Is that Maximum resolution of that 19" CRT or Recommended?

IIRC Quake II at 1600x1200 on my Voodoo 3 3000 AGP @ 190MHz wasn't all that fast. My voodoo 5 5500 AGP does kick some butt at that resolution.

I agree with swaaye. Love PQ of the V3~V5 (V3 at 16-Bit color though)

I am using a 430TX board with a K6-III+ at 450 MHz and 256 MB of RAM. Of the cards below, which would be the best choice for this system? It is a quad-boot system with NT4, Win98SE, W2K, and XP Pro. While the display driver for the GeForce 6200 functions, I could not find a driver revision which functions with D3D. I suspect the supporting hardware is holding it back.

I have a Rage 128 Pro II 4XL from China, which functions well enough, but the display quality at 1280x1024 made me pull the card out. The ATI Rage 128 VR card, on the other hand, has good display quality at 1280x1024.

ATI Rage 128 VR (32MB)
NVIDIA RIVA TNT (Diamond V550 or Creative) (16 MB)
NVIDIA TNT2 M64 (16 MB)
Matrox Millennium G450 (32 MB)
3DLabs Oxygen VX1 (32 MB)

I was going to put the G450 in the system but decided not to due to swaaye's comment

swaaye wrote:

G450 and Kyro are not really ideal for more than GUI use.

Regarding the GF6200: That's too new, older Geforce cards should run faster in most cases. I wouldn't go later than GF2, may GF4(MX) - with that CPU there is not much point in having pixel shaders.

I suppose you already checked driver support for all cards?
You did read http://www.vogonswiki.com/index.php/3Dlabs#Pe … _3_and_GLINT_R3? "Gaming performance-wise, the Permedia 3 and GLINT R3 are both slower than even NVIDIA's RIVA TNT and ATI's Rage 128 GL (both released several months earlier) in most situations, and they are definitely no match for the RIVA TNT2 or 3dfx's Voodoo3."
The TNT and the M64 should trade blows (M64 has more fillrate but less bandwidth), the M64 might be preferable since it should run cooler.
The G450 is generally faster than an M64, mainly because it uses DDR RAM which somewhat compensates its 64bit RAM interface. It's possible that it'll run slower at lower resolutions with a K6-3+, though.
A Kyro is at least as fast as a TNT2, if not much faster depending on the game, but you didn't mention that as an option.

If I disregard the cards in your list I'd say the best PCI VGA card is either a Voodoo3 or a GF2(MX), depending on the game. Both should work with a K6-3+ if you use an older driver for the GF2.

idspispopd wrote:

Regarding the GF6200: That's too new, older Geforce cards should run faster in most cases.

FX5600 is faster GF6200 and has better backwards compatibility with older games. Ones you want to avoid are the vanilla FX5200, FX5600 XT, and FX5700LE. Those cards are all pretty much useless when it comes to playing DX9 games. FX5500 is only slightly better, so probably shouldn't be on your shopping list if you plan to play the occasional DX9 game. For DX8 and older they would all probably be OK.

Geforce2 MX 400 would be a wise choice.

sliderider wrote:

FX5600 is faster GF6200 and has better backwards compatibility with older games. Ones you want to avoid are the vanilla FX5200, FX5600 XT, and FX5700LE. Those cards are all pretty much useless when it comes to playing DX9 games. FX5500 is only slightly better, so probably shouldn't be on your shopping list if you plan to play the occasional DX9 game. For DX8 and older they would all probably be OK.

I don't think you want to play any DX9 or even DX8 games on a K6-III+ ... or do you have a good example?

What about one of these cheap eBay special buy Radeon 7500 cards? If the signal quality is acceptable at 1280x1024, is a 64-bit Radeon 7500 better than the graphic card options I have presented for a K6-III-450/500?
$15.99 + $2.99 shipping. http://www.ebay.ca/itm/231431907247

Unfortunately, trying to source another PCI GF2 MX400 or Voodoo3-3000 would be too expensive. Although heavily under utilised, those cards will remain in the 486's.

I'm going to throw the Savage 4 PCI out there - good DOS compatibility and OKish DX speed.
I have one in my K6-3+/TX machine alongside a Voodoo2. It has excellent signal quality as it's highly integrated.
I would just get the cheapest Savage4 PCI, don't worry about memory size or speed.

How does DOS compatibility of the Radeon 7500 compare/contrast with that of the Rage 128 ?

