Perhaps the Quake 2 demo doesn't include the timedemos. I own the full version of the game and used that...

swaaye wrote:

Perhaps the Quake 2 demo doesn't include the timedemos. I own the full version of the game and used that...

I ended up installing the full version of Quake 2, v3.06 and it seems to work now, but the process to start the timedemo is a bit convoluted. Here is what eventually worked:

1) Start new game (ensure the CD is inserted)
2) Press: ` to bring up the console.
3) Type: timedemo 1
4) Type: map demo1.dm2
5) Quake returns "Dup connect received. Ignored" and just sits there.
6) Press: ESC
7) Select the Video option and set your desired resolution
8) Press: Enter
9) The timedemo now runs.

Swaaye, I noticed on your comparison bar chart that you have "no 8-bit"? How do you specify this in QuakeII?

FYI, I ran timedemo1 on a Pentium-266MMX using a Millennium G200 graphics card and got only 10.1 fps at 800x600, while you got 30 fps using a PIII-1400 MHz with the same card.

My test system is Windows 98SE. The latest Matrox driver version which does not have issues with 3DMark99Max is v6.28 (6.82 doesn't work well with several benchmarks for whatever reason).

feipoa wrote:

Swaaye, I noticed on your comparison bar chart that you have "no 8-bit"? How do you specify this in QuakeII?

There is a video menu option for 8-bit palletized (256 color) textures. 8-bit is probably faster on cards with hardware-based palletized texture support, but disabling this setting gives slightly better image quality because the textures are natively 16-bit AFAIK.

feipoa wrote:

FYI, I ran timedemo1 on a Pentium-266MMX using a Millennium G200 graphics card and got only 10.1 fps at 800x600, while you got 30 fps using a PIII-1400 MHz with the same card.

My test system is Windows 98SE. The latest Matrox driver version which does not have issues with 3DMark99Max is v6.28 (6.82 doesn't work well with several benchmarks for whatever reason).

I suppose that the game is CPU limited when run on a Pentium 266.

I think I used the newest driver, which has the bug with transparencies. I don't think I knew of older drivers working better before I ran the tests.

The fact that you were able to run 3DMark99 Max with driver v6.82 implies that your system didn't have the issues mine was having. I couldn't even run 3DMark99 Max with v6.82; the newest that would work for me was v6.28 - perhaps some motherboard hardware limitation? I was using a 430TX board and the PCI version of the Millennium G200, not the AGP version as in your comparison.

Do you happen know if Quake II specifically requires Pentium instructions to run? During the timedemo on my 5x86-133 (2x66) system, it will eventually spit out an illegal operation message, however DOS Quake1 worked fine. MDK Direct3D and MDK Win95 both seemed to work on this 5x86.

Retraction Edit: The Cyrix 5x86-133 (2x66) system completed the QuakeII timedemo with branch prediction disabled. It seems that some programs work with Cyrix 5x86 branch prediction enabled, while others do not. Benchmark score was 7.8 fps at 320x240 and 4.0 fps at 640x480.

swaaye wrote:

I suppose that the game is CPU limited when run on a Pentium 266.

Ok, I see now. You are probably using the "Default OpenGL" setting instead of "software", in which case my 266MMX with PCI Millennium G200 16MB SDRAM gets 21.4 fps at 800x600 8-bit off.

"Software" is the CPU render mode. G200 would just be a dumb framebuffer in that case.

If benchmarking numerous socket 7 CPUs, I wonder if "software" mode would be sufficient?

What do you mean by sufficient?
Software mode utilizes the CPU to render everything. Graphics cards don't really make any difference, that's why it is a good idea to use software mode as a benchmarking tool for CPUs.
Default OpenGL means that the game is utilizing the graphics card to render a lot , if not most, stuff that you see on screen and greatly lifts the CPU of a lot of work. The CPU tends to other stuff, like sound and thus the system runs faster.

I had originally typed "preferred" instead of "sufficient", but altered the wording to hopefully increase unbiased feedback response. Thank you for your input.

A clearer way of wording my question might be:
For the general trend of increasing game performance (fps) with increasing CPU frequency/complexity, will the use of "software" and "OpenGL" both scale up accordingly for the majority of CPUs in the socket 7 class of CPUs? All CPUs will be benchmarked with the same graphics card. Thanks.

Considering you are doing a CPU comparison, I think software mode makes the most sense. It will remove the video card from influence.

The tests I did here were on a very fast CPU (P3 1400) and I was looking specifically at 3D card performance.

Video card's memory and bus speed will still determine software mode performance. Just compare ATI Rage Pro vs a Intel i752 at Quake 😜

Awesome thread!

Here are a couple of my Kyro cards:

feipoa wrote:

I had originally typed "preferred" instead of "sufficient", but altered the wording to hopefully increase unbiased feedback response. Thank you for your input.

A clearer way of wording my question might be:
For the general trend of increasing game performance (fps) with increasing CPU frequency/complexity, will the use of "software" and "OpenGL" both scale up accordingly for the majority of CPUs in the socket 7 class of CPUs? All CPUs will be benchmarked with the same graphics card. Thanks.

Both software and accelerated comparisons could be significant since each would give different perspectives on CPU limitations (and possibly different driver limitations/bias/quirks as well).

Of course, you'd want to compare apples to apples (different CPUs with the same video card set-up) and not software vs accelerated. (assuming the goal was CPU benchmarking -and not comparing something else, like different system builds of approximately the same cost/timeframe)

Hardware and software rendering (as well as various drivers) would tend to have varying bias towards FPU vs ALU intensive tasks as well as varying MMX and 3DNow performance. (and varying bias for specific CPU architectures too -like pentium-specific vs 486 etc -or more generalized to make a good compromise for 486/pentium/5x86/6x86/K5/K6/WinChip/etc -or 386/486SX/DLC for that matter)

Using video cards without hardware T&L would probably be a more useful comparison (and make more sense from a contemporary hardware standpoint -ie not a 1997 CPU with a 2001 GPU), but T&L capable cards could be interesting to compare too. (and might push integer performance to greater importance on the CPU end)

You'd probably want to limit video cards/games/benchmarks to those that were stable/bug-free on all the hardware platforms being tested so as to not skew results from bugs that impact some hardware differently than others.

I'm running 3DMark99 MAX on a Matrox Millennium 2, and the two tests I've completed so far have had no textures on anything.

Is that right?

SquallStrife wrote:

I'm running 3DMark99 MAX on a Matrox Millennium 2, and the two tests I've completed so far have had no textures on anything.

Is that right?

How much memory does the card have? Sounds like you don't enough to load the textures.

It's a 4MB card. The Voodoo beside it is also 4MB, but it works perfectly.

Weird huh?

Reminds me of me testing a Matrox G450 a couple of years ago on my Pentium 3. I think it has 32MB. Anyway, I tried to play NFS 3, and the game was running without textures.

Yeah, everything's white, it looks weird huh?

http://www.youtube.com/watch?v=NvIWMwuUDzE

(You may hear my nose whistling for pretty much the whole video, I have a bit of a stuffy nose at the moment, so sorry about that. I used the Youtube audio swap thing, but as of now that's still processing, so my disgusting nose noises should be gone soon!)

Edit: Disgusting nose sound has been replaced with some lovely Beethoven.

yeap, it was just like that. 😀

Matrox Millennium II, Mystique and ViRGE do that in games as well, like Shogo and Blood II!