VOGONS


First post, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie

Hi guys. So after picking up some FPM ram yesterday I decided to finalize my 586 configuration by settling on a video card - so I began benchmarking the PCI and VLB video cards I have. I put the results down on paper, then I tough it might be useful for other users if I were to post them.

The 486 system I tested on has the following configuration:

- Cyrix 5x86-100GP @ 120MHz (3x30MHz) with register enhancements enabled
- FIC 486-VIP-IO 256KB cache, VIA chipset
- 2x16MB FPM ram
- MS-DOS 7.1 (win95)

So here are the benchmark results for the 486 platform:

table1.jpg
Filename
table1.jpg
File size
159.17 KiB
Views
3903 views
File license
Fair use/fair dealing exception
chart1.jpg
Filename
chart1.jpg
File size
88.2 KiB
Views
3903 views
File license
Fair use/fair dealing exception

And the cards I tested:

IMG_20150529_210106.jpg
Filename
IMG_20150529_210106.jpg
File size
3.36 MiB
Views
3903 views
File license
Fair use/fair dealing exception

The results are interesting. The S3 Trio64 and Virge cards are faster in Quake and PC Player 3D Benchmark, while the VLB Cirrus Logic cards are quite a bit faster in DOOM and 3D Bench. This might be to the poor PCI implementation on the VIA chipset, or the architectural difference between older VLB cards and newer PCI cards. A pentium platform benchmark will tell us more.

I tried benchmarking my millennium, but it freezes both systems at POST so I think I killed it somehow 🙁 I would have loved to have it in the charts since it's such a popular card.

Swapping in a 160MHz 586 from AMD (4x40MHz) yields interesting results:

a vs c.jpg
Filename
a vs c.jpg
File size
85.86 KiB
Views
3888 views
File license
Fair use/fair dealing exception

I'll also test the AMD 486 with a VLB video card tomorrow and compare again.

Now for the Pentium platform tests! I used a 133MHz non-MMX pentium on the Intel 440bx chipset, on DOS 6.22. Results are similar to the 486 platform albeit 2-3x higher:

table2.jpg
Filename
table2.jpg
File size
160.68 KiB
Views
3761 views
File license
Fair use/fair dealing exception

Fmc3Wcf.jpg

What surprised me is that there is almost no performance difference between the S3 Trio and the S3 Virge. The virge is slighly faster due to the clock speed difference and larger memory (benched with 2MB). It only really takes off in VESA modes, where it sores a solid 1.6 FPS difference (PCPBench VESA 640x480). The gap widens even more as resolutions increase. Testing VESA 102 mode, the trio64 with 2MB installed scores 9.4 FPS, while the Virge with 2MB scored 11.9.

The logical conclusion is that the Virge is rather wasted on DOS gaming systems, unless you plan to play newer games at higher resolutions. But then you could go for the slightly faster Matrox Millennium if you're ok with the latter's compatibility issues regarding older DOS games (and of course it's price).

The Alliance card makes a good effort - solid performance on Quake and PCPBench, but poor performance in Doom. The Cirrus PCI performs as expected of a common budget card. Good enough for a 486 and early Pentium.

Last edited by kanecvr on 2015-05-30, 12:30. Edited 10 times in total.

Reply 2 of 18, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie
firage wrote:

That's odd; 30-35 fps is a weak show in Doom. No experience with that board, though.

I know right? I'll retest the PCI cards using a UMC chipset board to see if things improve.

Reply 4 of 18, by kixs

User metadata
Rank l33t
Rank
l33t
firage wrote:

That's odd; 30-35 fps is a weak show in Doom. No experience with that board, though.

Also 3Dbench should be around 90FPS.

Requests are also possible... /msg kixs

Reply 5 of 18, by dirkmirk

User metadata
Rank Oldbie
Rank
Oldbie

I tested a PCI VIA board also with the Cyrix 5x86-120 and also got poor results, the highest results I saw for the Cyrix chip was 10.1fps 640x480 pcpbench in a umc board, the sis manages around 9fps, Quake scores should be around 14-15fps from what I remember, unfortunately the VIA chipset is no good for a high end 486/5x86.

