The PCI Trio64 is a bit faster than the PCI Vision968, but when I've tested my VGAs in DOOM it was 0,4 FPS difference with a Pentium 133 (you can see the results in Phil's benchmark database if you filter BIOS to "Compaq 04/07/97" and manufacturer to "S3").

feipoa wrote:

3) PCI-only systems outperformed the VLB-only system in every instance except for the Landmark test. I wonder what Landmark is testing for and why the VLB-only system came out ahead? Even more curious is why the VLB+PCI system was the top performer Did the PCI-only systems outperform the VLB-only system because the S3 Trio64V2/DX is much faster than the S3 Vision968?

That is quite strange...try VIDSPEED or my VIDBENCH benchmark (but don't forget to enable VBE 2.0 and use linear framebuffer for latter one!)

@feipoa: Could you please use graphs where the x-axis starts at zero?
At the moment the bar length suggest a much higher difference between the cards as there actually is.

Nice comparison, great work (as always)!

elianda wrote:

@feipoa: Could you please use graphs where the x-axis starts at zero?
At the moment the bar length suggest a much higher difference between the cards as there actually is.

If I did that, then you would get the other extreme. It is probably best to use a chart normalised to a certain system, or to express the values as a percent difference of some system. However, the values are also presented on the graph, so anyone can determine this themselves.

I was not intending to show favour for one type of system over another, just that PCI systems are not "slower" than VLB-only systems and that PCI on VLB+PCI systems sucks. I suspect the VLB on VLB+PCI systems would at least match that of VLB-only systems if the cache write cycle could somehow be set the same - perhaps if using a 33 MHz FSB.

The MS-4144 was a fantastic board and I miss mine. Nice to see it did well in these tests, makes me proud to have owned one. Recently advised a guy in Denmark to buy one and I'm sure he won't be disappointed.

Interesting test results indeed.

You can add PS/2 to the MS-4144 with far less effort than with the Chaintech 486SPM - Adding PS/2 mouse components to a Chaintech 486SPM
If I recall, most of the components are already there.
Why did you get rid of your MS-4144? A fine board indeed.

It had problems with cache when I got it over a decade ago. Things kept breaking and I never managed to track it down, it stopped working earlier in the year by letting out its magic smoke all of a sudden when I was using it. Sucks, I tried everything to keep it running and am amazed it got this far. So far as I can tell, someone had done something weird with the cache at some point as there were burnt traces under the sockets when I got it. Perhaps someone installed a chip backwards at some point? No idea, but it used to break CPUs and RAM from time to time and no amount of probing could ever tell me why. Under that though, I knew it was a kickass board and it used to rip holes in the world of performance very well indeed. It was free anyway, so I had more than my money's worth I guess, over a decade of service was more than I could hope for.

I like my MB-4DUPM now, so I have no reason to get another MS-4144 and therefore won't, but I still miss it. I haven't thrown it out, but I don't think it will ever run again. At least it managed to hang on long enough for the replacement to be well established before it gave up for good.

The MB-4DUPM is a UMC-based board with VLB and PCI. I would be very interested to see how it performs with the above noted tests if you so happen to have a Trio64

I have both a VLB and PCI one... In fact, I have a Diamond version of both but with only 1MB. I could borrow the RAM from my other Trio 64 (As it isn't in use at present) and give it a go later I guess. It has a Pentium OverDrive installed however and I would rather not change that.

Even if you leave the POD in place, I would be interested to see the comparitive results of the PCI Trio64 vs. that of the VLB Trio64 on the same system. And how does the Trio64 VLB compare to that of the Vision968 VLB?

I have updated the original post to include results from a UMC UM4980 system (PC Chips M912 v1.7). That system performed worse than the SiS 471-based system.

I shall test it when I get the chance.