Did you already check http://gona.mactar.hu/DOS_TESTS/ ?
It seems that Radeon DDR (= 7200) is slightly worse than Rage 128, I'd assume that this didn't change with Radeon 7500 which is mainly a die-shrink.

I did check that, but didn't know which card "ATI Radeon DDR" was in reference to. Before I look into spending money, I have decided to benchmark the 5 graphic cards noted above with a K6-III-500 (430tx). I am running Final Reality (DX5), 3DMark99Max (DX6), 3DMark2000 (DX7), GLQuake, Quake II (w/AMD patch), and Quake III timedemos. I am testing low resolution 800x600x16 and high resolutions 1152x864 (or 1280x1024). I tried Quake II with and without the AMD K6 patch, howver the frame rates were identical. DOS compatibility tests to come later. Based on reading numerous forums, it seems that the GeForce2 (non-mx200), GeForce 256, TNT2 (non-Ultra & non-M64), or Voodoo3 are good cards for a compromise of fast Windows-3D with DOS compatibility.

EDIT: I was hoping for 30 fps in Quake III at 1280x1024x32, but the Rage 128 VR did not come close. 8.0 fps. For some reason, I can only run Quake III timedemo four within NT4, but not Win98.

Regarding "Radeon DDR": The first Radeons were plainly called Radeon, like the first Geforce there was an SDR and a DDR model. About the same time the next generation Radeon 7500 was released a new model with 64 MB SDR-SDRAM called Radeon 7200 based on the old chip was also sold, so later all cards with that chip were called 7200.

Agree on the "good cards" list. Pity that ebay Canada doesn't have more old cards.

Regarding your Q3 bench: That's about what I would expect. Rage 128 VR only has a 64 bit memory interface.
Here a comparison benchmark: http://thandor.net/benchmarks.php/vga/vga?a=3 … &s=Submit&h=116
Rage 128 Pro scores 16fps under the same conditions. Assuming it's totally bottlenecked by VRAM bandwidth 8fps make sense.
For 30fps you have to go newer than TNT2/Rage 128.
Still, 30fps should be possible with that CPU (as you probably already checked at lower resolutions), see http://www.tomshardware.com/reviews/benchmark … hon,590-23.html

On the same P3 1400 system noted in the thandor.net benchmarks, they were able to obtain 46.0 fps on a Radeon 7500, which is nearly tripple the score of their Rage 128 Pro. I did not realise the jump in consecutive generation technology was so vast.

idspispopd wrote:

sliderider wrote:

FX5600 is faster GF6200 and has better backwards compatibility with older games. Ones you want to avoid are the vanilla FX5200, FX5600 XT, and FX5700LE. Those cards are all pretty much useless when it comes to playing DX9 games. FX5500 is only slightly better, so probably shouldn't be on your shopping list if you plan to play the occasional DX9 game. For DX8 and older they would all probably be OK.

I don't think you want to play any DX9 or even DX8 games on a K6-III+ ... or do you have a good example?

NASCAR Racing 2003 has a P-II 450 as the minimum processor, so a K6-2+ or III+ overclocked to 600mhz can probably handle it as it has similar FPU and better integer performance than a P-II 450.

Has anyone tested frame rates with graphic cards which are heavily CPU-bound on a K6-III-500 and compared them with older generation cards which are also CPU-bound?

For example, at 1280x1024x32 in Quake 2/3, how much increase in frame rate, if any, does a Radeon 9250 or GeForce4 MX440SE have over a GeForce2 MX400 or GeForce 256 on a K6-III-500? What about even more modern graphic cards? My K6-III system only has PCI, so I am limited to what is readily available in that form factor.

After extensively testing an ATI Rage 128 VR, NVIDIA RIVA TNT, NVIDIA TNT2 M64, Matrox Millennium G450, and 3DLabs Oxygen VX1, I was left with the feeling that the results could be much better. As such, I am looking for PCI alternatives.

I don't think there will be any gain in fps, the cards are simply waiting for the cpu.

I'm in the process of testing a few cards on a p3-450 (at 300mhz and 450mhz). Up to now i have tested tnt, tnt2 m64, geforce2 mx400, geforce 2Ti, mx440 and geforce4 ti4200 128mb.

I only used 3dmark 99 max. At 800x600 ALL cards are about equal with the cpu at 300mhz. They ofcourse score higher at 450mhz but again ALL are equal at 800x600 (cpu bound).

