My cache is getting quite warm. How hot is it supposed to get?

I set my HOST/PCI to 1:1/2 in the BIOS, could the host/PCI clock setting be causing the cache heat?

Should I back off the timings from 2,1,1,1 ?

The thing with stabily on these systems when pushing them to the limits is that sometimes it can be stable, and at other times, with the same conditions, not stable. No errors on MemTest after 30 hours is a good start, but not enough. Same with an error free Windows 95 and NT4 installation. It has taken me years of on-off tinkering to really uncover what causes faults with specific hardware combinations. Some of the CPU collectors on CPU-World have theorised that none of the Cyrix 5x86-133/4x chips are entirely stable at 3.60 V and that they all should have read 3.7 V. Another good test to do is let Quake (DOS) and GLQuake run in look for at least 90 minutes. That is, let it run the autoplay game preview. Letting an mp3 play at full sampling overnight is another essential test.

I remember some years ago I was having a really hard time determine what was giving me a BSOD in NT4 with the Adaptec 2940U2W. I turned the system upside down and was pulling my hair out. I eventually determined that the power connector to the SCSI HDD wasn't tight enough, which I think was not letting enough current get to the HDD with extended uptime. Most times, it would take 1-3 before it would crash. The clue which finally gave way was that occasionally (actually, rarely), I would hear the hard drive spin down, then back up again. I desoldered the power connectors and replace them and never had the problem again. Errors which only occur after extended uptimes can be the most frusturating to troubleshoot.

If reseating the cards fixed things, it makes me wonder if something got messed up in ESCD (which gets rewritten after reseating). If you are lucky, your issue is isolated to some bad HDDs. Could also be the cable. Make sure you don't have the HDD terminator jumpers installed if your cable is terminated.

Did Univbe improve the Matrox G200 scores at all? What about the Mystique?

If you don't care much for image quality in Windows and don't want to play any early D3D games, you could always switch to the Virge DX.

I have the Atomic version of Duke3D. How do I get the game to demo autoplay?

Oh nice, you can use double-banked 512 KB cache on the UM8810 motherboard! By the way, enabling Slow Refresh is the optimal setting. Most FPM supports it. Does setting the HOST-to-PCI Post Write to 0ws cause issues? What about PCI Posted Memory Write and Burst Copy-Back to enabled?

Both banks of cache get warm? What is warm? If you can leave your finger on it for 20 seconds, I think it is OK. Since you are using double-banked cache, I think 2-1-1-1 should be OK. There was one motherboard manual, forget which now, which actually went into depth about cache banking, size, and FSB for determing the cache timings. I sort of recall a 40 MHz FSB with 512 KB double-banked wanting 12 ns, but I may be wrong and it may be chipset-dependent. I would think a single stick of 64 MB or 32 MB FPM and 2-1-1-1 with 15 ns should probably be fine.

For comparison, an IBM 5x86-133/2x gets these cachechk results:
L1 = 274 MB/s (read)
L2 = 102 MB/s (read)
RAM = 70 MB/s (read)
RAM = 93 MB/s (write)

Thanks feipoa! I made some changes in the BIOS.

Enabled Slow Refresh
HOST-to-PCI Post Write set to 0ws
1:1 HOST/PCI clock

The cache are not hot enough to burn when you touch but they feel very warm so I guess they are okay. Yes both banks heat up. I can also confirm the UM8810 board did not like my Chinese reprints (IS61C1024-10N) for single banked 512k cache operation.

I might have two different versions of Duke3d. Ill look into it... The copy on the Cyrix 133 machine plays the demo in the background from start menu and the copy on the CYrix 120 machine wont play the demo. (edit: it was missing demo1 and demo2 - Duke3d auto plays demo1.dmo when the menu loads after the intro)

My RAM, L1 and L2 speeds have increased!

feipoa wrote:

For comparison, an IBM 5x86-133/2x gets these cachechk results: L1 = 274 MB/s (read) L2 = 96 MB/s (read) RAM = 70 MB/s (read) RA […]
Show full quote

For comparison, an IBM 5x86-133/2x gets these cachechk results:
L1 = 274 MB/s (read)
L2 = 96 MB/s (read)
RAM = 70 MB/s (read)
RAM = 93 MB/s (write)

Regarding the Matrox Mystique 220 and with the new bios settings:

Without Univbe
◙ pcpbench 320x200: 23.1
◙ pcpbench 640x400: 9.4
◙ pcpbench 800x600: 6.2
◙ pcpbench 1024x768: 4.3

With Univbe 6.7 (5.1a detects oak chipset)
◙ pcpbench 320x200: 23.1
◙ pcpbench 640x400: 9.4
◙ pcpbench 800x600: 6.0
◙ pcpbench 1024x768: 4.3

The S3 Virge (not DX)

Without Univbe
◙ pcpbench 320x200: 23.1
◙ pcpbench 640x400: 10.5
◙ pcpbench 800x600: 6.8
◙ pcpbench 1024x768: 4.8

With Univbe 5.1a
◙ pcpbench 320x200: 23.1
◙ pcpbench 640x400: 9.4
◙ pcpbench 800x600: 6.2
◙ pcpbench 1024x768: 5.0

Interesting - it looks like the IBM 5x86-133/2x system with 66 MHz FSB, which as the cache settings set to the slowest (3-2-2-2), are the same as a 40 MHz system with cache settings set to fastest (2-1-1-1). The RAM on the 66 Mhz system is faster, however, the 40 MHz system will have the PCI running at 40 MHz, which will help it quite a bit in games.

And for the most part, UNIVBE made the cards yield worse pcpbench scores.

Did Duke3D play any faster with the Virge cards compared to the G200?

Also, what are your DOS Quake scores with -nosound -nocdaudio -nonet -nomouse -nojoy using timedemo 1?

feipoa wrote:

did Duke3D play any faster with the Virge cards compared to the G200?

I haven't done any testing with the Matrox G200 in the Cyrix 120MHz system. Its permanently in the Cyrix 133MHz system and Its the only G200 I have. The Cyrix 120MHz system will have to make do with the Mystique 220 as the card to represent Matrox during tests.

Cyrix 120MHz

Duke Nukem 3D 1.3D Full Version (800x600) : start new game (dnrate) then exit back to title for demo (escape key to remove title menu)

S3 Virge with Univbe 5.1a:
◙ completing first level: 9-22 fps
◙ demo2.dmo(scuba shark/water level): 7-18 fps

Matrox Mystique with Univbe 6.7:
◙ completing first level: 4-20 fps
◙ demo2.dmo:(scuba shark/water level) 8-18 fps

Matrox Mystique without Univbe
◙ completing first level: 11-20 fps
◙ demo2.dmo:(scuba shark/water level) 8-18 fps

Duke3D looks and feels pretty good using either video cards at 800x600 and they both seemed to average around 10-14 frames per second throughout testing.

<timedemo demo1> with Matrox Mystique
◙ DOS Quake @ 640 x 480 was an unplayable 6.1 fps
◙ DOS Quake @ 320 x 200 looked awful but ran fast 16 fps

I'm not sure whats wrong with my Quake but it keeps ignoring my requests for variables so I wasn't able to test with -height 640 -width 480 -nojoy -nonet -nomouse -nocdaudio from the command line or using a shortcut like I did on my other Cyrix machine.

I've purchased two more 3DFX Voodoo2's to test SLI in the UM8810. I have two free PCI slots in this board, so when they arrive I'll post the results from GLquake. I'm unsure if there will be any benefit since the CPU must be choking pretty bad.

I also switched PCI Posted Memory Write and Burst Copy-Back to enabled in the BIOS of the UM8810 board. (note this did not appear to increase cache speed or speedsys results)

My test bench area.

Those DOS quake timedemo results are for the Cyrix 120 or 133?

You should be able to run the commands like,

c:\games\quake_sw\Quake.exe -nojoy -nonet -nomouse -nocdaudio -nosound -wavonly

Once the game starts, press ` to bring down the command console. Type timedemo demo1
Press ` again to hide the command console.

Those results are from the Cyrix 120MHz in the UM8810 board.

c:\games\quake_sw\Quake.exe -nojoy -nonet -nomouse -nocdaudio -nosound -wavonly

yeah quake seems to just ignore my requests.

` works for timedemo.

How do you know it is ignoring the requests? Does the timedemo play with sound enabled?

Edit: You could always remove the sound card.

My bad, I was trying to use my GLquake command line for DOS Quake (height/width)

While speedsys CPU results can vary slightly and I find pcpbench results to vary by 0.2 points often, Quake 1.08 Server seems to be throwing out consistent numbers (5/5) and yielding the best results out of the different versions I have.

