FS ATI interview on R500 + Block Diagram

[Nite_Hawk]

One does wonder why exactly the PS3 GPU has around twice as many transistors as the xbox360 GPU. From this perspective, it will be interesting to see where the ATI GPU is deficient in comparison. It certainly sounds like ATI is making better use of what transistors they have.

Nite_Hawk

 

[nAo]

R500 impress me more day by day..ATI did a stunning work

 

[therealskywolf]

Anand even noted that the Gpu could do AA for free at 1080p.....

No 1080p, only 1080i.

The Vpu can do it. Its the output that MS has in the Box that doesnt support it, but MS (Ballmer i think) as said that right now it is as it is......

I dount doubt that if they feel the need to, they will make the output support 1080p because the VPU can do it, as stated by the Ati guy.

 

[therealskywolf]

One does wonder why exactly the PS3 GPU has around twice as many transistors as the xbox360 GPU. From this perspective, it will be interesting to see where the ATI GPU is deficient in comparison. It certainly sounds like ATI is making better use of what transistors they have.

Nite_Hawk

Because it wasn made from scrath for the Ps3, they were doing it fo the PC and then addapted it to work with Cell. Thats why there's so many transistors..... (I guess)

 

[Geeforcer]

I think that R500 will be at least as good as RSX in 720p in qualitity per pixel, but will go dawn fast as rez go up.

BTW it seems to have some nice features for "Fluid Reallity"

From the looks of it, the resolution will not go up - the system seems to be restricted to 720p.

no, the system standard resolution mandated my MS is 720p, it can go all the way up to 1080p if needed

It could, but my understanding that it can't output above 720p per MS's decision.

 

[jvd]

in an interview they said 1080i

 

[Geeforcer]

Yes, you are right. Sorry about that.

 

[PiNkY]

quote Dave Baumann

... this will have ramifications beyond XBox 360 and even the PC.

one would be nintendo

 

[therealskywolf]

quote Dave Baumann

... this will have ramifications beyond XBox 360 and even the PC.

one would be nintendo

MS owns the Rights to it. The patents or whatever, i doubt MS is gonna let some of the tech that is only on their Chip to go to the Revolution's chip.

Just like SOny doesnt let IBm use Cell or whatever in the Xbox 360.

Thats why both chips are being made by two different Teams that have no contact with one another.

 

[Joe DeFuria]

quote Dave Baumann

... this will have ramifications beyond XBox 360 and even the PC.

one would be nintendo

I'm thinking this tech is going to make its way into handhelds in one form or another...

 

[vblh]

Could this mean that they used Fast14 tech as well? I know that you guys said no previously but the low transistor count just got me thinking that they used Fast14 to increase eff.

 

[Riddlewire]

As far as the Xenos transistor number is concerned, that's only counting the 'parent' core, right? The 150M doesn't include the EDRAM and its associated logic. So the total transistor count is considerably higher than 150M. Technically, the Xenos graphics system isn't half the RSX in transistor count. It's only slighter less.
Or have I misunderstood completely?

 

[eSa]

quote Dave Baumann

... this will have ramifications beyond XBox 360 and even the PC.

one would be nintendo

And other would be the whole "GPGPU" scene... If unified shaders/threading really works well, performance could be stunnig when runnig general algorithms.

 

[quest55720]

As far as the Xenos transistor number is concerned, that's only counting the 'parent' core, right? The 150M doesn't include the EDRAM and its associated logic. So the total transistor count is considerably higher than 150M. Technically, the Xenos graphics system isn't half the RSX in transistor count. It's only slighter less.
Or have I misunderstood completely?

I think IGN was the only place to ever say anything about tranny count on the xenos. The 150MT could be just for the e-dram for all we know. Until we hear more from either ATI or MS we can not even really guess the transistor count of the xenos. With 192 PS/VS3.0++ ALUs and 10Meg of ram I don't think the xenos is hurting in the transistor department.

Also from the info today is the GPU have 2x the shading power we thought?

 

[j^aws]

As far as the Xenos transistor number is concerned, that's only counting the 'parent' core, right? The 150M doesn't include the EDRAM and its associated logic. So the total transistor count is considerably higher than 150M. Technically, the Xenos graphics system isn't half the RSX in transistor count. It's only slighter less.
Or have I misunderstood completely?