Reply 6 of 18, by FGB

User metadata
Rank Oldbie
Rank
Oldbie

One shouldn't bench a Trio64 with only 1MB of DRAM installed. Any 64-Bit S3 card needs at least 2MB of memory installed for nominal performance. With only 1MB installed, the restricted memory interface will cripple the performance a bit.

www.AmoRetro.de Visit my huge hardware gallery with many historic items from 16MHz 286 to 1000MHz Slot A. Includes more than 80 soundcards and a growing Wavetable Recording section with more than 300 recordings.

Reply 7 of 18, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie
philscomputerlab wrote:

Nice work! I only have S3 PCI cards, they all work quite well.

Thanks, I was hoping to bench more cards, but my Millennium decided to trow a fit and I don't own any ATi PCI cards...

kixs wrote:
firage wrote:

That's odd; 30-35 fps is a weak show in Doom. No experience with that board, though.

Also 3Dbench should be around 90FPS.

I'm not so sure about that... I was expecting ~50-55 FPS, but not 90 - that sounds like pentium territory to me (benchmarks support that theory). It's possible I may have calculated the FPS using the wrong formula.

dirkmirk wrote:

I tested a PCI VIA board also with the Cyrix 5x86-120 and also got poor results, the highest results I saw for the Cyrix chip was 10.1fps 640x480 pcpbench in a umc board, the sis manages around 9fps, Quake scores should be around 14-15fps from what I remember, unfortunately the VIA chipset is no good for a high end 486/5x86.

From what I remember, my UMC motherboard scored quite a bit higher then the VIA in most benchmarks, including speedsys. It also seems to favor the Cyrix when Register enhancements are turned on, even if the AMD is clocked higher. I'll do the tests later today or tomorow, since I have to set up the UMC board (no bios battery, no PS/2, non-standard com header configuration so no mouse).

FGB wrote:

One shouldn't bench a Trio64 with only 1MB of DRAM installed. Any 64-Bit S3 card needs at least 2MB of memory installed for nominal performance. With only 1MB installed, the restricted memory interface will cripple the performance a bit.

As a mater of fact, I tested the S3 trio with 2MB at first (out of curiosity) - no difference whatsoever. It did score marginally better in VESA 102 (800x600) in PCPBench, but that's it.

Also, I belive you are mistaking. The 64 bit interface requires 1MB, not 2. Two memory chips => 2x512kb = 1MB / 32bit interface per memory chip, as proven by the fact that the board will not run with only one chip installed (I tried). When using 2MB, it actually slows down a bit in some tests due to the fact it needs to share the 64 bit interface across 4 chips).

I might of course be wrong, but this is what I understood after reading the S3 Trio64 chip's datasheet + a little experimentation with a Eagle S3 Trio64V+ that came with four sockets (no memory soldered).

Last edited by kanecvr on 2015-05-30, 19:06. Edited 3 times in total.

Reply 8 of 18, by elianda

User metadata
Rank l33t
Rank
l33t

You shouldn't be so curious about the S3 Trio64V scoring very similar to the S3 Virge in 2D tests as the 2D core is the same.

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 9 of 18, by soviet conscript

User metadata
Rank Oldbie
Rank
Oldbie
elianda wrote:

You shouldn't be so curious about the S3 Trio64V scoring very similar to the S3 Virge in 2D tests as the 2D core is the same.

whats strange is some time ago I did some benchmarks between a s3trio64/V2 and a virge/dx and I had the opposite effect where the trio was very very slightly faster. also I was playing Daggerfall at the time and the first part of the game I was running the Trio. half way in I switched to the Virge/DX and after I started noticing a few graphical issues with floors in dungeons. when a hallways was diagonal sometimes the floor texture would be stretched. Maybe this happened with the Trio as well but I don't recall seeing it until after I switched to the Virge. I know they supposedly use the exact same 2d core so I was a little confused by it all.