At 1024x768 the tnt and tnt2 m64 get a hit, but from geforce2 mx and higher ALL scores are equal at 1024x768. (16bit, so i can compare 3dfx)

Only differences are: AA and resolution

2Ti gets free 4xAA at 1024x768 (same score as without AA)
4Ti gets free 4xAA , 8x AF and 1280x1024 (same score as 1024x768 without AA)

As the p3-450 is more powerfull than any k6 cpu i think the scores will be lower but the conclusion the same. I will start testing 3dfx cards and see how they compare. I already did v2 sli and it is king on this setup in 800x600, getting a 10% higher score than any of the Nvidia cards. At 1024 it is on par. I will test v1, banshee, v3, v4, v5 and a matrox g400 also.

Test setup: Asus p3b-f, 128mb pc100 cl2, p3-450, win98se, forceware 40.72

When i'm done i will repeat this test on an Ali socket7 board with a k6-2 450 or maybe a k6-3 450.

If you tell me what quake to run (and where to get it as i don't play quake) i can test it on my other system for you.
I have a k6-3+@ 600mhz up and running with a v2-sli and geforce2 gts setup. (win98se)

I have a test matrix started and am testing 3DMark2000 for Direct3D and GLQuake/Quake II/III for OpenGL. For low res, I am testing at 800x600x16, for hi-res, I am testing 1152x864x32 (max. for most 16 MB cards) for all cards and 1280x1024x32 for cards with 32 MB+. It will be interesting to compare results. I dropped 3DMark99Max (DX6) because some cards gave out errors (TNT2 M64 and G450).

It takes a long time for this mini-project because I am also testing all possible driver revisions to determine which driver yields the best performance. I also tested Final Reality (DX5) but dropped it because of errors with Banshee and G450. I have completed 7 cards now but I am looking to increase this number if I can find appropriate pricing.

It would be interesting to compare your results to mine. I have left all Quake settings at default. I am running 3DMark2000 with and without D3D T&L accel. for cards which support it. I am running Quake 1 timedemo 1, Quake 2 timedemo 1, and Quake 3 "demo four". For 3DMark2000, I am using the 3DNow! optimisation. For Quake 2, I am using the 3DNow! optimisation, however it has very little effect. For Quake 3, I am using NT4 because I could not figure out how to get the timedemo running with Win98SE. All other tests are done with Win98SE.

The most shocking results thus far were how studding the Matrox G450 results were at GLQuake and how crummy they were in Quake 2. I suspect some cheating going on with Matrox's drivers for the timedemo.

I have tested these cards thus far,

ATI Rage 128 VR 32MB
3Dlabs Oxygen VX1 32MB
NVIDIA TNT2 M64 16 MB
NVIDIA TNT 16 MB
Matrox G450 32 MB
PNY GeForce2 MX400 64 MB
3dfx Voodoo Banshee 16 MB

but am considering adding another 4 cards to the mix. I'll post full results and bar charts when I have completed this task.

I think an FX 5500 might be the top of the line for available PCI cards. I was considering one of these from China, http://www.ebay.ca/itm/221373503126 , but I am conserned about the PCB and signal quality.

Same consideration with the Radeon 7500, http://www.ebay.ca/itm/231431907247
and perhaps Radeon 9250. The problem is cost and quality. The Americans seem to have everything as usual, but shipping anything out of the USA has become ridiculously expensive. Shipping out of China is still cheap, but quality....

I was able to source a GeForce4 MX440SE PCI for $12 shipped, so I will test that as well.

During my testing thus far, I have determined that to run a 430TX well at 83 MHz, do not use SDRAM. The timings are too fast for the chipset to be long-term stable, even on the slowest setting. Switching to 256 MB of EDO RAM on the fastest settings and removing the cache TAG DIP SRAM worked wonders.

I ran a comparison of various PCI graphics cards on an i430TX system utilising an AMD K6-III+ 500 MHz (6x83). I was trying to determine the fastest overall graphics adapter for the i430TX platform. Attached are the results sorted in ascending order by the higher resolution (1152x864x32). I set 800x600x16 as the "low resolution" test and 1152x864x32 as the "high resolution" test. The reason for setting 1152x864x32 as the upper resolution was because this the largest resolution the 16 MB graphics cards would accept at 32-bits.

3DMark2000
The FX5500 came out on top for the high resolution tests, however if T&L was disabled, the GF2 MX400 beat the FX5500 by 5%. It was interesting to see the Voodoo Banshee surpasing the Matrox G450 in this benchmark. 3DMark2000 is directX-based and it was my understanding that DirectX was a Matrox's strong point, so to see the Banshee surpass the G450 by ~9% was surprising. Also noteworthy was the NVIDIA TNT beating the FX5500 in the low resolution test when T&L was disabled.