I think IGN was the only place to ever say anything about tranny count on the xenos. The 150MT could be just for the e-dram for all we know. Until we hear more from either ATI or MS we can not even really guess the transistor count of the xenos. With 192 PS/VS3.0++ ALUs and 10Meg of ram I don't think the xenos is hurting in the transistor department.

It's 48 Unified shader ALUs (32 bit, 4-way SIMD + scalar)

Those 192 are fixed function ALUs for blending/stencil/z etc...

 

[DopeyFish]

As far as the Xenos transistor number is concerned, that's only counting the 'parent' core, right? The 150M doesn't include the EDRAM and its associated logic. So the total transistor count is considerably higher than 150M. Technically, the Xenos graphics system isn't half the RSX in transistor count. It's only slighter less.
Or have I misunderstood completely?

I think IGN was the only place to ever say anything about tranny count on the xenos. The 150MT could be just for the e-dram for all we know. Until we hear more from either ATI or MS we can not even really guess the transistor count of the xenos. With 192 PS/VS3.0++ ALUs and 10Meg of ram I don't think the xenos is hurting in the transistor department.

It's 48 Unified shader ALUs (32 bit, 4-way SIMD + scalar)

Those 192 are fixed function ALUs for blending/stencil/z etc...

the whole thing though is the edram has it's own logic that does the blending/z/stencil

we're talking about the block diagram which describes the shader pipeline as 192 ALUs (3 shader pipes with 16 shaders... 4 ALUs each)

 

[quest55720]

As far as the Xenos transistor number is concerned, that's only counting the 'parent' core, right? The 150M doesn't include the EDRAM and its associated logic. So the total transistor count is considerably higher than 150M. Technically, the Xenos graphics system isn't half the RSX in transistor count. It's only slighter less.
Or have I misunderstood completely?

I think IGN was the only place to ever say anything about tranny count on the xenos. The 150MT could be just for the e-dram for all we know. Until we hear more from either ATI or MS we can not even really guess the transistor count of the xenos. With 192 PS/VS3.0++ ALUs and 10Meg of ram I don't think the xenos is hurting in the transistor department.

It's 48 Unified shader ALUs (32 bit, 4-way SIMD + scalar)

Those 192 are fixed function ALUs for blending/stencil/z etc...

This is from the tech report article.

http://www.techreport.com/etc/2005q2/xbox360-gpu/index.x?pg=1


On chip, the shaders are organized in three SIMD engines with 16 processors per unit, for a total of 48 shaders. Each of these shaders is comprised of four ALUs that can execute a single operation per cycle, so that each shader unit can execute four floating-point ops per cycle.

This is were I got the 4 ALUs per shader for 192 total. I bet I am mixing up some terminalogy being a layman.

 

[j^aws]

FYI, I posted some patents here regarding R500+ EDRAM module,

http://www.beyond3d.com/forum/viewtopic.php?t=22260


Anyway, these tech reports all seem conflicting...

On chip, the shaders are organized in three SIMD engines with 16 processors per unit, for a total of 48 shaders. Each of these shaders is comprised of four ALUs that can execute a single operation per cycle, so that each shader unit can execute four floating-point ops per cycle.

AFAICS, they're still describing 48, 4-way SIMD units...or 48 ALUs which were in MS spec sheet....

In fact these provide 4 flops per cycle per SIMD unit, which seem *less* powerful than the 'leak'... :?

 

[j^aws]

...
The Xenon GPU is a custom 500+ MHz graphics processor from ATI. The shader core has 48 Arithmetic Logic Units (ALUs) that can execute 64 simultaneous threads on groups of 64 vertices or pixels. ALUs are automatically and dynamically assigned to either pixel or vertex processing depending on load. The ALUs can each perform one vector and one scalar operation per clock cycle, for a total of 96 shader operations per clock cycle.Texture loads can be done in parallel to ALU operations. At peak performance, the GPU can issue 48 billion shader operations per second.
...

ALU = vec+scalar ~ 2 Shader ops

48 ALUs ~ 2 shader ops per ALU * 48 ALUs ~ 96 shader ops per cycle

AFAIK, these agree with the official spec....

There just seems a liberal use of the term ALUs here...

If the article is correct, then each of the 48 shader unit/ALUs ~ 3-way SIMD + scalar ... (4 flops per cycle)

 

[ecliptic]

Uhm I thought vector is 8 and scalar is 2.