FP32 will make NV40 slow?

[Bob3D]

Looking at this bench from anand we can see that NV40 running PF32 mode is slow than R420 running with FP24.

http://www.anandtech.com/video/showdoc.html?i=2044&p=9

So, when (if?) games start using FP32 on Nvidia hardware they will run slow compared to ATI FP24?
We can also see that reducing FP32 to FP16(?) make the NV40 double the speed.
Do you guys think that Nvidia will still be using FP16 because the performance impact?

 

[Dave Baumann]

As far as games are concerned there is no "FP24" or "FP32", only normal (high) precision and partial precision. The precision running depends on the architecture its running on.

 

[Bob3D]

Thanks Dave.
Let's see if i catch it.
ATI FP24 is partial precision, and nvidia FP32 is full precision.
ATI hardware can only do partial precision, and nvidia can do full (32FP) and partial (16FP)

Question
1-ATI hardware will always run in PP, but nvidia can alternate between FP and PP. Looking at anand bench we can see that nvidia have some speed bost running PP against FP.
Let's take farcry as a example. If nvidia use FP we don't have some visual bugs, like the ones found when using FP16. But the performance will be worse than ATI PP?

 

[jvd]

Thanks Dave.
Let's see if i catch it.
ATI FP24 is partial precision, and nvidia FP32 is full precision.
ATI hardware can only do partial precision, and nvidia can do full (32FP) and partial (16FP)

Question
1-ATI hardware will always run in PP, but nvidia can alternate between FP and PP. Looking at anand bench we can see that nvidia have some speed bost running PP against FP.
Let's take farcry as a example. If nvidia use FP we don't have some visual bugs, like the ones found when using FP16. But the performance will be worse than ATI PP?

fp24 is not partial percision . fp 16 is . Ati hardware will allways run in fp24. Nv can do 32fp , 16fp , 16fx , 12 and 8 int (think i have them all right)

Fp24 is is considered full persicion.

 

[karlotta]

24bit is FULL minimum
32 is FULL

 

[akira888]

No, ATI's 24-bit FP precision is full precision according to the DirectX9 spec, as is Nvidia's 32-bit FP precision. NV's 16 bit FP on the other hand is partial precision.

 

[Bob3D]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

 

[karlotta]

welcome to DX9 hardware since the 9700pro and nvwhatever cards. or the last two years.

 

[Ostsol]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

Sorta. . . It's more of an example of CryTech forcing NVidia cards to use partial precision. Not exactly NVidia's fault (unless it was them who asked CryTech to do such a thing).

 

[Bob3D]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

Sorta. . . It's more of an example of CryTech forcing NVidia cards to use partial precision. Not exactly NVidia's fault (unless it was them who asked CryTech to do such a thing).

Ok
So if CryTek force nvidia to use FP we will see worse performance than ATI using FP?

 

[Ostsol]

Looking at the various reviews, even with PP NVidia performs worse.

 

[Bob3D]

Looking at the various reviews, even with PP NVidia performs worse.

Thanks Ostsol
I hope this is related to bad drivers. Soon nvidia should fix this, or not.
Strange to see shader performance in nvidia new top card with the same problems we see in NV3X family.
I can conclude that IF any game use FP on NV40 cards the performance will always be worse than R420 right?

 

[{Sniping}Waste]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

Sorta. . . It's more of an example of CryTech forcing NVidia cards to use partial precision. Not exactly NVidia's fault (unless it was them who asked CryTech to do such a thing).

Mabe not. If the R3xx show the same problems using the NV40 ID then its not a PP problem but somthing else because the R3XX force all pixel shaders to FP24.
http://www.driverheaven.net/showthread.php?s=&threadid=44253&perpage=15&pagenumber=1

and here is some more
http://translate.google.com/transla...e=UTF-8&oe=UTF-8&prev=/language_tools

 

[jvd]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

Sorta. . . It's more of an example of CryTech forcing NVidia cards to use partial precision. Not exactly NVidia's fault (unless it was them who asked CryTech to do such a thing).

Mabe not. If the R3xx show the same problems using the NV40 ID then its not a PP problem but somthing else because the R3XX force all pixel shaders to FP24.
http://www.driverheaven.net/showthread.php?s=&threadid=44253&perpage=15&pagenumber=1

there is no r300 path. Its the standard path. so nvidia should just be running it at 32fp , if you force ati to use the n30 path it will force fp16 which the radeons don't support.

So its not as simple as that

 

[Bob3D]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

Sorta. . . It's more of an example of CryTech forcing NVidia cards to use partial precision. Not exactly NVidia's fault (unless it was them who asked CryTech to do such a thing).

Mabe not. If the R3xx show the same problems using the NV40 ID then its not a PP problem but somthing else because the R3XX force all pixel shaders to FP24.
http://www.driverheaven.net/showthread.php?s=&threadid=44253&perpage=15&pagenumber=1

How can NV40 cards run with the same FP of R3xx cards if they don't have FP24?

 

[Bob3D]

Thanks JVD
That's my question.

 

[{Sniping}Waste]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

Sorta. . . It's more of an example of CryTech forcing NVidia cards to use partial precision. Not exactly NVidia's fault (unless it was them who asked CryTech to do such a thing).

Mabe not. If the R3xx show the same problems using the NV40 ID then its not a PP problem but somthing else because the R3XX force all pixel shaders to FP24.
http://www.driverheaven.net/showthread.php?s=&threadid=44253&perpage=15&pagenumber=1

there is no r300 path. Its the standard path. so nvidia should just be running it at 32fp , if you force ati to use the n30 path it will force fp16 which the radeons don't support.

So its not as simple as that

ATI R3XX don't support FP16 so it does FP24 and there is still banding if using the NV id on a R3xx so its not a color persetion problem but somthing else.

 

[jvd]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

Sorta. . . It's more of an example of CryTech forcing NVidia cards to use partial precision. Not exactly NVidia's fault (unless it was them who asked CryTech to do such a thing).

Mabe not. If the R3xx show the same problems using the NV40 ID then its not a PP problem but somthing else because the R3XX force all pixel shaders to FP24.
http://www.driverheaven.net/showthread.php?s=&threadid=44253&perpage=15&pagenumber=1

How can NV40 cards run with the same FP of R3xx cards if they don't have FP24?

by forcing the standard path (what the r3x0 series runs it will default to 32bit for the geforce series .

 

[jvd]

Thanks
So, here is a example of Nvidia using PP and ATI using FP?
http://www.tomshardware.com/graphic/20040414/geforce_6800-46.html
If yes can we conclude, based on anand shadermark results, that if nvidia use FP performance will be probaly worse than ATI in FP?

Sorta. . . It's more of an example of CryTech forcing NVidia cards to use partial precision. Not exactly NVidia's fault (unless it was them who asked CryTech to do such a thing).

Mabe not. If the R3xx show the same problems using the NV40 ID then its not a PP problem but somthing else because the R3XX force all pixel shaders to FP24.
http://www.driverheaven.net/showthread.php?s=&threadid=44253&perpage=15&pagenumber=1

How can NV40 cards run with the same FP of R3xx cards if they don't have FP24?

by forcing the standard path (what the r3x0 series runs it will default to 32bit for the geforce series .

it could just be crap in their drivers to make it run faster that is left over from the nv30 . Or it could just be the best image quality the card can offer.

But it should be running fp32 when detected as a r300
sssssss

 

[AlphaWolf]

r300 should quite simply ignore _pp hints and render fp24. If the banding/IQ problem still exists with radeon cards running nv30 path then _pp isnt the problem.