Final Reality
Perhaps someone could explain the poor performance of the FX5500 in the Final Reality benchmark? Does the FX5500 not handle DX5-based games well? Are newer graphics cards generally poor at older DirectX functions?

GLQuake
The FX5500 and GF2MX400 came out on top and were comparable. Interesting was that the Matrox G450 surpassed the GeForce cards by more than 10 fps in the low resolution tests. Also surprising was the landslide victory of the G450 over the Banshee. I suspect there is some Matrox driver optimisation for GLQuake which caused this large differential, especially considering how the Banshee beat the G450 in DirectX tests. The Banshee and ATI Rage 128 VR were comparable, while the TNT was close behind. Note that the GF2/4/FX driver did not disable VSYNC for this test and that I did not use a 3rd party app. to disable it. Thus the 800x600 results for these cards are peaked at around 60 fps.

Quake II
The stellar performance witnessed by the Matrox G450 in GLQuake was certainly not carried over to Quake II. How did the G450 end up at the bottom in Quake II? I was considering using the G450 in my K6-III build until I saw these results. The Banshee, Rage 128, and TNT all had acceptable low resolution scores but really fell short in the high resolution tests. This is where the GeForce cards showed their strength.

Quake III
Only the FX5500 broke the 30 fps threshold. It was the extra 3.6 fps of the FX5500 card compared to the GF2 which made up my mind to use the FX5500 in this system. Had the GF2 had better Quake III results, I'd have probably settled on the GF2 for my K6-III rig. It is interesting how the original RIVA TNT clobbered the Rage 128 in the low resolution test, however both cards exhibited the same high resolution result.

Final thoughts
Would a GF4MX440 show any improvement over a FX5500 or GF2MX400 in Quake II/III when using a K6-III? What are the disadvantages of the FX5500 compared to a GF2 or GF4? In general, the Oxygen VX1 was unremarkable and I see little gaming use for it. The Rage 128 VR is a much better overall substitute. While the G450 has really nice GLQuake and D3D capabilities, the Quake 2 test renders it useless. Perhaps there is some driver optimisation I am missing out on?

System specifications
Biostar MB-8500TTD, AMD K6-III+ 500 MHz, 256 MB EDO RAM, L3 cache disabled
All tests were performed using Win98SE with the unofficial service pack version 2.1. The exception is for Quake III, which used NT4.0.

Setup notes
3DNmark2000 used the 3DNow! optimisation. Oxygen driver optimisation was set for "Quake". This optimisation, for example, increased Quake II's result from 28.9 to 31.7 fps. Quake II was using the 3DNow! patch. I was unable to locate a functioning ATI Radeon 9250 driver version. My computer would not turn on with the ATI Radeon 9000 installed. Perhaps the i430tx or this particular motherboard are incompatible with the Radeon 9000 series of cards.

The driver versions which yielded the best scores were used. They are as follows:

ATI Rage 128 VR 32 MB
4.13.7078 (w98)
4.3.192 (NT4)

3Dlabs Oxygen VX1 32 MB
2107-0829R (w98)
2.16-0866 (NT4)

NVIDIA TNT2 M64 16 MB
53.04 (w98)
29.42 (NT4)

NVIDIA TNT 16 MB
2.08 (w98)
23.11 (NT4)

Matrox G450 MB
6.82 (w98)
5.06 (NT4)

NVIDIA GeForce2 MX400 64 MB
53.04 (w98)
29.42 (NT4)

3dfx Voodoo Banshee 16 MB
1.04.00 w/miniGL 1.46 (w98)
1.04.00 w/WickedGL 2.31

NVIDIA GeForce FX5500 256 MB
56.64 (w98)
56.64 (NT4)

NVIDIA GeForce 4 MX440 64 MB
56.64 (w98)
56.64 (NT4)

EDIT: 7 Nov. 2015 - Graphics have been updated to include data from a GeForce 4 MX440 with AGP8x. I'm not sure how the "with AGP8x" is involved; this is a PCI card. Overall, the GeForce4 MX440 was the best overall for the graphics cards I tested. In benchmarks where the FX5500 had strength over the GF2 MX400, and the GF2 over the FX5500, the GF4 combined these two strengths.

Lastly, I compared Quake scores for the FX5500 on 4 different Windows operating systems. Driver version 56.64 was used for each operating system. NT4 generally demonstrated good performance, while XP was slower. Interesting to note that Win98SE had the worst Quake II results, but the best GLQuake results. NT4 came out on top for 2 of the 3 tests.