Hi buddies!

I think that one of my classic computers is not performing ok for its specifications, specially under Direct3D. I were looking for benchmarks of similar systems on the internet and the results of my computer are way under... I want to know your oppinion.

My computer is:

Motherboard VIA Apollo MV3P based 100FSB, not sure of the brand.
Processor AMD K6-2 450Mhz
128Mb RAM
Graphics Voodoo3 3000 AGP with latest 3dfx reference drivers.
Sound Blaster Live! 1024 sound card.
OS Win98 SE.

And these are the results of some benchmarks I did:

Quake II v3.20 3DNow! optimized:
demo1: 58.9 fps
demo2: 56.2 fps
massive1: 43.5 fps

Quake III v1.11:
demo1: 25 fps
demo2: 24.4 fps

3Dmark99 Max:
1807 3DMarks
4740      CPU 3DMarks

Full 3Dmark99 results:
Rendering Platform      Voodoo3 AGP
Resolution      800*600
Color Depth      16-bit Color
Frame Buffer      Triple buffering
Refresh Rate      75 Hz
CPU Optimization      AMD 3DNow!(tm)
3DMark Result      1807.35      3DMarks
Synthetic CPU 3D Speed      4739.91      CPU 3DMarks
Rasterizer Score      2256.38      3DRasterMarks
Game 1 - Race      26.62      FPS
Game 2 - First Person      13.68      FPS
Fill Rate      154.17      MTexels/s
Fill Rate With Multi-Texturing      292.59      MTexels/s
2MB Texture Rendering Speed      247.43      FPS
4MB Texture Rendering Speed      236.06      FPS
8MB Texture Rendering Speed      166.15      FPS
16MB Texture Rendering Speed      1.63      FPS
32MB Texture Rendering Speed      0.81      FPS
Bump Mapping Emboss, 3-pass      100.81      FPS
Bump Mapping Emboss, 2-pass      131.55      FPS
Bump Mapping Emboss, 1-pass      0.00      FPS
Point Sample Texture Filtering Speed      102.51      %
Bilinear Texture Filtering Speed      100.00      %
Trilinear Texture Filtering Speed      57.54      %
Anisotropic Texture Filtering Speed      0.00      %
6 Pixel/individual      288.89      KPolygons/s
6 Pixel/strips      413.32      KPolygons/s
25 Pixel/individual      287.08      KPolygons/s
25 Pixel/strips      466.67      KPolygons/s
50 Pixel/individual      289.65      KPolygons/s
50 Pixel/strips      472.87      KPolygons/s
250 Pixel/individual      292.97      KPolygons/s
250 Pixel/strips      472.84      KPolygons/s
1000 Pixel/individual      146.69      KPolygons/s
1000 Pixel/strips      148.82      KPolygons/s
Processor Type      AMD-K6(tm) 3D processor
Processor Speed      450 MHz
L1 Cache size      64 KB
L2 Cache size      None
Physical Memory      128 MB

Well, I think that's all, if you want to know any other parameter please tell me. What do you think about the performance? ok or too low?

Thank you guys.

Two more benchmarks...

Quake II Crusher: 28 fps

Unreal Tournament UTbench.dem: min 7,37 fps, max 23,56 fps, average 13,91 fps.

That's quite low... Specially in UT, that is supposedly to be 3dfx and Glide territory...

Cheers

Try benchmarking Unreal v226f. I have benchmark (click) using a Pentium II 450. I don't have benchmarks using a K6-2 450. Keep in mind that the K6-2 wasn't very good with FPU-calculations. I won't be surprised if a Celeron 300A will perform equal to a K6-2 450 using Quake II software-rendered benchmarks.

Thandor is right, the problem is CPU. Voodoo3 needs a fast CPU or else the it will not perform very good. And extra 128 MB RAM might also help.

Thank you for your answers!

Well, I know the K6-2 has two weak points: FPU and no L2 cache. But its hard point is 3dNow!, that with aproppiate optimizations can match or outperform Intel equivalents, like in Quake 2 AMD version.
The problem is that I saw benchmarks of K6 performing way better than mine. That's because I think there is a problem with my system. For example: http://www.anandtech.com/show/160/10 A K6-2 333 with Voodoo2 performing better than mine in Q2, no way...