Unfortunately, I don't own any Vision cards. So it will be Trio only. I could test ET4000s on all BUSes but the only PCI one I currently have working is a SPEA branded one which is unusually slow, so I won't bother doing those tests. They may appear many months down the line when I get my WinFast one going. S3 tests though, those should be done within the next week provided I have time and can make room to work on the machine - I also have to fix the OS installation before I can do the tests and I don't have a working floppy, so I need to be at the internals of two things at once to get it going again.

How did the testing go?

Hasn't happened yet, moving is still getting in the way immensely and I currently have no space to put things in while I work on the machine to fix the DOS install and swap cards over. I will get it done, but it might take a while.

Sorry for the delay, I don't like making people wait.

OK, here we go;

UMC 8881/8886 Aquarius MB-4DUPM/E. 48MB RAM, 256KB Cache, Pentium OverDrive 83MHz. Nothing loaded at boot aside from an ANSI driver used by my benchmarking suite.

Diamond S3 Trio 64 VLB 2MB:
> Superscape 1.0c - 76.6
> Topbench - 268
> Landmark (Video) - BROKEN
> Doom - 2134 in 1752
> Quake - 18.0
> PCPBench (VGA) - 19.5

Diamond S3 Trio 64 PCI 2MB:
> Superscape 1.0c - 72.7
> Topbench - 203
> Landmark (Video) 11702.86 Chr/s
> Doom - 2134 in 2168
> Quake - 18.0
> PCPBench (VGA) - 19.5

Strange how some of the results are the same, implying these tests might be limited by other factors. Still, as suspected, VLB is faster on this 486 board, hardly a surprise given it's pretty much tied directly to the CPU. I used Windows 95's version of MS-DOS in "Safe mode command prompt" boot. I removed my Ethernet card when testing both cards as it would have to be absent for the VLB test anyway, thus eliminating skewed results - in short, the configuration was identical for both sets of tests here. The system will be in on the bench for a while, probably until I move, so if you want more tests let me know.

Very intersting. Thank you for this. My Trio64 VX2/DX PCI results were better than the VLB results. I wonder if the VX2/DX improved upton the original Trio inasmuch as 3DBench, Topbench, and Doom are concerned. DO you have a Trio VX2/DX card? If so, could you run these benchmarks again with the Trio64 VX2/DX card?

I have the Vision864 2MB in both VLB and PCI, both from Number Nine GXE series.

The problem is that I don't have a motherboard that can do VLB and PCI, so hard to get a real comparison, but on a POD83@100mhz vs a Pentium 100 on a 430FX:

System 1: Asus VL/I-486SV2GX4 motherboard /w 1MB L2 cache (SIS 471), 1x32MB FPM 60ns RAM
System 2: FIC PT-2003 motherboard (Intel 430FX), 2x16MB EDO 60NS RAM

I have updated my results to include data from the Biostar MB-8433UUD (UMC 8881-based motherboard, "PCI-only-UMC"). It appears as if this motherboard is slower than the SiS471 board, however the SiS496 board still beats the SiS471.

feipoa wrote:

I have updated my results to include data from the Biostar MB-8433UUD (UMC 8881-based motherboard, "PCI-only-UMC"). It appears as if this motherboard is slower than the SiS471 board, however the SiS496 board still beats the SiS471.

Nice, that fits with my results from my PCChips UMC 8881-based motherboard with fake L2 cache (turned off L2 cache on other motherboards to get comparable results).

It would be interesting if you could get hold of a 964 or 968 PCI card. I believe the differences you're seeing between SIS471 and 496 is mainly due to the VRAM and DRAM of the graphic cards and that Trio64 is newer and faster than the Vision series. I've gotten similar performance differences on my two VLB cards running in my Asus VL/I-486SV2GX4 rev. 2.0 with 1024kb cache, the DRAM based S3 Vision864 (Number Nine GXE#64 with 2MB) and the VRAM based Diamond Stealth64 VRAM - S3 Vision968 4 MB (the same card you have). I can post the benchmarks when I get home from work late tonight 😀

I can also switch out the CPU to the AMD 5x86 @ 160mhz