clueless1 wrote:

To reduce any CPU bottlenecking and see how these cards perform with more processing power, I tested all these cards in a Pentium III 933Mhz system.

Reducing the CPU and System Bus bottlenecks is not easy. They are likely to remain the major factor in DOS Vesa Mode framerates. Have you considered this?: Benchmark examples

gerwin wrote:

clueless1 wrote:

To reduce any CPU bottlenecking and see how these cards perform with more processing power, I tested all these cards in a Pentium III 933Mhz system.

Reducing the CPU and System Bus bottlenecks is not easy. They are likely to remain the major factor in DOS Vesa Mode framerates. Have you considered this?: Benchmark examples

If you mean higher FSB/PCI speeds and write combining, yeah, I have. I don't have a system with overclockable FSB/PCI. And I didn't feel write combining was appropriate since I am approaching this from a pure DOS point of view (as opposed to a Win9x/DOS mixed system where you're more likely to use write combining). From my perspective, Introducing write-combining will make any comparisons to the 100 and 200 Mhz results invalid. But it would still be neat to see those results separately, as standalone results. 😀 Might be something to add to the thread later.

OK no problem.
The compatibility notes are great info, just don't know yet how to value the framerates.
It ain't easy like Windows + Accellerated 3D graphics benchmarks.

For VGA 320x200 on anything Pentium and above, It hardly matters what card one uses. First it is the CPU that is holding things back, but when the CPU is fast enough framerate gets above 60FPS and then it doesn't matter anymore.

VESA modes: a K6-2 (stepping 8+) or Pentium II allows LFB write combining so you can get double the framerate at the cost of one line in your autoexec.bat. Like, why would anyone not want to use this free power? And it kinda shows how CPU and BUS bottlenecked all of this is. Because the videocards give double the framerate by just changing a CPU feature, not a videocard feature.

So I am looking at your result tables and wonder which differences are significant for the videocard, which are irrelevant and which are bottlenecked elsewhere...

Personally I tend to use Voodoo 3 and nVidia cards in all my retro systems (except, of course, on the 486 VLB which has a cirrus logic)

gerwin wrote:

VESA modes: a K6-2 (stepping 8+) or Pentium II allows LFB write combining so you can get double the framerate at the cost of one line in your autoexec.bat. Like, why would anyone not want to use this free power? And it kinda shows how CPU and BUS bottlenecked all of this is. Because the videocards give double the framerate by just changing a CPU feature, not a videocard feature.

So I am looking at your result tables and wonder which differences are significant for the videocard, which are irrelevant and which are bottlenecked elsewhere...

I'd say I'm targeting people with P233MMX and slower sytems, running pure DOS. Faster than that and most people are throwing Win9x into the mix. And in that case, most DOS games they'd be running can be run from the DOS prompt in Win. It's telling that the cards perform pretty consistently between the different cpu speeds (eg, the S3s finish in the same order), with the main difference being that the slower cards fall further behind the faster cards the faster the cpu. But I see where you're coming from, and this could be done from that perspective as well. Maybe with more interesting results to some people. 😀

I've always used write combining on Intel systems, but I wasn't aware of the K6-2 chips that support it. It was Elinda, AKA The Oracle, that showed me the ways 😊

The AMD Athlon also supports write combining. And then you can also get a boost with graphics card specific tools, for example that S3 VESA 2.0 tool boosts performance quite substantially.

But I really appreciate the work of clueless1 as they all are tested in the same environment. For future project ideas, it would be interesting if the ranking of these cards remain in a Slot 1 machine with write combining combined, or if some cards "write combine better" if you know what I mean.

I had a dead Number Nine Virge 325, so I pulled the BIOS chip and put it in my working (Fastware) Virge just for fun. Turns out it's slightly faster with the Number Nine BIOS than its own. If I had to guess, it has faster timings (the Number Nine had 40ns memory while the Fastware has 45ns). When I get time, I'll update the charts.

I also have a Number nine reality 332 Virge 325,nice little card, it's installed in my Cyrix 5x86 100 Windows 95 system. It's a good bit faster than my generic virge 325, image quality is far better too, blacks are black unlike many other Virge 325 cards. Here is a pic of the generic one.
[img]

[/img]

Hey clueless1, are you still looking to acquire any cards for your charts?

I've started doing some benchmarking with a mobo (w/ Pentium 90) that I keep around for newly acquired hardware testing. I've had an ARK2000PV (Diamond edition) card in a box that I've thought about putting in my main machine (P200MMX), and figured I would take it and a few others to do my own benchmark work. Maybe use some of the same stuff you do - although I haven't benched with Duke 3D or Descent II before. Did you have a certain path you would run or something while showing FPS?

