A few titles I have been testing.

MechWarrior 2 CD (Matrox edition)

Fifa demo(3DFX)

MDK 2 (3DFX) Low Detail (10.00)fps 640x480

Forsaken (3DFX) BIODOME demo (23.77)fps 640x480

Sweet! Looking good, rgart.

Try playing Forsaken with the Matrox G200 for a few minutes. Then compare it to the 3dfx and report back the difference.

Final stable system specs submission:

Cyrix enhancement Branch Prediction + DOS 7.1 - stable. (75.50)

Cyrix enhancement Branch Prediction + Windows 95 GUI loaded - unstable. (76.14)

Same motherboard, but with an IBM 5x86c-133/2x, Ultra100 ATA controller, and Yamaha OPL3 sound. Taken by booting into DOS 7.1 instead of Win95. I think the speedsys CPU score will increase a little if you boot directly from a diskette, bypassing all Win95-DOS drivers.

Notice the improvement in RAM and cache speed when using the 66 MHz FSB. The L1 speed is faster as well.

33 MHz vs. 66 MHz
L1: 183.5 vs. 198.4 MB/s
L2: 59.6 vs. 73.5 MB/s
RAM: 43.4 vs. 56.5 MB/s

3DFX Voodoo 2 card (12MB) getting some passive love.

MDK2 3DFX 640x480 - low detail - 7 fps (barely playable)

MDK2 3DFX 640x480 - high detail - 3 fps (unplayable)

A little off topic but none the less Cyrix related, fixed all the bent pins on this unmarked Cyrix CPU today and I tested it briefly, I can tell you its a 4x Stepping 0 Revision 5. I can also confirm it didn't like 133MHz operation at 3.6 volts and speedsys crashed on startup(with Cyrix enhancements on). I only spent 20 seconds on it but thought it was worth noting.

Did it work at 3.7 V or 3.75 V?

feipoa wrote:

Did it work at 3.7 V or 3.75 V?

It was not stable at 3.7 V or 3.75 V. I could benchmark Quake and run other tests but eventually it always crashed. I spent some time on it and then swapped back to the Cyrix 5x86 133MHz x 4 at 3.6V

Today I swapped out the Matrox G200 and tested a few different video cards in DOS 7.10 / Windows 95. -without univbe-

Rendition Verite V2200 (8MB)

Speedsys: 74.59
3DBench2: 30.8
Pcpbench /vgamode: 17.7
Pcpbench 640x400: 8.0
Pcpbench 800x600: 5.3
Pcpbench 1024x768: 3.6

I tried using reference drivers v2.3 and v3.0 for Windows 95 but kept getting a general protection fault and crash on boot. I admit I don't know much about these cards, was the V2200 competing with 3DFX Voodoo1 cards back in the day? Is a Voodoo2 superior? I was unable to get Windows 95 loaded and test it fully, DOS 7.10 games and apps ran slowly. This card appears to change your font in DOS and through the boot process.

I tested a bunch of S3 VIRGE/DX cards in DOS 7.10 and found there results varied among different revisions with one of them being a clear winner even surpassing the Matrox cards substantially at 800x600 and 1024x768 in pcpbench, the animation itself was noticeably faster with this S3. I noted I couldn't pick 640x400 resolution or above in Quake using DOS 7.10.

(S3 Virge)

Speedsys (v4.78): 74.79
3dbench2 (v1.0): 90.6
Pcpbench /vgamode v1.97 (320x200): 23.7
Pcpbench 100 v1.97 (640x400): 10.5
Pcpbench 103 v1.97 (800x600): 7.0
Pcpbench 105 v1.97 (1024x768): 5.0

(Matrox Mystique 220)

Speedsys (v4.78): 75.42
3dbench2 (v1.0): 90.1
Pcpbench /vgamode v1.97 (320x200): 23.7
Pcpbench 100 v1.97 (640x400): 8.4
Pcpbench 103 v1.97 (800x600): 5.3
Pcpbench 105 v1.97 (1024x768)3.6

(Matrox G200)

