Not to hijack the thread too badly, but I've got a 3d 2000 Pro 4mb card. Loved it back in the day, but there was always a question that bugged me...What is the 33-pin header on the top part of the card for?

I have a 2000 pro. I know that everyone has been praising it, but I thought that I would put in my praise as well.

I have a socket 3 Motherboard with a POD 83 in it. I love the old looking glass game Terra Nova. That is quite a strenous game for a pentium 83 (granted I played it on a 486 DX2/66 in high school; you should have seen the stutter!)

Anyways the 2000pro is that fastest vesa core of any card I have tried in that machine including my et6000. Terra Nova runs much faster than with the et6000.

Also is Terminal Velocity that great a game?

Well ViRGE is definitely not the fastest VESA card. Not even close. You need to look at later cards with 128-bit memory buses. Those become effectively infinitely fast framebuffers. I suggest NVIDIA, probably RIVA TNT or higher.

S3 cards are the compatibility benchmark though, for sure. I ran though about a dozen different VGA chipsets one day, messing with Cybermage and Terra Nova. The S3 chips were the only cards that ran Terra Nova without issues. I had to use a TSR VESA fix to get Terra Nova working right on 3dfx and NV.

3dfx and ATI cards are very fast too, but they are more problematic, especially ATI. There is actually a TSR that fixes most of 3dfx's VESA problems, however.

As for Terminal Velocity, it's a fun game. It was mostly impressive from a technical standpoint in its day. It is a pretty simple shooter.

Fastest card in dos i've tested is 3DFx Banshee AGP version, but theres many close cards like Voodoo 3, Matrox G series and all Rivas TNT2. Fastest PCI card is Voodoo 3, from oldest ones without 3D its Cirrus Logic CL-GD5480 and Tseng ET6000. If you need fast ISA card then look for Acumos AVGA2, Cirrus Logic CL-GD5428 or any WD90C3xx series.

Virge chips were depending on clocks - it varies alot. For example i have Virge/DX clocked at 50MHz and 66MHz - they have 21,38 and 27MB/s in graphic test in diag benchmark - in quake its 129,6fps vs 146,9fps in 320x240. But generaly VX version is slowest, GX2 fastest one. Anyway in 3D are usable only in preQuake games and in low resolutions like 320x240, max 360x480. And good only for filtering - still remembering playing motoracer on mine virge/dx 2Mb at amd 586 pr133 - some efects didnt worked (black smoke), but filtering was way better than software.

Unfortunately i dont have working tseng et4000 so i cant answer thread question - but for many other cards just check mine page with benchmark results - think you gonna like it 😉
quake - http://82.114.193.227/vga/view.php?cisloclanku=2007030003
Diag - not exactly same results as in quake (card in diag faster can be slower in quake), but mostly it coresponding with quake results and didnt found other bench working also at non vga cards (ega/cga/mda...)
http://82.114.193.227/vga/view.php?cisloclanku=2007030002

There are also some other benchs, but not exact as they maybe could be - glquake is ok but unreal 3d tests could be affected with drivers problems - dont have time to look which driver version is best for each card.

very cool. Thanks for the links and info, vlask.

It's interesting how at 640x480 in Quake you can see "tiers" of card performance levels. I think you really do need to run 640x480 or higher so that you introduce some potential bandwidth bottlenecks. At the lower resolutions, the scores are so close that the top cards can be considered equal IMO. The difference is small enough that it could be induced by measurement precision error. Considering that, you can also say that once you get into most graphics cards from 1998 or so, you are as fast as you are going to get for DOS speed.

What becomes more interesting than speed with such "infinitely fast" cards is whether or not the card will work with games. Basically any card outside of an ET4000 or S3 Trio64/Virge is problematic for some games. UNIVBE won't even get them working right for some games.

The S3 AGP cards may be the best choices overall for DOS speed and performance, but I don't know whether they remain as compatible as Trio64/Virge.....

I remember this thread, were it seemed quite troublesome to get a rating on Dos Vesa Video performance, because there are a lot of different configuration aspects to consider:
Most demanding DOS game, resource wise?

In Dos, I mess mostly with soundcards, because they have so much issues. Since they invented the TNT-2's to Geforce 4's I never had any trouble with the graphics department in Dos. 😀

Hi vlask, welcome to the forums!

Thanks for the benchmark results, you have an impressive VGA card collection. 😀

While the low resolution Quake timedemo results are quite interesting, the 640x480 tests don't say much about the speed of the faster cards, because they were run on an uncached framebuffer (that's why you got very similar fps for many cards).
In order to unleash the true power of these cards in the high-resolution VESA modes, you'll have to set your CPU's MTRRs to cover the VESA video memory ranges.

There are DOS utilities (i.e. FASTVID) to accomplish this, or you can boot into Windows 9x and use the "Restart in MS-DOS mode" option (Windows drivers usually enable framebuffer caching).

In that older thread there is also this measurement of mine:

PCPBench benchmark System: Soyo 440BX; Pentium-III-S 1000/105MHz; 512MB Ram, FASTVID set to enable 'linear frame buffer write co […]
Show full quote

PCPBench benchmark
System: Soyo 440BX; Pentium-III-S 1000/105MHz; 512MB Ram,
FASTVID set to enable 'linear frame buffer write combining:
Geforce 4 MX440, 64MB 128-bit DDR: 135.0 fps
Geforce 4 MX440, 64MB 64-bit DDR: 134.0 fps

So these cards were still not 'unleashed' to show their difference in memory bandwidth!?

5u3 wrote:

Hi vlask, welcome to the forums! […]
Show full quote

Hi vlask, welcome to the forums!

