@Administrators: Please move to Forum 63 (Marvin\Video).

@Nemo1985: May you add the amount of memory on your cards please?

What’s the un-colored (white) boxes represent? I’ll have to take a closer look at Permedia2.

BinaryDemon wrote on 2020-02-02, 00:28:

What’s the un-colored (white) boxes represent? I’ll have to take a closer look at Permedia2.

Uncolored because I'd have to use too many colors
Green is for the best performance, yellow second, orange third and red the worst.
The last 2 3dfx cards have white color because I wanted to try the Nvidia GeForceFX 5500 pci too but it wasn't boot on ss7 machine so I took it off from the comparison.
I also did the quake 640 benchmark (with vbe 2.0 cards only since it didn't work with 1.2) but the results were cpu limited so I took that off too.

Nemo1985 wrote on 2020-02-02, 00:18:

Every card has been tested with the maximum memory installed.

Nice ! Thank you. When you say Maximum memory did the Matrox cards have 8Mb or more ? I have some M2 with 8Mb but never tried Chris 3d test on them, odd to see good results on one test but black screen results other test. Maybe the card/matrox driver doesn't like the way the higher end test initiated.
Other thing: never thought a CL card would perform as good as your results on the 5446. Very interesting !

Horun wrote on 2020-02-02, 03:08:

Nice ! Thank you. When you say Maximum memory did the Matrox cards have 8Mb or more ? I have some M2 with 8Mb but never tried Chris 3d test on them, odd to see good results on one test but black screen results other test. Maybe the card/matrox driver doesn't like the way the higher end test initiated.
Other thing: never thought a CL card would perform as good as your results on the 5446. Very interesting !

Thank you for showing some interest 😀

I uploaded the version with mb memory for every card, I should say that for maximum amount I mainly refer to older cards with SOJ package.

About the matrox cards, I would be curious to check the other bios release or other release for the m220 or millennium 2 but I can't find them, they have a great quality output and if downgrade the bios (I found how to do it since I was able to downgrade from 1.8 to 1.2) improve the dos compatibility it would be the jack of all trades, believe me when I say that some of these cards have a very low quality, it can be seen on Chris 3D 320.

I agree with you the CL-GD5446, has good results, too bad I don't have the GD5440, it was the chip after the 5436, I found it for 20 eur, quite too much for my tastes. Another card I would be curious to test is the Ark2000mt, to my knowledge it could gives very nice results.

I did those test to decide which videocard to use on my amd 5x86, I will probably re-run the tests on that older machine but preliminary results shows it is mainly cpu limited.

Great job, man! Very interesting results, but mostly, a great round-up of cards. I've been trying to get my hands on an ET4000 and ET6000 for years but they're always so flippin expensive. The other really interesting number I saw was for the Avance Logic ALG2302.A. It is so much faster than the ALG2301.B. In my tests, the 2301.B is just a little bit faster than the GD5430, but your ALG2302.A is in the top third of your results. Any idea what brand and model the card with that chip is? Mine is a Hercules Stingray.

Some other observations on your results:
-I'm surprised at the relatively big difference between the Banshee and V3 2000 in certain benchmarks
-most of the cards I have line up with your results.
-you need to get a TNT2 M64 PCI. They are relatively cheap and easy to find and are always at or near the top of such comparisons.
-I wish you had the Ark Logic 1000/2000PV. I'd love to see how they compare to the ET6000 and Banshee. See, I have the Ark1000/2000 but not the ET4000/6000. We should trade for benchmarking purposes, then trade back when done. 😀
-it's interesting to see these results on a K6-3+ 550. I think it's actually a really good platform to test DOS PCI graphics cards. I would suggest you retest on a slower platform (P233MMX) to see how the results change. I have found that the faster the platform, the more exaggerated some of the differences between the cards become. Think about what the typical DOS retro PC would be built on today for gaming. Too fast becomes problematic for a lot of games, so the typical recommendation for a fast DOS gaming PC is somewhere between P200 and K6-2 550. I've been testing between P75 and P200 over the years and occasionally posting results here.
-you might want to test with Univbe as well. I have found that even at VGA 320x200, it slightly helps performance in certain benchmarks: Chris' 3D, Quake, Duke3D, and even occasionally 3dbench. So here's the scoop: For the S3 cards, there's S3VBE20.EXE, which only uses ~6kb but doesn't help as much as SDD/Univbe. Sci Tech Display Doctor 5.3a is what I use. It has a smaller footprint (~13kb) than Univbe 6.7 (~17kb) and equal performance. I test with and without SDD53a. It doesn't work with newer cards that already have VBE2+, like the Millennium II, Voodoo3, and TNT2 M64. But they don't need it because they already have native support.

Thanks for sharing these results. I'm always interested in DOS graphics card benchmarks. Cheers!

Thank you.
Avance logic card is one of my favourite for 1995 year, mainly because it's an almost unknown manufacture but to my eyes it's not that bad...
This is my card: https://i.imgur.com/MSmETIe.jpg I would have said it's an oem card but according to vga museum it's a powercolor.

I never heard about the tnt2 m64 pci, if you have some link feel free to send me a pm 😀

I have two et4000\w32p cards, one is the Hercules with vb v1.2 support the other is Siemens card (http://www.vgamuseum.info/media/k2/items/cach … da792716_XL.jpg) which weird enough supports vbe 1.32 but despite of this it has lower compatibility compared to Hercules (vbe 1.2).

Well actually that was the purpose of my test, since dos platform is strictly software the cards are cpu limited already (at least at 640x480 resolution), one of the reason I removed the quake 640 from the comparison it's because most of the cards were having the same result, imho lowering the cpu speed will flatten the results even more.

