Either way, my board is still behind on scores... If I could get a decent motherboard to post at 150MHz or higher, then I'd be all set for maxing out my benchmarks... as it is, I personally can't remember anything else scoring higher than 80 on speedsys with anything less than a Pentuim chip.

I've been wondering about the overall stability of the computer so I put it through several tests while @133 and discovered that WIN3.11 wouldn't start anymore. I instantly blamed the motherboard (afterall the CPU has proven itself at higher than 133 before) and started to dissemble the computer when I discovered that the power connector had started to detach itself from the hard drive. It would appear that some of my issues are data corruption related now. A fresh install should clear somethings up.

A bit off topic, I think we discussed the SiS Soyo 4saw2 motherboard (my first favorite 486 board) in the past as a possibility for a benchmark SiS. I just wanted to warn you that it has a list of problems of it's own.

Though it does in fact have a working PS/2 mouse connector and mine has a PS/2 keyboard connector, it lacked performance in speedsys when benching my Cyrix based chips. I don't have the benchmarks on hand, but they were several points lower than even my Shuttle HOT-433 scores (if I get it set up again, I'll post them)... the PCI bus seemed slower on the board and it didn't like posting at 512KB (took several tries with different chips). My biggest gripes against it were that due to certain jumpers being soldered over, I couldn't test it for 60/66/83 MHz bus speeds and I can't seem to get the CPU multiplier to work. Just figured that I'd throw my two cents worth in for that particular board. Although if you can work around some of those issues, it might be a nice performing board.

Side note: I managed to get the new setup to boot at 150MHz, but locked up before I could get any benchmarks done... hope still abounds.

Perhaps one day, I really will stand a chance at building my own 486 board, and all my issues will melt away... and of course make room for new ones, because fixing issues is half the fun. 😉

nemesis wrote:

I've been wondering about the overall stability of the computer so I put it through several tests while @133 and discovered that WIN3.11 wouldn't start anymore.

For how long had the system been running before it started to fail? 10% of the time, my IBM 5x86c-133 will fail after being powered up for more than 2 hours. What if you let it sit for an hour then try the Win 3.11 tests again? Are you using heatsink compound? What CPU voltage? You may need to do a quick mod to allow for variable voltages. I find the 3.70 - 3.85 V range to be good, with some CPU's doing fine at 3.7, while others requiring a little more juice at 3.85 V.

nemesis wrote:

I think we discussed the SiS Soyo 4saw2...it lacked performance in speedsys when benching my Cyrix based chips.

Do you have a Voodoo3 3000 to plop in there? Speedsys is just one benchmark, ideally you'd want to run a collection of benchmarks to check for the average performance. This will also provide some insight as to intermediate-term stability.

nemesis wrote:

My biggest gripes against it were that due to certain jumpers being soldered over...

That's fine. I find those soldered jumpers can be removed and jumper headers soldered in. I've had multiple motherboard success with this modification.

nemesis wrote:

it didn't like posting at 512KB

Perhaps a jumper wasn't set right or your cache was bad/incorrect? Try 4 pieces of those 1024, 10 ns modules and a different TAG (10 ns).

For how long had the system been running before it started to fail? 10% of the time, my IBM 5x86c-133 will fail after being powered up for more than 2 hours. What if you let it sit for an hour then try the Win 3.11 tests again? Are you using heatsink compound? What CPU voltage? You may need to do a quick mod to allow for variable voltages. I find the 3.70 - 3.85 V range to be good, with some CPU's doing fine at 3.7, while others requiring a little more juice at 3.85 V.

It had been running about 6 hours. That doesn't appear to have been the issue though. I might eventually have to mod the voltage like you showed me in a few posts to see if that's the issue though.

The reason it failed in Win3.11 was because the harddrive power had become partially disconnected and corrupted it... reinstall seems to be working.

nemesis wrote: […]
Show full quote

nemesis wrote:

I think we discussed the SiS Soyo 4saw2...it lacked performance in speedsys when benching my Cyrix based chips.

feipoa wrote:

Do you have a Voodoo3 3000 to plop in there? Speedsys is just one benchmark, ideally you'd want to run a collection of benchmarks to check for the average performance. This will also provide some insight as to intermediate-term stability.

No Voodoo 3 PCI cards 🙁.... but I do have a Matrox G200 32MB on it's way that I might be able to test in the Shuttle HOT-433 and Soyo 4SAW2 boards.

