Quak4 - Quake 4 tuning tool for NVIDIA G70/NV40/NV30 Series GPU's

[tertsi]

Originally Posted by Mr. Nice

Yeah your right, I said as much months back in the 3mood thread about fp16. On my 6800GT the performance advantage was quite negligible as well for this option.
As a point of order though, it's not fp12, that's fx12.

On another couple of the settings, both doom3 and quake4 actually use precision hint fastest in the standard interaction.vxp already, and quake 4 by default already does what the "swizzle" option does (which is mask the final result by xyz, ie skips writing out the redundant w channel). Indeed, quake 4's interaction.vxp is replete with quite aggressive write masks (as opposed to doom3's, which contains none). Finally, quake 4 already does the specular with maths, not a texture lookup, and indeed it does it with one less instruction then quak4/3mood . Quak4/3mood subtract's 0.75, squares it, then multiplies by 16, 3 instructions. Quake 4 uses a single MAD instruction to multiple by 4 then subtract 3, and then squares it, which is mathematically equivalent (multiplying by 4 before squaring is the same as multiplying by 16 afterwards, since 4 squared is 16, and similarly 3 is subtracted instead of 0.75 since 4x0.75=3).

Well older drivers has "ARB_precision_hint_fastest" and "FP16 precision in arithmetic operations" conflict, so it is recommended to use the 80 series drivers. Also "FX12 ALU operations" option with "Cube normalization" option may cause a very "blocked" highlights of specular. "FX12 ALU operations" option should be avoid on the NV4x and G7x.

I haven't seen any problems with "FP16" options on the NV4x and G7x but there could be a problems on the NV3x because I didn't test my Q4 tweaks on my 5800u or 5900u.

"SUB/MUL/MUL" should be most of cases 1-2fps faster in Doom 3 and Quake 4 than "MAD/MUL" though "MAD/MUL" was one cycle faster in my fragment program analysis.

EDIT! Quake 4 standard interaction.vfp is optimized for ATI Vec3D + Vec1D hardware....

[tertsi]

Originally Posted by Razor3DC

Hi tertsi,

what about this "demo1"?
Is this an official timedemo?

Per default it doesn't exist...

greets

Razor

that was Doom 3 bench with demo1... Quake 4's build-in timedemo mode (no shadows) is borked so it is recommended to use Fraps for benching!

[Mr. Nice]

Originally Posted by tertsi

WI haven't seen any problems with "FP16" options on the NV4x and G7x but there could be a problems on the NV3x because I didn't test my Q4 tweaks on my 5800u or 5900u.

It's those ubiquitous (in both doom3 and quake4) floor gratings, they look fairly fugly with fp16 on (and that's on my 6800GT, ie NV40). Using short temps though is fine.

Originally Posted by tertsi

"SUB/MUL/MUL" should be most of cases 1-2fps faster in Doom 3 and Quake 4 than "MAD/MUL" though "MAD/MUL" was one cycle faster in my fragment program analysis.

Hmmn? If SUB/MUL/MUL uses one more cycle then MAD/MUL (which seems fairly intuitive), why would it be 1-2fps faster? Especially since one of your other optimizations is to replace a MUL/ADD with a MAD? (Although I do see why it was left as MUL/ADD, it looks neater due to the ADD being the summation of the results of two lots of SUB/DP3/RSQ/MUL's).

Originally Posted by tertsi

EDIT! Quake 4 standard interaction.vfp is optimized for ATI Vec3D + Vec1D hardware....

Heh, well structurally the only noticable difference from doom3's is using the specular maths instead of a lookup, and all those masks, so either way I can't see this being an issue for Nvidia hardware. I did note that a comment mentioned the use of .w was more optimal for ATI then .x, presumably because ATI's ability to do 3 component vector plus one scalar in the same pass requires that the scalar is in the .w component (Nvidia hardware is more flexible in this regard).

[saweetnesstrev]

Tertsi i know your a busy man, but did you see my ?

[tertsi]

Originally Posted by saweetnesstrev

Tertsi i know your a busy man, but did you see my ?

Oops... yes... use 5600 or 5800 profile.

[tertsi]

Originally Posted by Mr. Nice

It's those ubiquitous (in both doom3 and quake4) floor gratings, they look fairly fugly with fp16 on (and that's on my 6800GT, ie NV40). Using short temps though is fine.

I haven't noticed any problems with/without ARB_precision_hint_fastest, FP16 ALU Operations or Short Temps (75 or 80 series drivers).

Originally Posted by Mr. Nice

Hmmn? If SUB/MUL/MUL uses one more cycle then MAD/MUL (which seems fairly intuitive), why would it be 1-2fps faster? Especially since one of your other optimizations is to replace a MUL/ADD with a MAD? (Although I do see why it was left as MUL/ADD, it looks neater due to the ADD being the summation of the results of two lots of SUB/DP3/RSQ/MUL's).

Well 1 cycle more gpu time in NVShaderPerf doesn't mean really nothing... like I said, "SUB/MUL/MUL" is faster IN DOOM 3 than "MAD/MUL"

MUL/ADD optimization is for vertex programs.

[saweetnesstrev]

Originally Posted by tertsi

Oops... yes... use 5600 or 5800 profile.

Thank you mr. tertsi

[saweetnesstrev]

FX 5700 Ultra - 81.87 drivers,, everythign is black in performance mode with this app.. just thought you should know

[OWA]

Originally Posted by Mr. Nice

It's those ubiquitous (in both doom3 and quake4) floor gratings, they look fairly fugly with fp16 on (and that's on my 6800GT, ie NV40). Using short temps though is fine.

Hmm, it looks okay to me. I can't really tell much of a difference, if any, with just the fp16 enabled.

Doom 3 default...........Doom 3 using mod with just fp16 enabled


Doom 3 default...........Doom 3 using mod.......Doom 3 using mod, no fp16


Quake 4 default..........Quake 4 using mod with fp16 enabled


I compared several other areas in Quake 4 and Doom 3 and only two out of 8 places I compared looked worse. In those two cases it was mainly the specular option making an already poor looking grating easier to see. I don't really see a difference with the fp16 option but maybe things are different with the 7800s or something.

[ssj4goku887]

Even if there is not that much IQ difference, the face is there IS an iq difference and why would you lower the iq to increase performance on a 6800 or 7800 when you can have max quality ANYWAY and not get a performance drop? why would you buy a 400+ dollar card only to lower the quality on it?

[OWA]

Because even with 6800s and 7800s there are some areas where an increase in framerate would be nice. Also, while I didn't state it absolutely, I couldn't tell any difference with fp16 on. If I can't tell a difference when comparing static screenshots, why turn it off since there is no way I'm going to notice anything when moving.

[halcy]

I've got a 7800GTX and looking for some speedups for multiplayer.. can this utility be used to give more speed instead of IQ? And if so what order of magnitude can be expected?

Thanks