Nice thread. Keep 'em coming 😀

Asus V9950 Ultra
Geforce FX 5900 Ultra
Driver used: 45.23

3dmark 99 MAX: 15424 3D marks.

3dmark 2000: 12900 3D marks.

3dmark 2001: 12758 3D marks.

GL-Quake 640*480 32bit: 568.8 FPS

GL-Quake 800*600 32bit: 560.4 FPS

GL-Quake 1024*768 32bit: 444.8 FPS

GL-Quake 1280*960 32bit: 303.3 FPS

Quake II 640*480 32bit: 440.8 FPS

Quake II 800*600 32bit: 447.1 FPS. I benched 640*480 and 800*600 3 times each just to be sure, 800*600 was faster every time.

Quake II 1024*768 32bit: 436.1 FPS

Quake II 1280*960 32bit: 366.9 FPS

Quake III 640*480 32bit: 225.2 FPS

Quake III 800*600 32bit: 223.7 FPS

Quake III 1024*768 32bit: 222.1 FPS

Quake III 1280*1024 32bit: 216.1 FPS

Unreal Gold (Direct 3D) 640*480 32bit: 152 FPS

Unreal Gold (Direct 3D) 800*600 32bit: 149.2 FPS

Unreal Gold (Direct 3D) 1024*768 32bit: 145.6 FPS

Unreal Gold (Direct 3D) 1280*720 32bit: 156.1 FPS. 1280*768 would not work, the increased FPS is mainly from the WS FOV not the resolution.

Overclocking

The stock speed of this card is 450 MHz core, 850 MHz memory.
The overclocked speed is 500 MHz core, 1000 MHz memory. (Not the max OC the card can take but enough)

Overclocked 3dmark 2001 result: 13038 3D marks. Bottlenecked by the CPU.

Quake III 1280*1024 OC score: 217.8 FPS. Bottlenecked by the CPU.

Quake I 1280*960 OC score: 338.2 FPS. Here we get a proportional FPS increase from of the OC.

PixelView Geforce 2 MX 400 64MB
200 Mhz Core, 332 Mhz memory (64bit).
Driver used: 45.23

3dmark 99 MAX: 10001 3D marks.

3dmark 2000: 5693 3D marks.

3dmark 2001: 3077 3D marks.

GL-Quake 640*480 32bit: 189.4 FPS

GL-Quake 800*600 32bit: 127.8 FPS

GL-Quake 1024*768 32bit: 82.1 FPS

GL-Quake 1280*960 32bit: 53.6 FPS

Quake II 640*480 32bit: 221.6 FPS

Quake II 800*600 32bit: 153.7 FPS

Quake II 1024*768 32bit: 102.5 FPS

Quake II 1280*960 32bit: 67.0 FPS

Quake III 640*480 32bit: 155.4 FPS

Quake III 800*600 32bit: 105.8 FPS

Quake III 1024*768 32bit: 66.2 FPS

Quake III 1280*1024 32bit: 39.4 FPS

Unreal Gold (Direct 3D) 640*480 32bit: 127.7 FPS

Unreal Gold (Direct 3D) 800*600 32bit: 91.8 FPS

Unreal Gold (Direct 3D) 1024*768 32bit: 59.8 FPS

Unreal Gold (Direct 3D) 1280*768 32bit: 49.2 FPS

Overclocking

The stock speed of this card is 200 MHz core, 333 MHz memory (64bit).
The overclocked speed is 230 MHz core, 360 MHz memory.

Overclocked 3dmark 2001 result: 3289 3D marks. No much of a gain.

Quake III 1280*1024 OC score: 42.8 FPS, Not much of a gain but the game is playable at this resolution.

Leadtek TNT2 32MB
143 MHz Core, 150 MHz memory (128bit).
Driver used: 45.23

3dmark 99 MAX: 8211 3D marks.

3dmark 2000: 3552 3D marks.

3dmark 2001: 1832 3D marks.

GL-Quake 640*480 32bit: 124.7 FPS

GL-Quake 800*600 32bit: 82.2 FPS

GL-Quake 1024*768 32bit: 44.9 FPS

GL-Quake 1280*960 32bit: 28.1 FPS

Quake II 640*480 32bit: 125.3 FPS

Quake II 800*600 32bit: 77.6 FPS

Quake II 1024*768 32bit: 54 FPS

Quake II 1280*960 32bit: 33.1 FPS

Quake III 640*480 32bit: 74.7 FPS

Quake III 800*600 32bit: 44.8 FPS

Quake III 1024*768 32bit: 30.6 FPS

Quake III 1280*1024 32bit: 17.3 FPS

Unreal Gold (Direct 3D) 640*480 32bit: 76.4 FPS

Unreal Gold (Direct 3D) 800*600 32bit: 47.7 FPS

Unreal Gold (Direct 3D) 1024*768 32bit: 33 FPS

Unreal Gold (Direct 3D) 1280*768 32bit: 26 FPS

Overclocking

The stock speed of this card is 143 MHz core, 150 MHz memory.
The overclocked speed is 175 MHz core, 183 MHz memory.

