lukas12p wrote on 2021-05-17, 00:02:

I have more of this motherboards and two with volt.mod so can adjust any voltage (used for IBM 5x86 100@133)

What Quake 1.06 score do you get when using your IBM 5x86c at 133 MHz with enhancements on and PCI at 66 MHz? Is PCI at 33 MHz the same result?

I had a heatsink slapped on CL5429s with 50ns RAM when I was running 60+, only had one that would do it out of a selection of 3, well one had 70ns RAM on so that was gonna be a turd, the other had 60ns and just got too hot on the "GPU" even with a sink on. The 50ns one just about coped.

Edit: BTW a trick for preselecting your cards I used was to see if they played nice on 40MHZ with no wait states set, if they coped with that, then they were usually good above 50Mhz with the wait states set.

Anonymous Coward wrote on 2021-05-17, 03:45:

pshipkov wrote on 2021-05-17, 02:47:

Ark2000PV lights-up at 66MHz and is a hair faster than Matrox Millennium in DOS.
If lukas12p has one of these - it may be better for the DOS tests.

I had good luck with my ark1000vl at 66MHz on VL-bus as well.

I have an ARK1000 PCI card, so the chip is quite OK with using at 66MHz regardless.

The Serpent Rider wrote on 2021-05-17, 01:14:

Try VESA 320x200. […]
Show full quote

4C under load and 20.6 fps in Quake1

Try VESA 320x200.

What voltage would be safe?

4V should be safe for prolonged use.

I use PhilsComputerLab dosbench, like Peter 'cpu-galaxy' for testing, so it is comparable

feipoa wrote on 2021-05-17, 03:46:

lukas12p wrote on 2021-05-17, 00:02:

I have more of this motherboards and two with volt.mod so can adjust any voltage (used for IBM 5x86 100@133)

What Quake 1.06 score do you get when using your IBM 5x86c at 133 MHz with enhancements on and PCI at 66 MHz? Is PCI at 33 MHz the same result?

Cyrix 5x86 120@133 (2x66) PCI66 = 18.8 fps
Cyrix 5x86 120@133 (2x66) PCI33 = 18.0 fps, but I don't remember if it was the same card 😒

I use PhilsComputerLab dosbench, like Peter 'cpu-galaxy' for testing, so it is comparable

We're squeezing as much FPS as possible at this point and using VESA is still a legit way to obtain higher score.

The Serpent Rider wrote on 2021-05-16, 09:34:

Depends in which test. Quake in VGA mode has very small increase, scoring only 0.1-0.2 fps more with 66 Mhz PCI bus. It should s […]
Show full quote

How much benefit performance benefit did running the graphics at 66 MHz award, that is, if the 1/2 FSB-to-PCI multiplier was enabled, how much lower would the score be?

Depends in which test. Quake in VGA mode has very small increase, scoring only 0.1-0.2 fps more with 66 Mhz PCI bus. It should score slightly more in VESA modes and probably 1+ fps more in Mode X, compared to normal 33 MHz PCI bus results.

When the motherboard is inside a case, is there any condensation build-up on or around the CPU which could cause a short if it were to drip.

I doubt that such experiment was for daily use. But you can insulate the area around CPU socket with plasticine. Which is a common practice during below zero Celsius runs in overclocking community.

Does the ISA bus speed run at a fraction of the PCI bus clock or the FSB?

He ran it at minimal possible value on LuckyStar board, which still should be somewhat overclocked at 66 Mhz. Maximum possible value is 22 Mhz or 1/3 of PCI clock. It works perfectly fine with some sound cards, video cards or controllers.

The integrated FDC on my lucky star stops working at 1/3PCLK when the FSB is 50. Had to back off the ISA to 1/4. (PCI is at 1:1 FSB)

The Serpent Rider wrote on 2021-05-17, 14:59:

We're squeezing as much FPS as possible at this point and using VESA is still a legit way to obtain higher score.

I noticed you mentioned this couple of times.
In my experience low-res VESA modes in DOS, established with UniVBE and tested with bunch of PCI/VLB/ISA cards, result in same or slightly worsened frame rates.
What is your magic formula ?

In my experience low-res VESA modes in DOS, established with UniVBE and tested with bunch of PCI/VLB/ISA cards, result in same or slightly worsened frame rates.
What is your magic formula ?

vid_wait 0
vid_nopageflip 1

Also CPU Galaxy can't utilize VESA with his current card, due to very buggy support on Permedia 2 chips. Better choices are Matrox, Nvidia, some late S3 cards and 3dfx. ARK cards were mentioned to work with 66 Mhz bus, but their VESA performance is yet to be seen. Tseng ET6000 has excellent VESA performance, but may or may not be compatible with doubled PCI clock.

