Nice comparison.

Usually instead of speed compatibility with different old games is more important. There is a thread about that somewhere.

I believe if you are using something like Geforce 6800 and Athlon then you'll be using it Windows games too. So it's important to make somekind of a compromise between DOS and WIN performance/compatibility or switch graphics cards as needed (at least I have a lot more graphics cards then motherboards&CPUs).

I don't have any reliable way of testing SVGA modes but I do believe anything that does exceptionally well in the VGA side would also do significantly well on SVGA modes as well. So likely the cards at the top of my list would probably be the go-to choice for those interested in SVGA game modes, which also require a decent CPU too.

Besides that, if one wanted to for example go with a Pentium-100 system for compatibility with older dos games, then they might be interested in what I found as the fastest PCI card.

I have another box with another 20 or so video cards somewhere, I just have to find it, I might look for it and expand this list later today.

I ment more like Windows 98/XP 3D game performance as AthlonXP and Geforce 6800 is quite good match. But not as good in pure DOS.

Actually I've found AthlonXP chips to be fantastic for MS-DOS SVGA modes for gaming, at least in descent II, Duke3D, and Quake, if they can be paired with a decent video card to give good dos VGA/SVGA performance. And My A-XP systems are the only ones I can find that can manage above 60 FPS in walk-around mode in 800x600 in Tex Murphy, at least it did with my voodoo3.

EDIT: Turns out I was a little wrong, my Athlon-XP thing is doing 1280x1024 in Descent2 just smoothly and even shows some tearing when spinning around (above 60 FPS), so apparently I was a tad wrong and what I have is good enough, at least for Descent2. It's been almost a year since I played with it I forgot how good it was. 😀

kithylin wrote:

I don't have any reliable way of testing SVGA modes but I do believe anything that does exceptionally well in the VGA side would also do significantly well on SVGA modes as well. So likely the cards at the top of my list would probably be the go-to choice for those interested in SVGA game modes, which also require a decent CPU too.

I think this assumption is incorrect. There are several problems:
Basic VGA uses banked I/O area between 640 kB and 1 MB.
SVGA is no standard but just a name for capabilities of the card for modes that go beyond 640x480 at 16 colors, e.g. as implemented in FractInt for various cards.
What you probably mean is the VESA standard, where you have to make a difference between VESA 1.x with banked SVGA modes and VESA 2+ with cards that support Linear Frame Buffer (of course in combination with USWC BIOS setting/FASTVID).

Speed wise you can not deduce from banked VGA accesses the speed for Linear Frame Buffer accesses. The data already present from benches here in the forum shows this, on some cards it even depends on the mode set. I think the main reason is that you don't know if for the specific card a memory layout translation for the mapping of a certain mode as seen in the frame buffer from the CPU to the graphics cards memory takes place. This can strongly slow down the effective transfer speed.

The second point to consider is the PCI speed vs. CPU speed. Ideally you have a CPU that is so fast that it can reach the maximum PCI transfer speed, if the graphics card is sufficiently fast on the back end. A faster CPU would not give you any gain. However if you have a board that has a lower maximum PCI speed actually a faster CPU system can perform slower. Athlon XP boards are not so well known to have ideal PCI transfer speeds. So with a certain CPU speed al that matters is just PCI transfer rate, so it might be that as example a 440BX with a low/mid range P3 performs even faster than an Athlon XP on a VIA based board (e.g. KT133A).

The third point to consider is bus latency. If you don't have simple fullscreen framebuffer copy loops, but more random accesses the impact of latency increases. So PCIE (66 MHz PCI for workstations) may be faster than PCI than AGP etc.

Before you start benching I would also suggest to review the performance data already collected here in the forum.

I haven't found any performance data for standard VGA 320x240 for video cards around here, and I did search.

All I meant was I do not know of a full benchmark suite for dos like the one Phil had set up that utilizes 800x600 mode (of any type), for anything, only this VGA mode I showed in the original test. I wasn't concerned with VESA/SVGA, whatever it is none of the tests I have use more than 320x240 anyway.

kithylin wrote:

I haven't found any performance data for standard VGA 320x240 for video cards around here, and I did search.

Well, probably because 320x240 is no standard VGA mode.
Standard VGA graphics modes are:
320x200 16 colors
640x350 16
640x480 2
640x480 16
320x200 256
all colors are palettized from a 6:6:6 palette.

Everything else is some custom mode that may not work on every VGA card.

.. I meant the basic VGA modes, what ever it is in the games that run, doom, quake, etc, in this test suite, 320x200, or whatever, I never really paid attention I just ran the tests to figure out which video cards are fastest. I don't know that it really matters either, VGA is VGA, anything below 640x480 pretty much.

Hi,
i have 2 questions:
1) Was FASTVID or MTRR used for this measument?
2) Were these data imported to Phils Mega sheet?

so strange
i got 2 3dbench and 3dbench2

with one i get 25fps on a geforce4 PCI
and with 3dbench2 i ger around 120fps

all with a pentium3 1000mhz.