader units is 240 floating-point shader ops per cycle, or 120 billion shader ops per second at 500MHz.
???
The 48 ALUs are divided into three SIMD groups of 16. When it reaches the final shader pipe, each of the 16 ALUs has the ability to write out two samples to the 10MB of EDRAM. Thus, the chip is capable of writing out a maximum of 32 samples per clock. At 500MHz, that means a peak fill rate of 16 gigasamples. Each of the ALUs can perform 5 floating-point shader operations. Thus, the peak computational power of the shader units is 240 floating-point shader ops per cycle, or 120 billion shader ops per second at 500MHz.
???
Jaws and Jawed, could you guys give some feedback on the relevance on Shader Op/s here?
As for power draw: http://www.teenja.com/p/articles/mi_zdext/is_200405/ai_ziff125906
As for power draw: http://www.teenja.com/p/articles/mi_zdext/is_200405/ai_ziff125906
Acert93 said:
I've posted my interpretation earlier in this thread,
http://www.beyond3d.com/forum/viewtopic.php?p=526435#526435
1 shader op per cycle ~ 1 shader execution unit
1 shader execution unit ~ vector unit or scalar unit
e.g. ALU = 1 scalar unit + 4-way SIMD unit ~ 2 shader ops per cycle
It's simply a 'count' of the number of execution units, more specifically, shader execution units, AFAIK...
mckmas8808 said:
The Cell is of cause more suitable for vertex processing than the xCPU, but it would also be possible to offload vertex operations to the xCPU. That makes it is unfair compare the RSX + Cell against the xGPU alone. And most of the time the CPUs will be used to do other work anyway and vertex shading will be left to the GPU.
The RSX is around 50+% faster than the xGPU in peak theoretical performance pretty in pretty much every way. How much of this advantage it can retain in real life is a hypothetical question until we see some real hardware. I think it will lose most if not all of this advantage but the picture is far from complete and I would not be very surprised if I was wrong.
Jawed said:
35W is damn cool the top range desktop GPUs are using 70-90W, I have my doubt about 35W it seems low but the eDRAM helps power consumption as it reduces IO-power usage quite a bit.
If it really only is 35W I don't understand why they did not try to clock it more aggressive (500MHz is quite low for modern 90nm GPU).
Edit: forgot that the 35W does not include the GDDR3 memory and the 70-90W does - still 35W is quite low.
Jaws said:
I hate to nitpick, but I'm awfully confused about the 196 flops per cycle number. If there are 48 Shader units and each can do 4 floating point ops, doesn't that equal 192 flops/cycle?
I find it especially confusing because in the interview, he addresses both numbers (192 [processors] and the 196) but IMO the 196 flops/cycle is incorrect?