Cyrix 120MHz with UM8810 + Matrox Mystique

DOS Quake Shareware 1.06 with -nosound -nomouse -nojoy -nonet -nocdaudio +timedemo demo1

300x200 = 16.2 fps
640x400 = 6.9 fps
640x480 = 5.9 fps
800x600 = 4.2 fps

DOS Quake Full Version 1.08 with -nosound -nomouse -nojoy -nonet -nocdaudio +timedemo demo1

300x200 = 16.6 fps
640x400 = 7.2 fps
640x480 = 6.3 fps
800x600 = 4.5 fps

Cyrix 120MHz with UM8810 + S3 Virge

Without Univbe you can't choose 640x480 or 800x600+

DOS Quake Shareware 1.06 with -nosound -nomouse -nojoy -nonet -nocdaudio +timedemo demo1 + Univbe 5.1a

300x200 = 16.2 fps
640x400 = 5.5 fps
640x480= 5.9 fps
800x600= 4.2 fps

DOS Quake Full Version 1.08 with -nosound -nomouse -nojoy -nonet -nocdaudio +timedemo demo1 + Univbe 5.1a

300x200 = 15.5 fps
640x400 = 7.2 fps
640x480 = 6.3 fps
800x600 = 4.5 fps

DOS Quake Full Version 1.08 with -nosound -nomouse -nojoy -nonet -nocdaudio +timedemo demo1 + Univbe 6.7

300x200 = 15.5 fps
640x400 = 5.7 fps
640x480 = 6.3 fps
800x600 = 4.5 fps

SVGA modes for DOS Quake should probably be left to the Pentium class for true play-ability. GL Quake on the other hand is a completely different story, it looks amazing and plays at speed with sound in SVGA modes. I'm looking forward to the Voodoo 2 SLI test results when they arrive.

Wowah! Why did you decide to bench with 640x480? It is standard to bench at 320x200 and use 1.06 demo. I noticed that 1.08 yields slightly faster results, however for comparative purposes, most users are using 1.06 demo so it is best to stick with that.

Ive added the tests that you mentioned, thanks!

So you get,

Cyrix 5x86-120+Mystique
16.s fps

Cyrix 5x86-120+Virge
16.2 fps

but what about the Cyrix 5x86-133?

I am assuming branch prediction and the other goodies are enabled.

Cyrix 5x86 120MHz @ 40MHz FSB + Matrox Mystique 220 (MGA-1164SG) (Quake.SW.106)

300x200: 16.2 fps

Cyrix 5x86 120MHz @ 40MHz FSB + Matrox Millenium II(MGA-IS Storm R2) (Quake.SW.106)

300x200: 17.1 fps

Cyrix 5x86 133MHz @ 33MHz FSB + Matrox Millenium G200(MGA-G200) (Quake.SW.106)

300x200: 17.1 fps

DOS 7.10
Himem.sys loaded
DOS Quake with -nosound -nomouse -nojoy -nonet -nocdaudio +timedemo demo1

BTB_EN=ON, FP_FAST=ON, LSSER=OFF for Cyrix enhancement gains and all else OFF in the Peter Moss utility for stability throughout all tests.

OK, thanks for checking that.

On my Cyrix 5x86-133/4x with Voodoo3, I get 17.8 fps.
On my IBM 5x86-133/2x with G200, I get 19.8 fps.

The difference on my two systems is puzzling. I was not expecting such a gap. I figured your 133/4x system would have at least equalled mine, but perhaps the Voodoo3 is that much faster? Or maybe it is the SiS chipset.

This hardware does not work with my ECS UM8810P-AIO motherboard but works fine with the Biostar MB-8433UUD.

[PCI Video Card]

Rendition Verite 2200 : red light, wont turn on.
ATI RAGE 128 VR : red light, wont turn on.
ATI RAGE 128 PRO : red light, wont turn on.
Matrox MGA-G200A Quad : green light, black screen. ( doesn't work in MB-8433UUD or UM8810P-AIO)

[RAM]

Samsung KM44C16100AK-6 (1 x 64 MB module) : red light, wont turn on

[CACHE]

IS61C1024-10N (single banked 512K operation) : hard crashes during boot / testing.

◙ Cyrix 5x86 120MHz
◙ 512K Cache
◙ 64MB FPM RAM
◙ ECS UM8810P-AIO
◙ Matrox Millenium-2MB
◙ 10GB IDE Hard Disk
◙ IDE CDROM

So I'm not sure which sound card to use. I need a sound card that will play well with different MIDI devices and I would like it to work well in GLQuake/Warcraft 2/Doom 2/Hexen/3DFX games in Windows 95. Operating System of choice is Windows 95C.