Hopefully I'll have a fresh set of benchmarks to display when it arrives.

nemesis wrote:

I do have a Matrox G200 32MB on it's way that I might be able to test in the Shuttle HOT-433 and Soyo 4SAW2 boards.

Are you referring to the Matrox Millennium G200 32MB Quad display card, or the single display Matrox MIllennium G200 16 MB (contains 8 MB upgrade module)? I've had trouble getting the Quad card to work reliably on my UMC 486.

It is in fact the Quad display. I guess I'll find out soon enough if it will work.

I'm at a much more stable 75.53 Speedsys score with loop and rstk off and BTB on. Though I had an issue with installing Doom after running the computer for a few hours.

At 133 Mhz, I have in my test log that RSTK should always be off, irregardless if BTB is on or not. If BTB is on, you should also disable LOOP and BWRT. From my test log, these are the two configurations I run,

IBM 5x86c-133 (2 x 66 MHz) [step0rev5]
Enabled: LSSER (=0), LOOP_EN, BWRT, FP_FAST, LINBRST, MEM_BYP
Disabled: BTB_EN, RSTK_EN

Register/Index HEX DEC
[PCR0] 4 4
[CCR1] 02 2
[CCR2] D6 214
[CCR3] 1C 28
[CCR4] 38 56

IBM 5x86c-133 (2 x 66 MHz) w/Branch Prediction under DOS/Windows [step0rev5]
Enabled: BTB_EN, LSSER (=0), FP_FAST, LINBRST, MEM_BYP, DET_E
Disabled: LOOP_EN, RSTK_EN, BWRT

Register/Index HEX DEC
[PCR0] 2 2
[CCR1] 02 2
[CCR2] 96 150
[CCR3] 1C 28
[CCR4] 38 56

DET_E may show improved stability at 133 MHz when disabled (depending on the particular CPU), in which case [CCR4] = 28 HEX, 40 DEC

I've never even touched the HEX DEC sections or the BWRT (is that one on by default?), and I've almost always turned off LOOP and RSTK now that I've discovered that they slow down when branch prediction is on.

I have a whole new set of benchmarks to run now.

I think you are confused; BTB, LSSER, FP_FAST, etc, are all part of sections of the registers: PCR, CCR1, etc. What I have shown is a table, whereby PCR0, CCR1, etc are the Register/Index, and the numbers to the right of the specific register are the respective HEX (hexidecimal) and DEC (decimal) values for each whole register (there are 8 features per register and 11 registers in total).

This is difficult to explain in words and you should probably refer to Cyrix 5x86 Register Enhancements, particularly the table on page 2. You'd want the HEX and DEC values if you are using the Evergreen Technologies program to enable features, which has a few more options than the Peter Moss Utility. You also may want to use the IBM Dos GUI utility, which also has more features than the Peter Moss Utility. Again, refer to the tables on page 2 and 4 of Cyrix 5x86 Register Enhancements Revealed, as well as the tables in Appendices 2-5.

You may want to read through Cyrix 5x86 Register Enhancements Revealed. Most features, including BWRT, are benchmarked already with charts. In the best case benchmark (LINPCOD), BWRT showed a 6% improvement. The charts contained therein go through the features one-by-one. The PDF file uploaded to the that thread is a lot easier to follow than the thread text.

BWRT is disabled by default. If memory serves me correctly, BWRT will hang-up HOT-433 boards... or was that LINBRST, or both? PC CHIPS M919 boards should be fine with these two features.

feipoa wrote:

I think you are confused; BTB, LSSER, FP_FAST, etc, are all part of sections of the registers: PCR, CCR1, etc. What I have show […]
Show full quote

I think you are confused; BTB, LSSER, FP_FAST, etc, are all part of sections of the registers: PCR, CCR1, etc. What I have shown is a table, whereby PCR0, CCR1, etc are the Register/Index, and the numbers to the right of the specific register are the respective HEX (hexidecimal) and DEC (decimal) values for each whole register (there are 8 features per register and 11 registers in total).

This is difficult to explain in words and you should probably refer to Cyrix 5x86 Register Enhancements, particularly the table on page 2. You'd want the HEX and DEC values if you are using the Evergreen Technologies program to enable features, which has a few more options than the Peter Moss Utility. You also may want to use the IBM Dos GUI utility, which also has more features than the Peter Moss Utility. Again, refer to the tables on page 2 and 4 of Cyrix 5x86 Register Enhancements Revealed, as well as the tables in Appendices 2-5.

