Carmack blurp on NV shader performance.

micron

Diamond Viper 550
Veteran
Gamers Depot
We emailed id Software guru, John Carmack about his experience with NV3x hardware on Pixel Shading performance, and this was his reply:

GD: John, we've found that NVIDIA hardware seems to come to a crawl whenever Pixel Shader's are involved, namely PS 2.0..

Have you witnessed any of this while testing under the Doom3 environment?

"Yes. NV30 class hardware can run the ARB2 path that uses ARB_fragment_program, but it is very slow, which is why I have a separate NV30 back end that uses NV_fragment_program to specify most of the operations as 12 or 16 bit instead of 32 bit."

John Carmack
http://www.gamersdepot.com/hardware/video_cards/ati_vs_nvidia/dx9_desktop/001.htm
 
i don't know but it just doesn't sound like a carmack answer. It's way to short and untechie :D
 
tEd said:
i don't know but it just doesn't sound like a carmack answer. It's way to short and untechie :D

This is basicly the same as JC said a long time ago in one of .plans (the NV30 was half as fast the R300 using the ARB2 path).
 
K.I.L.E.R said:
So, what's new? :)

A couple things are "new"

1) As stated earlier, it's more or less confirmation that FX12 is being used, not just FP16, compared to the ARB2 path.

2) IIRC, Carmack's earlier plan talked about "nVidia promising" to increase ARB2 performance to get it up to par. (All those driver "sensitivities.") It appears that hasn't happened. Not a real surprise to most people, but it is at least a sort-of confirmation.
 
Joe DeFuria said:
K.I.L.E.R said:
So, what's new? :)

A couple things are "new"

1) As stated earlier, it's more or less confirmation that FX12 is being used, not just FP16, compared to the ARB2 path.

2) IIRC, Carmack's earlier plan talked about "nVidia promising" to increase ARB2 performance to get it up to par. (All those driver "sensitivities.") It appears that hasn't happened. Not a real surprise to most people, but it is at least a sort-of confirmation.
How bad is it going to look using FX12? Will it have to be noticable or can JC make it work?

DaveBaumann said:
Except FX12 is being removed from NV35 and on...
So the new FX's won't run D3's nVidia path? :?
 
DaveBaumann said:
Except FX12 is being removed from NV35 and on...

I find that strange, since their value and mid range line-up need it, NV31 and NV34. The only value NV35 is a 5900 -Non-Ultra which is still quite expensive. ?
 
This is new news to me :p So are the INT12 units from NV30 completly gone in NV35? Does NV35 jsut convert INT12 to FP16? Is NV35 just strictly FP 16 and FP32?
 
We emailed id Software guru, John Carmack about his experience with NV3x hardware on Pixel Shading performance, and this was his reply:

GD: John, we've found that NVIDIA hardware seems to come to a crawl whenever Pixel Shader's are involved, namely PS 2.0..

Have you witnessed any of this while testing under the Doom3 environment?

"Yes. NV30 class hardware can run the ARB2 path that uses ARB_fragment_program, but it is very slow, which is why I have a separate NV30 back end that uses NV_fragment_program to specify most of the operations as 12 or 16 bit instead of 32 bit."

John Carmack

Thats very interesting because carmack said earlier on that the NV30 is a full float point pathline now unless there is NEVER any rounding when he does the FX12 calls then he has lied and all the ATI boys can finally yell about how they only got full float point graphics.
 
Back
Top