Reply 10 of 18, by Tertz

User metadata
Rank Oldbie
Rank
Oldbie

If you'll do testing in Quake with "viewsize 120" (removes a panel at bottom) then your results may be compared with Phil's testing. It gives 10-15% lower fps in 320x200. If you did not do so, certainly.

DOSBox CPU Benchmark
Yamaha YMF7x4 Guide

Reply 11 of 18, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie
elianda wrote:

You shouldn't be so curious about the S3 Trio64V scoring very similar to the S3 Virge in 2D tests as the 2D core is the same.

Theoretically yes, but the Virge is supposed to be clocked faster.

soviet conscript wrote:
elianda wrote:

You shouldn't be so curious about the S3 Trio64V scoring very similar to the S3 Virge in 2D tests as the 2D core is the same.

whats strange is some time ago I did some benchmarks between a s3trio64/V2 and a virge/dx and I had the opposite effect where the trio was very very slightly faster. also I was playing Daggerfall at the time and the first part of the game I was running the Trio. half way in I switched to the Virge/DX and after I started noticing a few graphical issues with floors in dungeons. when a hallways was diagonal sometimes the floor texture would be stretched. Maybe this happened with the Trio as well but I don't recall seeing it until after I switched to the Virge. I know they supposedly use the exact same 2d core so I was a little confused by it all.

Well, the Virge is supposed to have perspective correction.

Tertz wrote:

If you'll do testing in Quake with "viewsize 120" (removes a panel at bottom) then your results may be compared with Phil's testing. It gives 10-15% lower fps in 320x200. If you did not do so, certainly.

I didn't. I used the default view size in quake.

Reply 12 of 18, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie

Well I can't get my UMC board to work (one of the cache chips has gone bad) and I don't want to tear down my working 586, so out of frustration I benched the trio64 and the Virge on my Pentium MMX platform.

chart4.jpg
Filename
chart4.jpg
File size
115.6 KiB
Views
3696 views
File license
Fair use/fair dealing exception

As expected, performance improves quite a bit on the Pentium MMX 233, some benchmarks showing a 50% performance increase. The Virge only really shines in PCP Benchmark at 640x480 where there is a 3,2 FPS difference. Both cards were tested with 2MB of VRAM.

Reply 13 of 18, by idspispopd

User metadata
Rank Oldbie
Rank
Oldbie
kanecvr wrote:
elianda wrote:

You shouldn't be so curious about the S3 Trio64V scoring very similar to the S3 Virge in 2D tests as the 2D core is the same.

Theoretically yes, but the Virge is supposed to be clocked faster.

There are no real standard clocks for S3 Trio or Virge cards, especially not for RAM access. The RAM speed depends on the actual RAM chips and the settings in the BIOS (just like with modern cards).
You could check the clocks with MCLK in DOS or maybe PowerStrip in Windows. From the photo you posted I can't recognize any inscription on the RAM chips.
For further information you could read http://www.geocities.ws/liaor2/myutil/mclk.html.

kanecvr wrote:
soviet conscript wrote:
elianda wrote:

You shouldn't be so curious about the S3 Trio64V scoring very similar to the S3 Virge in 2D tests as the 2D core is the same.

whats strange is some time ago I did some benchmarks between a s3trio64/V2 and a virge/dx and I had the opposite effect where the trio was very very slightly faster. also I was playing Daggerfall at the time and the first part of the game I was running the Trio. half way in I switched to the Virge/DX and after I started noticing a few graphical issues with floors in dungeons. when a hallways was diagonal sometimes the floor texture would be stretched. Maybe this happened with the Trio as well but I don't recall seeing it until after I switched to the Virge. I know they supposedly use the exact same 2d core so I was a little confused by it all.

Well, the Virge is supposed to have perspective correction

That only matters if the game actually uses the 3D acceleration features of the chip, which Daggerfall doesn't do to my best knowledge.