Univbe etc get you nothing on mode X but usually noticeably faster on a 640x480 mode vs standard VGA 640x480 in my experience.

The Serpent Rider wrote on 2021-05-17, 23:40:

Also CPU Galaxy can't utilize VESA with his current card, due to very buggy support on Permedia 2 chips. Better choices are Matrox, Nvidia, some late S3 cards and 3dfx. ARK cards were mentioned to work with 66 Mhz bus, but their VESA performance is yet to be seen. Tseng ET6000 has excellent VESA performance, but may or may not be compatible with doubled PCI clock.

CPU Galaxy did his 200 mhz record setter with a banshee based CT6760 at 66mhz.

It is ok to alter the sw/hw configuration (the environment) for best performance, which is actually the goal, but the test itself should be constant.
For good or bad Phil set his tests in a particular way. They ended up being the adopted "standard" for interactive DOS graphics.
Let's not change that, so we have common starting point for comparison.

pshipkov wrote on 2021-05-18, 03:19:

It is ok to alter the sw/hw configuration (the environment) for best performance, which is actually the goal, but the test itself should be constant.
For good or bad Phil set his tests in a particular way. They ended up being the adopted "standard" for interactive DOS graphics.
Let's not change that, so we have common starting point for comparison.

Eh? No, nothing Phil did is a "standard". As far as Quake is concerned, if there is a decent "standard" out there, then it is not Phil's but actually thandor.net's because he has useful numbers to compare against: https://thandor.net/benchmark/33

As for Doom, Phil's benchmarks are anything but standard. The "standard" is almost always to install the shareware and run it at default screen size, not full screen. And that way you at least have comparable scores to benchmark with, again courtesy of thandor.net: https://thandor.net/benchmark/69

I don't have strong stance about which should be the standard set of tests and rules, as long as there is a standard of sort, so we can compare apples to apples.
If tomorrow the community here moves to another tests and rules - fine by me.
As for today - everybody seems to be going with Phil's, which makes it the standard.
But als0 - in my opinion Phil's test suite is a bit too basic and not that great of a fit.
The only two useful things there are Quake 1 and Doom.
The rest is some arbitrary synthetic benchmarks that are kind of meaningless.

The only synthetic DOS benchmark that I have found to be a pretty accurate measuring yard for overall speed is NSSI. It almost pinpoint accurately measures speed compared to other CPUs, even under weird conditions like disabled L1 or L2 cache.

Of course, it's not in Phil's suite so nobody uses it. Instead we have Youtube videos of people running Landmark on Pentium III systems..

They ended up being the adopted "standard"

That's not the main point of this "race". I consider VESA as just another tweak, since resolution don't change.

So, following your logic - can I go in code and make the vertex animations of the entities skip N out of M frames ?
Will you accept that as just another tweak, since render resolution didn't change ?
I am not trying to be difficult, but hope you see my point - both of us will be altering the test and skewing results to a different degree.

@appiah4
Yes, I also think that NSSI is one of the better ones.

Will you accept that as just another tweak

No, same as viewport downsizing or other image quality degrading tricks. Simple. VESA does not alter image quality.

I upgraded my Cx5x86's video card today, from a Macronix MX86200 PCI 2MB to a SiS 86C326 PCI 4MB, and it turned out to be quite the downgrade. Quake 320x200 went from 16.6fps to 15.7fps, and I went from 83.3 fps in 3DBench to 62.something and 22.2 fps in 3D Player Benchmark to 20.7. Around 10-15% slower all around, kind of shocking. I will stick with this card because I want the S-Video out for capture purposes but I really need to find a PCI Riva128/TNT with S-Video it seems.

Interestingly, this card is faster in VESA supporting games. Sim City 2000 is much smoother, and Duke3D VESA 320x200 is smoother (although 640x480 is still out of the question). That said, how can Quake be slower then? Quake doesn't use VESA modes for 320x200, just plain 13h VGA? If that's the case, is it possible to leverage VESA in Quake 320x200?

Quake doesn't use VESA modes for 320x200, just plain 13h VGA?

It does. Type "vid_describemodes". Quake can't see VESA modes only if they above 1280x1024 resolution.

320x200 can be set in 3 modes: 13h, Mode X and VESA (only with VBE 2.0 or newer). Mode X could be more favorable with some video cards.

I get your point, but still think that the test setup should not be altered from inside.
There are reasons why tests-tests such as 3D Mark or whatever don't provide options to tweak things internally.
Same applies here, despite the test software being a game that was not really intended for such use and leaves its settings wide open.