Speedsys (v4.78): 75.50
3dbench2 (v1.0): 90.1
Pcpbench /vgamode v1.97 (320x200): 23.7
Pcpbench 100 v1.97 (640x400): 8.4
Pcpbench 103 v1.97 (800x600): 5.3
Pcpbench105 v1.97 (1024x768) 3.6

(Rendition V2200)

Speedsys (v4.78): 74.59
3dbench2 (v1.0): 30.8
Pcpbench /vgamode v1.97 (320x200): 17.7
Pcpbench 100 v1.97 (640x400): 8.0
Pcpbench 103 v1.97 (800x600): 5.3
Pcpbench105 v1.97 (1024x768) 3.6

I switched back to the Matrox G200 after testing and with a fresh Windows95 install I think this system might be complete.

◙ SilverStone S1803212HN-3M 180mm 3Pin Case Fan

GLQuake @ 23.5 FPS AVG with sound @ 800x600.3DFX
GLQuake @ 26.3 FPS AVG without sound @ 800x600.3DFX

◙ Cyrix 5x86 133MHz
◙ 512K Cache
◙ 64MB FPM RAM
◙ Biostar MB-8433UUD
◙ Matrox Millenium G200-8MB
◙ Creative AWE64 GOLD-28MB
◙ 3DFX VOODOO2-12MB
◙ 20GB SCSI Hard Disk
◙ SCSI Plextor CD-ROM
◙ Sony 17" TFT LCD SDM-S73
◙ Honeywell-SUH keyboard
◙ Microsoft 2-Button PS/2 Mouse 2.0
◙ X-Fighter Thrustmaster Joystick
◙ Roland SC55-ST MIDI Device
◙ JBL GO Portable Bluetooth Speaker
◙ Full Tower case with MHz LED display.

I have noticed that the G200 in pcpbench at 640x480+ is slower than the Virge cards, but performs well in VGA mode. Not sure why this is. Maybe there is a way to improve it?

A lot of these Virge cards don't have a very clear display, but depends on the brand and BIOS. I have one with a nice display and 4 MB, but the G200 still looks best in Windows and works well with many early Direct3D games. Do any Trio/Virge cards work with Direct3D?

I thought you switched your Cyrix 5x86-133 system to run at 3.7 V. Why did you switch back to 3.6 V?

feipoa wrote:

I have noticed that the G200 in pcpbench at 640x480+ is slower than the Virge cards, but performs well in VGA mode. Not sure why this is. Maybe there is a way to improve it?

A lot of these Virge cards don't have a very clear display, but depends on the brand and BIOS. I have one with a nice display and 4 MB, but the G200 still looks best in Windows and works well with many early Direct3D games. Do any Trio/Virge cards work with Direct3D?

It seems the S3 Virge does perform very well at 640x480 and 800x600 in DOS. The difference in pcpbench is extremely apparent by how smooth the animation is.

feipoa wrote:

I thought you switched your Cyrix 5x86-133 system to run at 3.7 V. Why did you switch back to 3.6 V?

In the end I found stability at 3.6 V

(my other machine) The Cyrix 120MHz (S1R3) in the OP is currently in pieces and decommissioned. I set up a small test bench today using the same hardware.

CPU: Cyrix 5x86 (S1R3)
Motherboard: UM8810-PAIO Revision 2.1 (with VR-102 REV1.0 3.6 V)
RAM: Toshiba TC5117400CSJ-60 (64MB)
Video: Matrox Mystique 220 (Rev.A)
HDD: Seagate Medalist 10GB (IDE)

OS: Windows 98 SE + Unofficial patch

Cyrix Enhancements loaded prior to entering Windows 98 and so far branch prediction is working stable and without crashing. I did note to turn BTB_EN on I had to turn RSTK_EN off.

The UM8810 is somewhat picky about your hardware choices but is a solid and stable work horse with plenty of BIOS options. The motherboard wouldn't turn on with either the Rage 128 VR or Rendition Verite V2200 inserted. (black no signal screen)

What criteria did you use to determine the system was initially not stable at 3.6 V, then stable at 3.7 V, then again stable at 3.6 V? Personally, I would at least run it at 3.65 V. Also, 512 KB of single-banked cache is not as stable as double-banked cache. If you continue to have odd things happening, you might try 256 KB of double-banked cache and set the L2 to WT mode if using 64 MB of RAM.