Overclocked 3dmark 2001 result: 2252 3D marks. (The driver number in the image is a typo, it should say 45.23)

Overclocked Quake 3 results

1024*768, 38.7 FPS (Seemingly fluid)

1280*1024, 22.6 FPS (Unplayable)

Pine VARM645PS TNT2 M64 32MB
125 MHz Core, 143 MHz memory (64bit)
Driver used: 45.23

3dmark 99 MAX: 5280 3D marks.

3dmark 2000: 2308 3D marks.

3dmark 2001: 983 3D marks.

GL-Quake 640*480 32bit: 73.5 FPS

GL-Quake 800*600 32bit: 49.2 FPS

GL-Quake 1024*768 32bit: 25.3 FPS

GL-Quake 1280*960 32bit: 14.2 FPS

Quake II 640*480 32bit: 72.6 FPS

Quake II 800*600 32bit: 48.1 FPS

Quake II 1024*768 32bit: 31.3 FPS

Quake II 1280*960 32bit: 17.1 FPS

Quake III 640*480 32bit: 40.5 FPS

Quake III 800*600 32bit: 26.5 FPS

Quake III 1024*768 32bit: 17 FPS

Quake III 1280*1024 32bit: 8.5 FPS

Unreal Gold (Direct 3D) 640*480 32bit: 41.6 FPS

Unreal Gold (Direct 3D) 800*600 32bit: 27.5 FPS

Unreal Gold (Direct 3D) 1024*768 32bit: 17.6 FPS

Unreal Gold (Direct 3D) 1280*768 32bit: 13.1 FPS

Overclocking

The stock speed of this card is 125 MHz core, 143 MHz memory.
The overclocked speed is 175 MHz core, 185 MHz memory.

Overclocked 3dmark 2001 result: 1359 3D marks.

Overclocked Quake III result: 35.6 FPS (800*600)

Would be good to add here results with AA and AF, also higher resolution 1600x1200.
And to make a table with all results.

Tertz wrote:

Would be good to add here results with AA and AF, also higher resolution 1600x1200.
And to make a table with all results.

This project is resting for now, if I continue it will only be to bench some more cards, not adding resolutions and AA.

The resolution cap was because the TFT I used only would do 1280*1024 max and I did not feel like carrying my 19" Samsung SyncMaster 959NF back and forth between my retro gameing setup and the benching setup.

Skyscraper wrote:

This project is resting for now, if I continue it will only be to bench some more cards, not adding resolutions and AA.

You may make new testing. 😀 For example P4 3 GHz for Win9x gaming seems better, while P3 for Win+DOS rig. Or, at least, expand the testing for new additions.
AA is important as it's closer to practical use. I doubt someone needs fps higher than 60, so he will set AA+AF in case of say 150 fps gotten without postprocessing. Also you've reached CPU bottleneck and using AA allows to overcome this.
As for resolution. The quantity of people wich prefer LCD for old games rises with time. So the importance of resolutions higher than were used on CRTs rises too.

Some testing is redundant: I'd keep only one 3dmark, instead of 3. 640x480 is not interesting beginning with TNT2 era, 1280x960 is exotic. You may keep only 1024x768x32, while 800x600 is useful only for slow cards so 16 bit mode is more appropriate. For overclocking some game should be shown too, for example Unreal in 1024x768x32.

The reason for choosing this motherboard is that is has Universal AGP so I can test all cards from the Voodoo Banshee to the Geforce 6800 Ultra using the same system and benchmarks. I have not got to the really slow ones yet though. The main reason this project halted is that this motherboard did not like some of my "almost working" AGP cards which work in some less picky boards, I have more cards to choose from now though.

Skyscraper wrote:

The reason for choosing this motherboard is that is has Universal AGP so I can test all cards from the Voodoo Banshee to the Geforce 6800 Ultra using the same system and benchmarks.

There would no be significant loss without pre-TNT2 cards as not many ones use them for 3D, besides Glide wich is a different story. Also, until you'll find sometimes P4 MB with UAGP support, you may use results for those early cards with Athlon as those cards may to be bottleneck and faster CPU may not change much the situation.
And you may use P3 1 GHz (or Tualatin 1.4 GHz) system with AGP 2x slot for cards up to GF5 (GF6 for some brands). It is not obligately that 4x and 8x modes would give significant boost in your tests, anyway.
Those systems are more common and hence practically more interesting than Athlon 2400.

The main reason this project halted is that this motherboard did not like some of my "almost working" AGP cards which work in some less picky boards

I suspect MB on Intel chipsets are more compatible and this may to be another reason to try them.

We will se what happens later when I have got things better organized.

As it is now my apartment is a mess so all projects except for testing some new stuff are on hold until the mess is stored away. The 100 large ESD bags I ordered from the UK for storing motherboards are already one week overdue, thanks Royal Fail!

Well I might find time and space for rebenching some systems with the correct (faster) Doom 3 version for the Doom 3 project but that is not very time consuming and I still have one non clogged work bench.