Why is this board so fast?

[Moogle!]

Well, fast for a given value of fast, anyway. Nonetheless, with an Intel DX2 66 (write through) and a Cirrus Logic GD5422 (STB Horizon), I get 6.8 FPS in the Quake Time Demo. I cannot get any of my other 486 boards to match that, using the same CPU and video card. The all ISA Symphony board gave me 6.4, an Sis board gave me 6.6, and an Opti board (with a Mr. BIOS no less) gave me 6.7. All boards have been tweaked with the most optimal timings they will actually run with.

Here's a picture of the wonder pig itself. It has 8Mb ram and 128K cache, the other three had 16MB and 256K cache. (For reference, the aforementioned Opti board manages 7.8 FPS with my fastest VLB card with the DX2.) The Big Board has its own crystal for the ISA bus, which runs at 8Mhz, the other boards have clock dividers and the tests were run at 8.33Mhz, though speeding the BUS clock up, where possible, did not seem to produce any results, and I tried it on all four boards. The big board has a Phoenix BIOS, The A486, v1.01. I cannot find an upgrade, and while several other BIOSes will POST, none of them really work properly IE, freeze after post, no cache enabled, ect. . The Phoenix BIOS is provided with only basic options for shadowing and cache, and performance is sadly lacking with a Trinityworks 133 Mhz upgrade.

Any thoughts?

 

[maxtherabbit]

cool board, got a manufacture date?

 

[Stone]

Nonetheless, with an Intel DX2 66 (write through) and a Cirrus Logic GD5422 (STB Horizon), I get 6.8 FPS in the Quake Time Demo. ... The all ISA Symphony board gave me 6.4, an Sis board gave me 6.6, and an Opti board (with a Mr. BIOS no less) gave me 6.7.

Do you really think that these 2.56%, 1.28%, and .64% respective deviations you listed are even statistically significant?

All these observed values may well reside within the acceptable margin of error.

This looks more like nit-picking, not statistical analysis.

 

[Moogle!]

cool board, got a manufacture date?

A quick scan of the board turns up the 18th week of 1991 for the most recent date.

Do you really think that these 2.56%, 1.28%, and .64% respective deviations you listed are even statistically significant?

All these observed values may well reside within the acceptable margin of error.

This looks more like nit-picking, not statistical analysis.

You may call it nitpicking, but this is, for me, anyway, part of the fun. To see how things compare. To see what makes a difference and what doesn't. To possibly challenge views and beliefs long held by myself and others. I would never have expected this discrete logic ISA board with less cache, less memory, and slower expansion bus to be as close in performance as the much more tightly integrated integrated VLB system with better specs and a more advanced BIOS. While the ISA system is probably more in-line with what a real person would have had back in the day, the VLB system has had a few upgrades. I would have expected the gap to be bigger than literally one frame. I can only sit here and imagine how much more one would have had to pay for the VLB system vs the ISA system back in the day, just to gain 14 percent more speed.

Not sure why the seemingly sour attitude about it. I thought I left that crap behind on Vogons.

 

[bakemono]

The question is whether there is any real difference between 6.6, 6.7, or 6.8 fps. Did you run the tests more than once and get the exact same result each time? When I ran Quake benchmarks on my 5x86 I recall that the fps could hop around by +/- .2fps at least.

If the difference is real, there are a lot of things that could account for it. Bus timing, cache, DRAM refresh rate, maybe the crystal driving the system timer is dragging a bit, who knows...

 

[Moogle!]

I have noticed some variance, but usually these are done a couple times. The 6.8 FPS is indeed consistent.

 

[Stone]

But 2.56%, 1.28%, and .64% deviations are defined as within the margin of error. IOW, the deviations aren't large enough to warrant further probing since they are too small to certify that they even exist in actuality.

 

[rittwage]

I would never assume a discrete logic-based board would be slower than one with VLSI. It just depends.

With such a slow framerate, I doubt the video is being pushed at all, so ISA vs VLB for video will make no difference. The system is not waiting to fill the framebuffer.

This is more to do with CPU and RAM. That board with so many slots RAM looks like a server board- it may have a slightly optimized RAM interface/pipeline that shows up as 0.1FPS of speed here.