Yes, apparently some PCI cards can have really terrible DOS performance. Like the Matrox MGA Impression IS-Athena stuff.

Yes, I remember testing this VLB card with dual processors. One for Windows, one for DOS. Diamond Viper P9000 I believe. Another one was a Trident VLB. It was just a tiny bit ahead of the top ISA cards.

Actually the P9000 is not VGA compatible and used in Windows only. Typically for VGA there must be a separate VGA chip on the card. The solution from Weitek is the Weitek 5186, which is really low and and covers only the most basic functions.
e.g. like on the Genoa VideoBlitz: http://retronn.de/imports/hwgal/hw_graphics_c … 00vl_front.html
Some card have other VGA chips. These Windows accelerator cards are not targeted for VGA/DOS (same is valid for the Matroc Impression Lite cards).
The 5186 was later included in the P9100.

Rendition Verite cards, both V1x00 and V2x00, have comically abysmal DOS VGA and Mode X performance, about 386DX levels, although they are very fast on VESA modes.

Check mine benchmark part od site - there are some, but not many. Usually because of bugs (Rendition) or aiming purely at Windows performance (Matrox), or they just use for dos separate lowcost chips....
http://vgamuseum.info/index.php/benchmarks/diag-graphic
http://vgamuseum.info/index.php/benchmarks/quake-320

flash.... really? T.T

Number Nine Imagine 128 also needed a separate VGA chip.

ElementalChaos wrote:

Rendition Verite cards, both V1x00 and V2x00, have comically abysmal DOS VGA and Mode X performance, about 386DX levels, although they are very fast on VESA modes.

Renutil fixes everything except Mode X. Of course Mode X is rather important. Oh well. 😀

as some of you already mentioned, some pci cards simply neglected dos performance by design, and would only show their power in windows.
so, to compare their performance and find which is slower, better compare benchmark results in windows.

for dos performance, most nvidia cards prior to geforce4 are very fast, although i believe the geforce2ultra is a bit faster than anyone else.

swaaye wrote:

Renutil fixes everything except Mode X. Of course Mode X is rather important. Oh well. 😀

I was not able to get Renutil working properly on my V2200 for whatever reason. I can only get it to show a messed up picture, cloned on a 2x2 grid, stretched horizontally and shifted to the top right of my monitor. It doesn't help that there's zero remaining documentation for the program.

vlask wrote:

Check mine benchmark part od site - there are some, but not many. Usually because of bugs (Rendition) or aiming purely at Windows performance (Matrox), or they just use for dos separate lowcost chips....
http://vgamuseum.info/index.php/benchmarks/diag-graphic
http://vgamuseum.info/index.php/benchmarks/quake-320

I'm looking at your graph and am noticing that the Tseng ET400AX ISA receives the same score as a Trident 8900CL. Why is that?

feipoa wrote:

vlask wrote:

Check mine benchmark part od site - there are some, but not many. Usually because of bugs (Rendition) or aiming purely at Windows performance (Matrox), or they just use for dos separate lowcost chips....
http://vgamuseum.info/index.php/benchmarks/diag-graphic
http://vgamuseum.info/index.php/benchmarks/quake-320

I'm looking at your graph and am noticing that the Tseng ET400AX ISA receives the same score as a Trident 8900CL. Why is that?

There's some more weird stuff there... The CL-GD5460 is the fastest 2D dos card I've ever tested, (it eats S3 cards for breakfast) but it's really low in that chart. The trident Blade 3D has excellent DOS performance, but again it's really low in the chart. Odd.

feipoa wrote:

I'm looking at your graph and am noticing that the Tseng ET400AX ISA receives the same score as a Trident 8900CL. Why is that?

Well first there are different Revisions of the ET4000AX with different performance. Also the Trident 8900Cx (as well as the 8900D) belong to the faster Trident cards.
You should also consider the tested modes. Cards vary a lot if you look at regular 320x200, ModeX or SuperVGA Modes either banked or Linear Frame Buffer (for local bus cards).

The main point here is the difference between planar modes and non-planar modes (also called unchained / chained). Now if you have the regular 320x200 a DWORD write to graphics memory with STOSD writes 4 bytes. While these are consecutive in main memory they are each 4 bytes apart in graphics memory. So you transfer from
01 02 03 04 to 01 xx xx xx 02 xx xx xx 03 xx xx xx 04 xx xx xx. Considering just performance, it happens that on the ISA bus it is split to 16 bit transfers for a 16 bit card, so e.g. 01 02 is written and the graphics card needs now two memory writes, even if it has 32 bit memory bus width.
In planar mode a 32 bit wide write would be split into two 16 bit transfers over the ISA bus and could be combined by the graphics card to a single 32 bit write to graphics memory.

You see it really depends on the mode and on the buffering the graphics chip does. The performance really depends on the mode you look at. There won't be a single best card that handles every situation at top speed. You have a per mode ranking.

kanecvr wrote:

There's some more weird stuff there...

Indeed. Like a Voodoo3 2000 scoring more in Quake than a 3000 and that 3000 scoring more than a 3500. 3 cards that their only difference is memory speed are ranked in the opposite-than-expected order. All AGP. That's some bad ass V3 2000 🤣

I just got the urge to bench the oddity that I bought this weekend the Macronix mx86200. Seems an extremely nice/compatible card for old DOS games.

mrau wrote:

flash.... really? T.T

Sadly haven't found better plugin for showing charts in joomla. You can always use picture only version - http://vgamuseum.info/index.php/benchmarks

kanecvr wrote:

There's some more weird stuff there... The CL-GD5460 is the fastest 2D dos card I've ever tested, (it eats S3 cards for breakfast) but it's really low in that chart.

I can't see there anywhere any CL-GD5460 - Odd. 🤣

kanecvr wrote:

The trident Blade 3D has excellent DOS performance, but again it's really low in the chart. Odd.

thats because its PCI version and probably also underclocked compared to official cards. I've tested also AGP version and its 96,9FPS vs 140,2FPS in quake 320x240 (agp version is not in the graphs - long time since i updated them). So scores depends also on bus and card maker - some might be underclocked, have not best available memory configurations, worse memory timing....might not like used Via KT133A chipset.... many posibilities....

vlask wrote:

Sadly haven't found better plugin for showing charts in joomla. You can always use picture only version - http://vgamuseum.info/index.php/benchmarks

thanks, the blonde obviously couldnt find that -.-

i wonder why that b69k card is so fast in pure dos, but is relatively slow with fastvid; any ideas? also its pretty obscure id say?

as for slowing down i guess any of these cards with a dos based overclocking program will do, just underclock until you see every pixel drawn :>