I catalogued all of the sound cards I have available:

◙ CT1740 SB16
◙ CT1750
◙ CT1750 SB16.MCD
◙ CT1910 Wave Blaster
◙ CT2770 SB16.VALUE
◙ CT3620
◙ CT3670
◙ CT4810
◙ CT4830 SB.LIVE

◙ SB0280 SB.AUDIGY2
◙ SB0060 SB.LIVE.5.1

◙ YAMAHA 718-S
◙ ESS AUDIODRIVE ES1868F
◙ AVANCE LOGIC (unknown)
◙ AUREAL VORTEX AU88E0R

Can you recommend a sound card for this system?

Where is there a red light which turns on when you install the Samsung KM44C16100AK-6 module?

Did you ensure that the same physical IS61C1024-10N modules work properly in the MB-8433UUD? You need to use known working pieces. I have an SRAM tester and found a 10% failure rate in the IS61C1024-10N modules. Without the SRAM tester, it was almost impossible to single out the exact faulty module.

Does the UM8810P-AIO board work with any 64 MB sticks? Or do you need to use 2x32MB?

Does the Voodoo2 work in the UM8810P-AIO board?

I am not the best sound card guy to ask about those sound card options. I don't think I would use any PCI sound card because they will be running at 40 MHz and DOS games work best with ISA cards. Generally, I stick with the AWE64 Gold/Value or use an Audican32 plus (Yamaha 718-S) with a MIDI daughter card (NEC XR385). This way you get OPL3 and general midi. The issue with the Audican32 is that the card is not full height, so you need to make a special interface cable for the XR385 and offset the XR385. Seems that some people like the ESS AUdiodrive cards. Maybe look for some sound recordings. I have the ES1868 in my 386, but I haven't really compared it against the 718-S. CT3670 and CT3620 are AWE32. I do recall reading many people liking these but I do not recall the reasons. There are ample threads on sound card comparisons.

feipoa wrote:

Where is there a red light which turns on when you install the Samsung KM44C16100AK-6 module?

I mean the red light on the monitor that changes when it receives a signal from the video card.

With that RAM module in the light stays red and the mobo doesn't turn on.

feipoa wrote:

Did you ensure that the same physical IS61C1024-10N modules work properly in the MB-8433UUD? You need to use known working pieces. I have an SRAM tester and found a 10% failure rate in the IS61C1024-10N modules. Without the SRAM tester, it was almost impossible to single out the exact faulty module.

I have no way of knowing for sure as they are spare modules I've never used.

feipoa wrote:

Does the UM8810P-AIO board work with any 64 MB sticks? Or do you need to use 2x32MB?

That's the only 64MB module I have and it doesn't work. I use 2 x 32 MB Toshiba modules.

feipoa wrote:

Does the Voodoo2 work in the UM8810P-AIO board?

Yeah it works fine.

feipoa wrote:

I am not the best sound card guy to ask about those sound card options. I don't think I would use any PCI sound card because they will be running at 40 MHz and DOS games work best with ISA cards. Generally, I stick with the AWE64 Gold/Value or use an Audican32 plus (Yamaha 718-S) with a MIDI daughter card (NEC XR385). This way you get OPL3 and general midi. The issue with the Audican32 is that the card is not full height, so you need to make a special interface cable for the XR385 and offset the XR385. Seems that some people like the ESS AUdiodrive cards. Maybe look for some sound recordings. I have the ES1868 in my 386, but I haven't really compared it against the 718-S. CT3670 and CT3620 are AWE32. I do recall reading many people liking these but I do not recall the reasons. There are ample threads on sound card comparisons.

Yeah, I would use the AWE64 GOLD but don't have a spare. I thought I could find something useable in the pile I have on hand.

rgart wrote:

feipoa wrote:

Did you ensure that the same physical IS61C1024-10N modules work properly in the MB-8433UUD? You need to use known working pieces. I have an SRAM tester and found a 10% failure rate in the IS61C1024-10N modules. Without the SRAM tester, it was almost impossible to single out the exact faulty module.

I have no way of knowing for sure as they are spare modules I've never used.

Only way to be sure is to remove the modules from the Biostar board and test them in the UM8810P.