MDK2 + T&L on a V3?!

[cool]

Yes you've read right! When I activate "T&L" in the MDK2 settings I got 35.2fps, when I deactivate it then I get 41.6fps. How is that possible? Is MDK2 'emulating' T&L with the SSE instruction set?

I've also noticed a big improvement in image quality/detail.
What do you think guys?

Here's my system spec:
PIII-650E
V3-3000AGP
160MB-RAM

MDK2 settings:
Default OpenGL driver
1024x768
Texture quality set to highest
16bit
Bilinear filtering
Mipmap
Full Screen
T&L on/off

 

[Finality]

Not really.

SSE does not 'emulate' T&L settings. It is the actual fpu of cpu that does most of the work. You need to stop listening to the PR departments.

You are fill rate limited by the V3 maybe its time to upgrade.

 

[cool]

Hmm... I was wondering why the fps decreased when I activated T&L on a card that doesn't have a T&L-unit. There must be a reason for that. And the image quality/details went also up. Maybe MDK2 uses the FPU/SSE unit to do the additional calculations when it doesn't find a T&L unit, I don't know...

Fill-rate limited: Yes, I know but I bought the card one year ago and there are some titles coming out that will work fine with my V3. DNF, MP, TF2 etc...

 

[BenSkywalker]

When you enable hardware T&L, you enable high quality specular lighting versus the lesser quality, but much faster, diffuse lighting.

Hardware T&L can only calculate specular lighting on current consumer boards, and it is also the highest quality option that can be reasonably used.

Your CPU isn't emulating hardware T&L, it is simply executing some of the instructions that hardware T&L is hardwired to decode. The code is native to x86, the hardware T&L unit is built to handle certain calculations on its' own. Your CPU can do anything a dedicated unit can, it will just be much slower most of the time.

You want to see the real difference drop your resolution down to the lowest allowed(320x200?? can't remember and don't have it handy) and all quality settings to lowest and then compare the two, hardware T&L vs software.

This situation unfortunately was made more confusing by 3DMark2K which in its' high poly test compares two different things when running hardware T&L vs software/SIMD. Hardware T&L runs the higher quality specular lighting, which is significantly more computationaly intensive, and the CPU/SIMD test runs diffuse lighting. The quality difference can't be seen in that test, but MDK2 demonstrates this fairly well.

BTW- I'm pushing 86.9FPS using the same settings as you with a GeForce1 @130(520MTexels peak vs ~366MTexels for your board if you are running at default). You should be pushing closer to 60FPS if it was a fill limited situation.

 

[cool]

Thanks Luke , for the precise answer. But what did you mena actually with this:

"You should be pushing closer to 60FPS if it was a fill limited situation."

I think it is a fill-rate limited situation as I don't get higher numbers in other OpenGL-games.

In Q3A for example I have the same fps score as in MDK2.
demo001: 42.1 (with WGL 44.9)
demo002: 44.5 (47.0)
quaver: 38.2 (38.2)
nv15demo: 19.5 (14.3)

all in 1024x768 with max. settings and newest V3beta-drivers

 

[BenSkywalker]

Cool



<< &quot;You should be pushing closer to 60FPS if it was a fill limited situation.&quot;

I think it is a fill-rate limited situation as I don't get higher numbers in other OpenGL-games. >>



Q3 has stiffer fillrate demands though. My numbers, same settings as yours-

Demo1 - 73.2

Demo2 - 74.5

Quaver - 66.5

A decent drop over my MDK2 numbers(15FPS give or take). Of course, the best way to see if you are fill limited is to test your own rig at multiple resolutions

 

[cool]

Like said, at 1024x768 all OGL-games are fill-rate limited. If I switch the resolution back to 800x600 then I get over 60fps (demo002) in Q3A and over 50 in MDK2.

 

[cool]

Just a 'small' correction: I enabled in the '3dfx Advanced Features' under OGL/Glide the 'triple buffering' feature. Then I benchmarked MDK2 and the fps increased from 41.6 (double buffered) to 54.3fps. That's an increase of 30%! Cool!

This setting has no effect in Q3A. Maybe there's not enough space in the video-RAM or it's disabled by the OGL-drivers by default.