You may want to read through Cyrix 5x86 Register Enhancements Revealed. Most features, including BWRT, are benchmarked already with charts. In the best case benchmark (LINPCOD), BWRT showed a 6% improvement. The charts contained therein go through the features one-by-one. The PDF file uploaded to the that thread is a lot easier to follow than the thread text.

BWRT is disabled by default. If memory serves me correctly, BWRT will hang-up HOT-433 boards... or was that LINBRST, or both? PC CHIPS M919 boards should be fine with these two features.

Confused? That's an understatement.
Well, not so much now. Thank you for the reference material, and the hint on using the Evergreen Utility. I'll have to dig that program up sometime soon.

The DOS version of the Evergreen 5x86 program requires HEX units while the Windows NT version requires DEC units.

Thus far I haven't changed any settings, but I did finally get a Quake 1.06 timedemo demo1 score: 12.2!

As low as that is, my friends have demonstrated their Pentium 100s and 133s and yielded almost the same scores (slightly lower with the 100 and slightly higher with the 133). I'm looking forward to seeing the scores after I tweak it a little.

Ouch, that's Intel DX4-100 speed.

I'd like to see you get 18.4 fps.

feipoa wrote:

Ouch, that's Intel DX4-100 speed.

I'd like to see you get 18.4 fps.

Yeah, I'm hoping that it's just the video holding me back. They (my friends) had very weak video cards as well when they benched theirs.

I'm expecting my new video card this week, and if it works with the motherboard, it should reveal whether the video memory speed issues pertain to the video cards or the motherboard itself.

Link to the Evergreen 5x86 utility can be found online here,
http://homepage3.nifty.com/sandy55/Program/et9603.exe

To the best of my knowledge, the Evergreen utility is the only NT driver available for the Cyrix 5x86 features. Luckily, it also works in Windows 2000.

I know I'm getting off tracka bit here, but I did a very small comparison between the Soyo 4saw2 compatibility and the Hot-433 rev4 video card list. I was able to get every video card that works in the Hot-433 to post in the 4saw2 and also was able to use one of the CF cards that didn't work in the HOT-433 as a HDD.

The ATi Rage 128 also worked fairly well (some screen flicker during boot up) in the 4saw2 and as I type this, I'm trying to set up an OS on the HDD to run benchmarks.

Sadly, you were right about the matrox quad not working on the UMC and it also didn't work in the SiS.

EDIT: A bit more on track, I was able to begin using my Voodoo 1 in the HOT-433 again without issues so far. It seems that it was impacted somehow by the corrupt HDD. Not sure how, though.

Try to source a Matrox Millennium G200 8 MB card and the additional 8 MB expansion module, for a total of 16 MB of video ram. If you at least have this card, you can directly compare your systems to the results obtained in the Ultimate 486 Benchmark Comparison. This is a good first step for troubleshooting slow or buggy motherboards. I figure most 1996-era UMC boards should perform similarly to the UMC board of the U4BC.

feipoa wrote:

Try to source a Matrox Millennium G200 8 MB card and the additional 8 MB expansion module, for a total of 16 MB of video ram. If you at least have this card, you can directly compare your systems to the results obtained in the Ultimate 486 Benchmark Comparison. This is a good first step for troubleshooting slow or buggy motherboards. I figure most 1996-era UMC boards should perform similarly to the UMC board of the U4BC.

I finally found a Matrox 8 MB card at a price that I could eventually forgive myself for paying and bought it... I had to get the expansion memory seperatly, though. Thanks for the clarification.

In other news, I changed the timings to 2/2 from 3/3 and, sadly, my speedsys score dropped to a 75.47 but my Quake FPS jumped up to 14... makes me think about my FIC 503+ build OC issues.

Are there any more folks doing "super socket 3" builds? I haven't seen very many on Vogons so far.

I have begun to run some benchmarks on the new video card and decided again to start experimenting with certain BIOS options that previously have caused crashes. To my delight, PCI burst write now works (I'm thinking that I might have screwed up some of the other settings before and thought that it was the PCI burst). My speedsys score is still ~75.5 (the numbers there just don't feel right) but my 3dbench2 score is now over 90. My Quake 1 timedemo only increased to about 15 FPS, though. Overall I'd say it's an improvement.

I'm still waiting on my 8MB modules, so this is only an 8MB card (not that I think the memory size is making a big difference with these benchmarks).

Speedsys now reports my video memory speed at a little over 48MB/s.

EDIT: Matrox G200 runs at 60MB/s on a 66MHz FSB... In case anyone didn't know if it could POST at 66MHz. I haven't done any long term tests though.