Yes Univbe is a great idea, it was always adviced back in time, probably will enable the 640x480 resolution on quake and give some higher results as you suggested, I will consider that, thank you!

Nemo1985 wrote on 2020-02-02, 14:42:

Thank you. Avance logic card is one of my favourite for 1995 year, mainly because it's an almost unknown manufacture but to my e […]
Show full quote

Thank you.
Avance logic card is one of my favourite for 1995 year, mainly because it's an almost unknown manufacture but to my eyes it's not that bad...
This is my card: https://i.imgur.com/MSmETIe.jpg I would have said it's an oem card but according to vga museum it's a powercolor.

I never heard about the tnt2 m64 pci, if you have some link feel free to send me a pm 😀

I have two et4000\w32p cards, one is the Hercules with vb v1.2 support the other is Siemens card (http://www.vgamuseum.info/media/k2/items/cach … da792716_XL.jpg) which weird enough supports vbe 1.32 but despite of this it has lower compatibility compared to Hercules (vbe 1.2).

Well actually that was the purpose of my test, since dos platform is strictly software the cards are cpu limited already (at least at 640x480 resolution), one of the reason I removed the quake 640 from the comparison it's because most of the cards were having the same result, imho lowering the cpu speed will flatten the results even more.

Yes Univbe is a great idea, it was always adviced back in time, probably will enable the 640x480 resolution on quake and give some higher results as you suggested, I will consider that, thank you!

Oh, I have a predecessor of your ALG graphics card:
The ALG2301.A from 1994. (1 MB + possible another 1 MB in 8 sockets)

It was my first PCI graphics card.

Nemo1985 wrote on 2020-02-02, 14:42:

I never heard about the tnt2 m64 pci, if you have some link feel free to send me a pm 😀

They are quite common. It was a low-end nvidia card for Win9x but had surprisingly good DOS performance. Just search ebay (or wherever you shop for retro gear) for 'tnt2 m64 pci' and you'll find plenty of results. 😀

The only PCI cards in my collection not on your list are: ATI Mach64 2mb, ATI Rage XL 8mb, Viper 330 (Riva 128 4mb), and a WDC WD9710 "Paradise Pipeline PCI" think it is a 1mb but expandable. I may look for TNT2 M64 PCI, was sure I had one but might have given it away or it's an AGP variant in a differant box of stuff 🙁

added: just found a CL-5434 2mb pci and Trident 9440 both in low end pentium boxs.

Trident 9440 series is horrible. It's more or less fine as VLB card for a 486 system (CPU is too weak anyway in most cases), but PCI version does not scale gracefully.

I did the same tests with the Amd 5x86 133 paired with Biostar MB-8433UUD-A, 256kb cache and 32mb FPM memory, settings for cache an memory are the fastest possible.

Results are just silly, the videocards are heavily cpu limited.
For this reason I also tried the ET6000 with overclocked cpu and bus (40x4) 160 mhz, the improvment is huge.
In the end with such pci vga cards doesn't really matter which one you are going to use.

Why don’t add a column to your spread sheet for total score of all the benchmarks to which card has the best overall score.

Intel486dx33 wrote on 2020-02-07, 11:25:

Why don’t add a column to your spread sheet for total score of all the benchmarks to which card has the best overall score.

Thank you for the advice.
Honestly I do not think it would be useful, I just compared the overall results of the Millennium 2 and Trident card, the first get an overall higher total results but despite of that in 1 test it gives black screen, that means it is less compatible than trident card.

Also Chris3D and 3DBench are totally synthetic and does not represent anything useful.

Shit, that SiS 6326 is an absolute dog..

I will be doing a similar test on my current Socket 7 PC at some point. I have a crap ton of PCI cards to test with it.

Yeah, I always like seeing benchmarking roundups of old hardware ^^
I plan to do some benchmarking too in the future ^^ I have a wide variety to choose from even if I'm missing certain noteworthy cards due to prohibitive costs nowadays. (I'm looking at you ET6000&co)

appiah4 wrote on 2020-02-07, 12:48:

Shit, that SiS 6326 is an absolute dog..

I will be doing a similar test on my current Socket 7 PC at some point. I have a crap ton of PCI cards to test with it.

Sis 6326 is bad with the 486 pc, but with the k6-3 wasn't that bad (not great, not terrible), still I don't get why some cards gets lower results despite being more powerful with the 486, I will probably test the cards with another 486 mb...

Hello good people!
I redid the roundup, added the fastdoom results and pc player bench 640 has been run with LFB.
PC: k6-3+ 550, aopen ax59 pro, 128mb memory, tightest timings.
I did choose this configuration because I wanted to find the video cards limit.
All the virge models have been tested at 85 mhz, they were overal stable except the GX which showed 1-2 pixel flashing on quake.
Cirrus Logic CL-GD5446 have been tested at the stock freq (80 mhz and 85 mhz).
Since from last time I collected a couple of new cards (ET 6100 and OPTI), they were not present in the older version.

All the quirks are almost the same as last time:
Sis 6202 had troubles running benchmarks at 640 with 2mb of memory, banding and screen corruption, it doesn't happen when 1mb is installed and try different chips didn't change the result.

Matrox cards, older bios are more compatible (the Mystique with bios version 1.2 gives a bit higher results and works on Chris 3D 640, while it doesn't if updated to bios version 1.8, unlucky I haven't been able to find any other bios release.), still black screen on Chris 3D 640

Trident 3DImàge 9750 has screen corruption on PC Player 640 on the higher and lower border if benchmarked WITHOUT LFB, everything is ok with LFB.

For doom I have chosen to use the tick value instead of fps, so lower values means higher performance.