My workbench monitor is LCD, and unfortunately that ARK2000PV has vertical banding on it. I imagine if I decide to go with the ARK and put it in an actual system (which uses a CRT), I won't have to worry. Still, that is too bad.

CkRtech wrote:

Hey clueless1, are you still looking to acquire any cards for your charts?

I've started doing some benchmarking with a mobo (w/ Pentium 90) that I keep around for newly acquired hardware testing. I've had an ARK2000PV (Diamond edition) card in a box that I've thought about putting in my main machine (P200MMX), and figured I would take it and a few others to do my own benchmark work. Maybe use some of the same stuff you do - although I haven't benched with Duke 3D or Descent II before. Did you have a certain path you would run or something while showing FPS?

My workbench monitor is LCD, and unfortunately that ARK2000PV has vertical banding on it. I imagine if I decide to go with the ARK and put it in an actual system (which uses a CRT), I won't have to worry. Still, that is too bad.

Yeah, I still have my eyes out for a couple of other PCI graphics chipsets, but not spending any money for them, so the odds of finding them for free are pretty low. 😊

There's really no repeatable way to get in-game framerates for Duke and Descent II except to start a new game and measure the standing FPS, which is how I do it. In Duke, start a game, type DNRATE to bring up the frame counter, then hit ESC, start a new game again, wait for the rocket to crash and the framerate to settle before taking the result. Same idea in Descent II, except the code to access the frame counter is FRAMETIME.

I hope im not stepping on any toes here, but I have a bunch of PCI and early AGP cards benchmarked using the Phils Computer Lab DOSBENCH package, I can post some results if you need more. I did them all on a P3-933 on an i815EP motherboard with 256mb RAM and Win98SE/DOS6.22 (Similar to the specs in your second post?) Someone asked about a Cirrus 5464 and here are the results I got on that system:

Cirrus Logic CL-GD5464 4MB RAMBUS PCI

Option 2: 167.7 (3DBench 1.0c for Faster PC's)
Option 4: 52.8"score" / 31.6FPS (Chris's 3D Benchmark 640x480)
Option 6: 14.4FPS (PC Player Benchmark 640x480)
Option B: 66.5 (Doom Max Details) (1123 realticks, i think the formula is FPS=74690/realticks (which was 1123)?
Option C: 93.2FPS (Quake 320x200)
Option E: 11.5FPS (Quake 648x480)

For Grins in Win98SE, Final Reality 1.01: 2.76 "score"
3Dmark99 and 2000SE wouldnt run with it.

Results put it among the slowest of all the cards I tested so far.

(edited for Typos)

How do you tell the difference by visual inspection if the ATI Rage 128 GL is for Mac or PC? I almost never see anyone with the PC version of the 128 GL PCI. What does that sticker on the BIOS say? Is that what gives it away?

feipoa wrote:

How do you tell the difference by visual inspection if the ATI Rage 128 GL is for Mac or PC? I almost never see anyone with the PC version of the 128 GL PCI. What does that sticker on the BIOS say? Is that what gives it away?

Honestly, when I found this card, I had no idea there was also a Mac version. So I'm not sure how to tell the difference. But here are a couple of photos that may help you.

Great, thank you.

I wonder if a full 128-bit RAM TNT2 (non M64 or Vanta) would be significantly faster . Do they exist in PCI variants ?

darry wrote:

I wonder if a full 128-bit RAM TNT2 (non M64 or Vanta) would be significantly faster . Do they exist in PCI variants ?

I don't think it would be any faster. I've also tested (not shown here) AGP cards in DOS and the TNT, TNT2 M64, and GF3 Ti200 all perform about the same. When you get to GF4 and GF5, DOS performance starts to go down. And of course, AGP is faster than the same card in PCI. So you get about the same DOS performance from TNT to GF3.

I bought me a Hercules Stingray 64 with ARK2000PV based on the tests here. Its the 2MB Version and i run it with the latest P200 MMX.
I also did a quick quake bench on VGA with result of 42.0 FPS. Very happy with the GPU, many thanks for your testing guys!

I also did a quick quake bench on VGA with result of 42.0 FPS

That's a bit too slow even with sound. Which motherboard?

really? Here is my MB: https://stason.org/TULARC/pc/motherboards/A/A … um-MS-5146.html
i used the quake 640x480 bench from dosbench (so without sound). But after the bench i checked the resolution of the demo in the options and it say 320x200. What is the best way to do a quake bench? is there a console command or something?

Ah, SIS boards. Then it's correct probably, they're slower than Intel.

is there a console command or something?

Timedemo demo1

What is the reason for Quake running on PIII 933 to produce less FPS even at 640x480?

What is best way to get Quake to have better frame rate, not the hard ware listed here for "PCI graphics round up - dos"?

Cheers,