On your Biostar system, what is your min and max frame rates in Duke3D at 800x600 using the Matrox G200? It would be interesting to see what the S3 Trio/Virge cards get in Duke3D. They will probably be a little higher. You might want to see if using UNIVBE improves the Duke3D scores in SVGA modes.

Curious why your UM8810-PAIO system was decomissioned?

I generally leave RSTK and LOOP off when enabling BTB for Windows. Also, BTB has minimal impact on Quake benchmarks. I think 0.3 fps at most.

feipoa wrote:

If you continue to have odd things happening, you might try 256 KB of double-banked cache and set the L2 to WT mode if using 64 MB of RAM.

Both systems are stable and I haven't had any stability issues for some time.

feipoa wrote:

On your Biostar system, what is your min and max frame rates in Duke3D at 800x600 using the Matrox G200? It would be interesting to see what the S3 Trio/Virge cards get in Duke3D. They will probably be a little higher. You might want to see if using UNIVBE improves the Duke3D scores in SVGA modes.

Would be interesting to see the result!

feipoa wrote:

Curious why your UM8810-PAIO system was decomissioned?

The original UM8810 motherboard in this thread with the Phoenix BIOS died. This is a different board that I received with the voltage module (VR102)

Here is the speedsys result of the Cyrix 5x86 120MHz with Elitegroup UM8810-PAIO and the Cyrix 5x86 133MHz with Biostar MB-8433UUD

Wait, so how are you establishing stability? You stated the contrary just two weeks ago,

rgart wrote:

I have found 3.6 volts on my 3.6 volt chip to be very close to the desired voltage of the chip but not quite. I get the occasional crash I now don't get at 3.7.

Its been pretty heavy use and minor tweaking to the bios in those two weeks till now to determine the stability. I use Prime95, Memtest86, Windows 95 and a number of DOS games with a couple of DOS benchmarks to test stability.

I thought the voltage had been related to a general protection fault crash I was receiving every time I tried to install Windows but it turned out to not be. I had further less subtle crashing due to too many bad hard disks.

Initially when I turned this system on after so much downtime I was getting some odd readings from the hard drive and controller with quite a few errors, it may have been dust and time though as everything just needed to be cleaned, reseated and reinserted.

Both of these systems are running stable at 3.6 V

feipoa wrote:

On your Biostar system, what is your min and max frame rates in Duke3D at 800x600 using the Matrox G200? It would be interesting to see what the S3 Trio/Virge cards get in Duke3D. They will probably be a little higher. You might want to see if using UNIVBE improves the Duke3D scores in SVGA modes.

I tested Duke3d on the Matrox G200 at 800x600 with graphics on (high) and shadows (on) version1.3D and the demo plays at between 9 - 16 fps. The demo tester is a whiz and doesn't stand still much, actually playing the game I get up to 24fps.

I tried the S3 Trio card pictured above using SVGA modes in pcpbench with univbe version 5.1a and 6.7.

univebe 5.1a yelied the best results for the S3 but the Mystique needed univbe 6.7 to detect the chipset properly.

3dbench2: 99.4 (Mystique 99.6)
pcpbench 300x200: 22.9 (Mystique 22.9)
pcpbench 640x400: 10.5 (Mystique 9.4)
pcpbench 800x600: 6.7 (Mystique 6.0)
pcpbench 1024x768: 4.8(Mystique 4.3)

I found another S3 Trio card that yielded 6.8 in pcpbench 800x600 mode with Univbe 6.7, looks almost identical to the other S3 Virge but this one is the DX variety.

Today I upgraded from 256K to 512K L2 Cache on the UM8810 board. A little bit of a mongrel mix but it seems stable, both rated - 15ns. I'm using the tightest timings and a 1:1 HOST/PCI clock with the Cyrix 5x86 120MHz which seems to contradict the UM8810 manual,