Thandor, your page is great. I've also tested with 3dmark2000 and got 1003 points, about 500 less than your PII. I think is too much.

I will also test with Unreal. How do you do a benchmark with it?

Thanks. 3DMark2000 is CPU-limited. Look at the small differences between Voodoo 2, 3, 4 and 5 (or even Banshee). The Pentium III-S 1400 can score around 5000 Marks which is a lot higher than 1500. (note: for some reason I haven't included Voodoo 3 3000 <> Pentium III-S 1400 results. Still on the todo list.)

My benchmark method with Unreal (click):

Quote:

I used the command 'timedemo 1' in the console. The console can be accessed using the ~ key. Run it twice or more to be sure you have a correct score. All the graphic options were left at defaults which means 'High texture detail'.

I start the timedemo directly from the intro/menu. I don't start a new game or whatsoever.

Well, I did the test with Unreal and these are the results:

1280*1024: 26.5 fps
1024*768: 29.8 fps
800*600: 32.5 fps
640*480: 33.1 fps

What do you think about it?

I've also did a benchmark with Quake 2 in software mode and got 15.5 fps. I see your K6 450 did 15.2 in that test. I am using the 3dnow optimized 3.20 version. Are you using the same? If so, it looks like they are performing similar under software render.

I'll put them in a list:

Resolution 800x600:

Voodoo3 3000 16MB AGP(P2 450)	62.6
Banshee AGP 16MB AGP (P2 450)	32.06
Your V3 3000AGP (K6-2 450)	32.5

The Pentium II is clearly faster. Even with slower graphic cards it can match or excess your speed. I'm not sure how Unreal runs on different CPU's (from Intel, Cyrix, AMD) but I do recall that I've also run an MVP3-based motherboard with K6-2 which ran quite slow. Perhaps the lack of on-die L2-cache kills the performance? this Tom's Hardware article shows a benchmark using Unreal Tournament and the scores are as following:

Celeron 500	28.8
AMD K6-2+ 500	21.1
AMD K6-2 500	13.3

The difference between the K6-2 and the K6-2+ (having no on-die L2-cache or 128KB on-die L2-cache) is huge! Can you check the L2-cache on your motherboard and see if it's enabled or actually there at all? The chipset can support up to 2MB L2-cache but support might also differ between motherboard brands or just the fact that memory chips are soldered and you can't do much with them. Also check BIOS settings and make sure that everything is configured well.

In order to ensure performance it's also important to have your memory running correctly. It should run at 100MHz and lower latencies are better.

I have also used version Quake II 3.20. It seems that the K6-2 and Pentium II perform almost equal in this version.

Hey! Sorry for the long time since my last answer, I was so busy...

Well, I think that the lack of on die L2 cache is decisive, but I am still thinking there is something wrong with my system. Even more after trying another Unreal engine game, Deus ex, with very poor performance, almost unplayable. And it does't matter if I put the video settings to max or min, I get the same horrible performance, very annoying...

My motherboard has 512Kb of external cache, not sure if it is ok or a bit low, and I have checked BIOS cofing a thousand of times, everything is ok, CPU and motherboard caches are enabled. I think the RAM is running at 100 MHz, because the FSB is, but I have no idea of latencies. The 128Mb are divided into two 64Mb modules, but I don't think this could be bad.

Any other ideas?

Thank you! :)

UPDATE:

I also have another K6 system in stand-by but I've put it working for compare. It has a K6-2 300 processor and a Voodoo 1 4mb graphics card.
With UTBench I get 14.9 fps (about 1fps faster than the 450). I can't believe... A K6-2 300 with Voodoo1 beating a K6-2 450 with Voodoo3 in Unreal Tournament! :o
With Quake 2 in software mode, I get 15.5 fps, exactly the same than the 450 did.
The last test I did was Quake 3, I use MesaFx as opengl driver for the Voodoo 1 and I got 20fps in demo1. The 450 did 25fps, a little bit faster but I think it is not enought considering the power difference...

I think that definitely there is something wrong with my 450 system, but don't know what, the system is stable and does not give any errors...