Reply 14 of 18, by lazibayer

User metadata
Rank Oldbie
Rank
Oldbie
kanecvr wrote:
As a mater of fact, I tested the S3 trio with 2MB at first (out of curiosity) - no difference whatsoever. It did score marginall […]
Show full quote
FGB wrote:

One shouldn't bench a Trio64 with only 1MB of DRAM installed. Any 64-Bit S3 card needs at least 2MB of memory installed for nominal performance. With only 1MB installed, the restricted memory interface will cripple the performance a bit.

As a mater of fact, I tested the S3 trio with 2MB at first (out of curiosity) - no difference whatsoever. It did score marginally better in VESA 102 (800x600) in PCPBench, but that's it.

Also, I belive you are mistaking. The 64 bit interface requires 1MB, not 2. Two memory chips => 2x512kb = 1MB / 32bit interface per memory chip, as proven by the fact that the board will not run with only one chip installed (I tried). When using 2MB, it actually slows down a bit in some tests due to the fact it needs to share the 64 bit interface across 4 chips).

I might of course be wrong, but this is what I understood after reading the S3 Trio64 chip's datasheet + a little experimentation with a Eagle S3 Trio64V+ that came with four sockets (no memory soldered).

I read the datasheets for both Trio64 and Trio64V+. The sticker on the video BIOS hints the chip might be a Trio64V+ (86C765).
According to page 7-1 from both manuals, both Trio64 and Trio64V+ can operate with either 32bit or 64bit bus, and the bus is 32bit wide when 1MB VRAM is installed. The VRAM chips on the expertcolor card should be 16bit wide each.

Reply 15 of 18, by overdrive333

User metadata
Rank Newbie
Rank
Newbie

My am5x86-p75@160mhz (40*4) is ~1,5-2,0 times slower than p75 (50*1.5) in tomb raider 1.
Is it normal or there is something wrong with my 486?

P75 - 430vx, 32mb edo, win98, s3 trio64
am5x86 - taken pci400-4 (SIS496) , 8mb fpm, win95, s3 trio64. (8mb ram is enough for tr1)

Reply 16 of 18, by amadeus777999

User metadata
Rank Oldbie
Rank
Oldbie

The VIA chipset seems way too slow - how are your settings regarding cache/ram/isa timings?
In NSSI I get around 69000 Dhrystones and 21300 KWhetstones.
3dbench2 - 77.6fps.
Doom demo3 - ~60fps

Reply 17 of 18, by FGB

User metadata
Rank Oldbie
Rank
Oldbie

Hi,
digging out this thread to add some maybe useful information:

I recently bought a small stock of the cool FIC 486-VIP-IO boards and configured one board for a customer for a Cyrix 5x86 CPU.
While the board has a lower memory throughput than other late 486 boards, the CPU scores are very good and the graphic scores are quite fast.
For a good graphics performance, a proper VLB card is a must. The PCI bus on this board is bridged from the VLB bus (VIA505 bridge chip), this eats performance when you use a PCI graphics card.

Talking about scores, the board with 256KB of L2 cache and the Cyrix 5x86 100GP CPU without any tools to optimize performance gives you:

76,9 frames/s in 3DBench 1.0 with a fast VLB S3 Vision864 graphics card with 2MB of DRAM
62,5 frames/s in 3DBench 1.0 with a fast PCI S3 Vision864 graphics card with 2MB of DRAM

kanecvr's previous posted graphics scores are much lower and may well reflect misconfiguration and or wrong BIOS settings or old underperforming BIOS-versions. I recommend BIOS 4.26ES4 for good performance.

Hope that might be helpful for anyone 😀

Greetings
Fabian

www.AmoRetro.de Visit my huge hardware gallery with many historic items from 16MHz 286 to 1000MHz Slot A. Includes more than 80 soundcards and a growing Wavetable Recording section with more than 300 recordings.

Reply 18 of 18, by punkture

User metadata
Rank Newbie
Rank
Newbie

Hi

The via chipset seems slow compared to the sis chipset. My c5x86 (LS486e@2x66) with Matrox storm scores 19.3 in Quake, and 30.1 in glquake. In Doom i get around 58 fps (Doom seems to score higher with 40 fsb)