K
Keanu
I've posted this as a reply to a review, and in the doom forum.
It's not new.. But easy to miss.
If you own an excellent radeon card and wish it ran faster
and better with Doom3; You can start by getting the ATI
Catalyst beta 4.9 HOTFIX.. It works rock solid and improves
the framerate quite a bit.
But the last part really makes doom3 smooth:
In case you missed the tweak. Here it is in ZIP format.
http://alexharis37.free.fr/doom/humusd3atitweak.zip
It's some sort of hotfix for the pixel shader to be ATI radeon friendly.
It does some float operations better, improves speed as well as
framerate. It also fixes specular glitches (pixies along some axis).
The speed gain to be expected is around 18%
Just unzip in your doom folder and it'll create a glprog within the base
folder with the now famous interaction.vfp by Humus, now working
for ATI... So this is not gonna break anything.
Hope it helps..
And people wondering about the game being dark should
dig informations on how to edit the DoomConfig.cfg and
up the gamma.... Brightness alone is washing the colours away.
r_gamma "1.4" is your friend !......USE IT !...
Hope it helps,
N³o
Looking around I found that Humus now works for ATI... CEWL !,
http://www.xbitlabs.com/news/other/display/20040808022148.html
Good for him.
Here for the records Humus own explanations for the increased speed:
-------------
My theory is that it simply causes too much texture cache trashing. Remember that the specular component may vary a lot between
nearby pixels, so the access behavior will be quite erratic. Another thing could be that the number of texture accesses are so
high. There are seven different textures accessed in that shader. Now I don't have that much insight in how our hardware works,
maybe someone with more knowledge about that can fill me in. (sireric, where art thou? )
I was thinking that it wouldn't make much difference on nVidia, or perhaps be slower. I don't think Carmack just threw it in out of
pure habit, but he probably had a rational reason for it. I can see that it may not make much difference on a 6800, but I wouldn't
be surprised if the GF 5x00 series cards were much slower using math, they weren't that strong on floating point shaders as we all
know, so moving it to a texture lookup probably improved performance on that card, which I guess he used at the time he wrote that
shader.
No, both coordinates are the same as they are written by the DOT instruction. I would guess he has bound a Nx1 texture to that
unit.
Humus.
--------------- http://www.beyond3d.com/forum/viewtopic.php?p=342441#342441
It's not new.. But easy to miss.
If you own an excellent radeon card and wish it ran faster
and better with Doom3; You can start by getting the ATI
Catalyst beta 4.9 HOTFIX.. It works rock solid and improves
the framerate quite a bit.
But the last part really makes doom3 smooth:
In case you missed the tweak. Here it is in ZIP format.
http://alexharis37.free.fr/doom/humusd3atitweak.zip
It's some sort of hotfix for the pixel shader to be ATI radeon friendly.
It does some float operations better, improves speed as well as
framerate. It also fixes specular glitches (pixies along some axis).
The speed gain to be expected is around 18%
Just unzip in your doom folder and it'll create a glprog within the base
folder with the now famous interaction.vfp by Humus, now working
for ATI... So this is not gonna break anything.
Hope it helps..
And people wondering about the game being dark should
dig informations on how to edit the DoomConfig.cfg and
up the gamma.... Brightness alone is washing the colours away.
r_gamma "1.4" is your friend !......USE IT !...
Hope it helps,
N³o
Looking around I found that Humus now works for ATI... CEWL !,
http://www.xbitlabs.com/news/other/display/20040808022148.html
Good for him.
Here for the records Humus own explanations for the increased speed:
-------------
Chalnoth said:Why would a single instruction on a 23-instruction shader make that much of a difference?
My theory is that it simply causes too much texture cache trashing. Remember that the specular component may vary a lot between
nearby pixels, so the access behavior will be quite erratic. Another thing could be that the number of texture accesses are so
high. There are seven different textures accessed in that shader. Now I don't have that much insight in how our hardware works,
maybe someone with more knowledge about that can fill me in. (sireric, where art thou? )
By the way, I tried it on my GeForce 6800, and could measure no performance difference. But, that doesn't mean much, as I'm pretty
heavily CPU-limited in the timedemo.
I was thinking that it wouldn't make much difference on nVidia, or perhaps be slower. I don't think Carmack just threw it in out of
pure habit, but he probably had a rational reason for it. I can see that it may not make much difference on a 6800, but I wouldn't
be surprised if the GF 5x00 series cards were much slower using math, they weren't that strong on floating point shaders as we all
know, so moving it to a texture lookup probably improved performance on that card, which I guess he used at the time he wrote that
shader.
exponentiate >two components (2D texture lookup), so shouldn't you use 2 POW instructions?Regardless, it looks like you're only exponentiating the x-component of the specular vector, while the LUT looks like it will
No, both coordinates are the same as they are written by the DOT instruction. I would guess he has bound a Nx1 texture to that
unit.
Humus.
--------------- http://www.beyond3d.com/forum/viewtopic.php?p=342441#342441