More info about RSX from NVIDIA

[passby]

Bits here and there

- vertex shaders triangle setup speed up
- pixel shaders MADD speed up - devs dissect this:
http://www.hardspell.com/hard/showcont.asp?news_id=14372&pageid=3104

- Found the translations! TSAA and TMAA = Transparency antialiasing, multisampling or supersampling.
- suspect 16x FSAA in SLI mode.
- normal map compression that R420 first introduced.
- faster Z-buffer (whatever?)
- UltraShadow II. Supposed to be 2x performance of Nv40. Demo
http://www.hardspell.com/hard/showcont.asp?news_id=14372&pageid=3113

Lots of sceens in that article, so can actually browse without worrying too much.

 

[nAo]

nAo,
RSX ~ 56 vec4 units ~ 56 DOT4/cycle and not 52 DOT4/cycle then!

Yeah..30+ GDot/s

That chart should show vec3 + scalar or vec4 in the pixel shader, and not vec4 + scalar

I believe there is a new scalar unit per pixel pipe, too bad I can't read chinese

 

[N00b]

TSAA Transparent S(?) Anti Aliasing

sorting (?)

TSAA = Transparent Supersampling Antialiasing
TMAA = Transparent Multisampling Antialiasing

 

[Shifty Geezer]

Is this a new type of AA unique to G70 & RSX then? No details I presume.

 

[_phil_]

thanks NOOB.makes much more sense .

 

[dukmahsik]

so looking at this chart alone we can assume RSX is way more powerful than xenos?

 

[Dave Baumann]

Yeah, those Xenos numbers are wrong - they are from NVIDIA based on the early discussions. Obviously they had no deeper knowledge of the architecture beyond the 48 ALU's with Vec4 + Scalar.

 

[nAo]

Yeah, those Xenos numbers are wrong - they are from NVIDIA based on the early discussions. Obviously they had no deeper knowledge of the architecture beyond the 48 ALU's with Vec4 + Scalar.

Well, it's not Nvidia fault it ATI doesn't disclose all their numbers
Obviously Xenos ALUs should have some reciprocal unit too, but ATI never mentioned a thing about them.

 

[Fafalada]

Yeah, those Xenos numbers are wrong

Then why do they match up exactly to the MS ones?

 

[jvd]

I've cleaned up goshes posts and posts in response to gosh . Please ignore him and do not responed to him in the future. It will make cleaning up his posts much easier . Thank you

 

[nAo]

Nvidia quotes 27 flops per pixel pipe: we have 2 fmadd + 1 scalar + 1 nrm per cycle: 16 + 2(1?) + x = 27
Umh..9 or 10 ops for a normalization?
Moreover on NV40 a one clock nrm work on 16bits data, not 32 bits.

 

[AlNom]

Would somecare care to update the chart (for C1) and upload the image, please?

 

[Carl B]

Wait wait - so are we believing that this chart from NVidia is real? I'm not questioning it in a big way, it's just I find it strange that news so big would have been routed through the Chinese press; not the G70 which I know the Taiwanese leakers would be all over, but the confirmation that G70 = RSX. Can we verify that the RSX stuff is official and not extrapolation/speculation on the chart authors part?

 

[nAo]

They say the chart is from Nvidia. We'll see what other reviews will say about it

 

[jvd]

nvidia has already said it was using a next gen pc part. Its sony that was saying it was custom.

The g70 fits if they want to launch anywhere in the world by spring 2006. They would need millions of units completed ahead of time so they would have to start making parts in november rish of 2005 . The g80 is most likely coming out spring 2006but the lead in time would be much less than that of a console

 

[Brimstone]

So the G70 clockspeed is going to be much lower than the RSX? Sony has awesome fabrication ability then.

I suspect the G70 will be power hungry and produce a lot of heat. RSX will be a power guzzler as well, .90 nm compared with .11 nm won't help that much.

 

[Glonk]

G70 is 110nm, RSX is 90nm...isn't it?

The 550MHz for RSX still seems abit optimistic to me.

 

[Carl B]

They say the chart is from Nvidia. We'll see what other reviews will say about it

Oh yeah I know what they're saying.

It's just at the moment it seems SO like G70 that it's a little dissapointing in the novelty department. Do those numbers leave any room for interaction with Cell to perhaps not be reflected there? It just seems like a waste if they don't end up using Cell for anything display-wise.

 

[jvd]

So the G70 clockspeed is going to be much lower than the RSX? Sony has awesome fabrication ability then.

I suspect the G70 will be power hungry and produce a lot of heat. RSX will be a power guzzler as well, .90 nm compared with .11 nm won't help that much.

Dropping the micron process shoudl allow them to run the chip cooler and faster . SO i don't see why they couldn't make it 550 at 65nm 6 months down the line

 

[Rockster]

The following snippet is from the Xenos article shader array portion:

"Additional to the 48 ALU's is specific logic that performs all the pixel shader interpolation calculations which ATI suggests equates to about an extra 33% of pixels shader computational capability."

Is this referring to free norm, or something else?