Thanks for the benchmark results, you have an impressive VGA card collection. 😀

While the low resolution Quake timedemo results are quite interesting, the 640x480 tests don't say much about the speed of the faster cards, because they were run on an uncached framebuffer (that's why you got very similar fps for many cards).
In order to unleash the true power of these cards in the high-resolution VESA modes, you'll have to set your CPU's MTRRs to cover the VESA video memory ranges.

There are DOS utilities (i.e. FASTVID) to accomplish this, or you can boot into Windows 9x and use the "Restart in MS-DOS mode" option (Windows drivers usually enable framebuffer caching).

Hmm didnt known about this, i tried only univbe but it wont helped at new cards, mostly dont support em (actualy using it only when i try to enable 640x480 on old cards which dont support it) - so i was thinking that i'm limited by CPU or agp bus (PCI cards all ends at 20fps).

Btw if its true, that means that mine work that i did last year is mostly useless 😲
Have to try it with some dos stuff, dont like idea installing drivers of all these cards again in windows. If i found difference, then ill make new upgraded graph, we'll see at last difference....

Thx for advice, its hard to find someone who knows what to do with these cards in dos....

Btw if you know any good card benchmark - best will be also supporting prevga cards, let me know, if you do rebenching all these cards, i can run one or 2 more tests too.

swaaye wrote:

Well ViRGE is definitely not the fastest VESA card. Not even close. You need to look at later cards with 128-bit memory buses. T.

number9 made a number of 128bit cards

My setup is very close to the system you used for the benchmarks, and my V5 AGP nailed out 227.6fps at the default setting and LFB enabled. Out of my small collection I topped out with NV6 at 238.0fps.

5u3 wrote:

In order to unleash the true power of these cards in the high-resolution VESA modes, you'll have to set your CPU's MTRRs to cover the VESA video memory ranges.

Good point, 5u3. I've occasionally dabbled in doing that with PPro and K6 systems. I don't usually bother though because the extra speed usually isn't needed for DOS gaming, and as you said, Windows drivers almost always take care of it. For benchmark purposes it might be interesting.

Don't some later system BIOSs take care of it too?

Davros wrote:

number9 made a number of 128bit cards

That's true. You don't really want to use a #9 Imagine 128 for DOS games though. Especially the I128s that use the secondary chip for VGA!

Not sure what to expect from a Revolution IV / Revolution 3D.

After i get my Stealth 3D 2000, i will pair it with my AGP Voodoo3. Perhaps will have to mess with cabling but i think it's fine.

Provided you have AGP/PCI or PCI/AGP boot sequence option in your mobo's BIOS, what do you think about this combination, a fastest AGP VESA card + a most-compatible PCI S3 Trio/Virge?

I thought this little trick has been used in the past (maybe it is still), i.e., booting with a 'backup' PCI card to facillitate flashing the 'main' gfx card.

Just got a Virge DX chipset (375) with 4MB on board...testing it...

archsan,

yup I've used the BIOS PCI/AGP vid card option to run a Voodoo3 PCI alongside a GeForce FX. Worked great. In Windows 9x the result is switching one of the cards into secondary mode.

vlask wrote:

Btw if its true, that means that mine work that i did last year is mostly useless 😲

Not really. Most people didn't know about caching the framebuffer, and some never cared about it when playing DOS games. So your scores are still valid and demonstrate how the cards performed on 95% of all machines...

vlask wrote:

Have to try it with some dos stuff, dont like idea installing drivers of all these cards again in windows. If i found difference, then ill make new upgraded graph, we'll see at last difference....

If you get around to update the bar graphs, how about overlaying the uncached and cached scores for each card, so people can easily compare? That would make a very good reference 😉

vlask wrote:

Btw if you know any good card benchmark - best will be also supporting prevga cards, let me know, if you do rebenching all these cards, i can run one or 2 more tests too.

Sorry, I don't know any good pre-VGA benchmarks and test programs, but I would be interested as well. There is a nice (S)VGA benchmark called PCPBench, it was used to define hardware classes for a German PC-games magazine. Many VOGONS regulars test their systems with it and post their scores.

swaaye wrote:

Don't some later system BIOSs take care of it too?

Most BIOSes turn on Write Combining for the original VGA range >1MB. I've never seen a BIOS which takes care of the VESA framebuffers.

archsan wrote:

Provided you have AGP/PCI or PCI/AGP boot sequence option in your mobo's BIOS, what do you think about this combination, a fastest AGP VESA card + a most-compatible PCI S3 Trio/Virge?

Yeah, this is the best compromise, since many fast AGP cards have lousy VESA BIOSes and are not supported by UniVBE. The only drawbacks are the monitor cabling and the hassle with hardware profiles under Windows.

5u3,

Does caching the framebuffer ever cause compatibility problems with games?

swaaye wrote:

5u3,
Does caching the framebuffer ever cause compatibility problems with games?

Good question. Write combining improves transfer speed over the bus by neglecting the order of read/write operations.

I reckon most games try to write out their data into screen memory as fast as possible and be done with it. But on the other hand, if a game somehow depends on the data in the framebuffer, weird things could happen.

Another issue would be games that rely on a certain video memory speed to get their timing right.

Ok so just run little test - S3 Virge/DX at 66MHz PCI 1 cycle EDO 3,5ns

Old results are for 320x240 - 360x480 - 640x480
146,9 - 72,7 - 20,2
Results with win dos are
196,2 - 72,5 - 76,5
Results with fastvid
197,4 - 72,7 - 76,6

So results with windos and fastvid are almost same, diag results hasnt changed at all. Interesting is that when i start with 320x240 resolution i have there about 141fps, then i go for higher resolutions, test em and then switch back to 320x240 - then i have 196-7 fps - thats 30 more fps. Have to try it also with other cards. Another interesting thing is that i have 4 more fps in 640x480, than in 360x480.