Got it.
I usually try to keep test results as compact as possible otherwise we start mixing things up.
Thanks for the clarification.
This board works best s3 trio64 vlb graphics cards indeed.

GOOD NEWS ABOUT THE HOT-433!

I am pretty sure to have found the problem of instabillity of this board.

As already told in the past, I have some of those boards here (more accurate: 3 pieces of ver. 1 with support of FPM-RAM only, and 1 piece of ver. 4 with support of EDO-RAM).

All of them have big stability-problems at FSB over 40 MHz - this was my assumption.

The instability means a really MASSIVE drop of the CPU-performance in WIndows (few minutes to about a half hour after loading it).

Some weeks ago I wanted to build a "standard" fast 486-system using an AMD 486 DX5-133 - nothing special itself.
And I had to wonder - the system was still unstable exactly the same way even at a FSB of 33 MHz...Look at that...

Tried all 4 of my boards with some different CPUs of the same kind and using different L2-modules (brands and amounts of 256 KB/512 KB/1 MB) - but the problem remains.

Knowing all of this, I concentrated now of the RAM - AND THIS WAS THE PROBLEM!

For all 4 boards, the problem occurs ALWAYS if modules of 64 MB are used! Regardless of the brand, chip-organization, BIOS-timings, and regardless if EDO-RAM or FPM-RAM (EDO-RAM only for ver. 4 because of its northbridge). All of the 64-MB-modules shown in the picture causes the problem in any possible combination with any board at 100 % !

The more irritating thing is: if there is EDO-RAM used together with an UMC-northbridge onboard not supporting EDO-RAM (without "E**" in the second line on its imprint), there is still a boot possible and the board SEEMS to be working good - but various failures and errors occur in Windows (except of the performance-drop of the CPU). So please do NOT use EDO-RAM in this case. Other 486-boards without support of EDO-RAM just not boot if EDO is inserted - but this is not the case here with this board, so more attention have to be paid. On the other hand, using EDO-RAM on my board ver. 4 (having an E-imprint) does not take a performance-advantage compared to the other boards ver. 1 using FPM-RAM (at the same RAM-timings in BIOS).

So, back to the RAM-modules, it was now only necessary to try modules of 32 MB or 16 MB - AND THE PROBLEM DISSAPPEARED! Again, some different brands of EDO/FPM-RAM of 60/50 ns and 32/16 MB were tested - NO more problems in any possible combinations on any board (of course, EDO-RAM only with the appropriate board).

Now, the new goal was to find out, if some board is stable at a higher FSB.
And this was the case for the board rev. 4 and at least one board rev. 1 (not all of them tested for now) - full stability at 60 MHz FSB!
For the board with rev. 1, the PIO-mode has to be decreased from 4 to 3 (not necessary for the rev. 4).

So it was now possible to build a very fast 486-system using a Geforce 2 MX-400, an AMD 486DX5-133 @ 180 MHz/5 Volt and 32 MB FPM-RAM with 50 ns together with a HOT-433 Rev. 1 having 1 MB of L2-cache at timings of 3-2-3. Its very impressive, that a full amount of 1 MB cache is possilble at this FSB! Sadly, as they are "standard" L2-modules used, no better/faster timings are possible...The rest of the hardware: Creative Soundblaster 16 Value PNP (Vibra16C, CT2960) and Opti FireLink 82C861 (USB-Controller 1.1).

The results are:
- QLQuake: 27,5 FPS (best one)
- Quake II: 16,5 FPS
- doom (DOS): 68 FPS (1097 realticks)
- 3dbench 1.0c (DOS): 101,5
- pcpbench (DOS): 30,1

For instance:
- the same VGA and CPU run pretty fine together with the board rev. 4 (32 MB EDO-RAM with 50 ns), but its maximum L2-cache of 512 KB results in a reduction of about 0,3 FPS in GLQuake and Quake II
- there is NO difference in GLQuake or Quake II using a Geforce 2 MX-200 (video-memory-bus at 64 bit) compared to the Geforce 2 MX-400 (video-memory-bus at 128 bit)
- using the Radeon 9250 shown here (3 (+3 more) retro battle stations) results in very low FPS (about 15 FPS for GLQuake), so this is not a really option
- no run of 3DMARK 99MAX with the detonator-drivers 6.31, 8.05 and 12.41 because of Direct3D-problems in generally

And yes, both mainboards are really stable - last week they were on for 2 days for about 10-12 hours per day. Each few hours a test of GLQuake and Quake II was done, with no one performance-drop and no one error.

In the next days I will try to found out, if it is possible to use more than one piece of 32 MB RAM and if so, how this affects the results.

next pictures

next pictures

next pictures

and the last pictures for now

Very cool. Will link your post in the dictionary at the beginning of the thread.

I don't remember if i tested with 64Mb modules the HOT-433 V4 board here, at least cannot remember hitting such profound issues.
The notes i left in the thread are all about 32Mb EDO modules, so this is an interesting finding you got.

I think it is mandatory to use extension IDE/SCSI cards in combination with this UMC chipset to offset its weak IDE.

One interesting thing i am noticing the PC Player Benchmark result is very good.
Maybe because of the larger L2 cache buffer of 1Mb. What is the number you get with the V4 mobo and 256Kb L2 ?
I found that smaller buffer at higher timings is in general better than bigger one with relaxed wait states.

Did you run more complex tests, or only games for now ?
Asking because i remember that gaming can be done with all settings on max, except CACHE SPEED OPTION = 2-2-2 (best is 2-1-2), but for complete stability with harder grinding i had to slow the V4 board down to
CACHE SPEED OPTIONS = 3-2-3
DRAM READ WAIT STATES = 1
DRAM WRITE WAIT STATES = 1

Didn't see any notes about CPU voltage. What was it ?

-

As shown above, the CPU is running at the maximum 486-voltage of 5 Volt.

Sadly, I still do not have any appropriate L2-modules to build 256KB/512KB/1MB with better/faster timings than 3-2-3 at 60 MHz FSB.
For all modules I have, any attempt to reduce the timings even at 3-1-3 (@ 60 MHz) results into a freeze of the system during the first or second POST-screen. It's a really big problem for my L2-modules, so no chance for now...

Yes, an external HDD-controller would help, if some more or larger HDDs (without using of a HDD-manager) are needed.

About more complex tests: believe me, in the past I tried some times to run 3D Studio R3 on different 486-/386-systems, but after the software is loading, no control via mouse or keyboard is possible (except of some few buttons on the keyboard).
Maybe I am the first one user encountered the problem, but this is the really reason for me not using it...And I still remember, I asked about this problem few years ago here in the forum, but no one could help. I don't know what goes wrong here.

To be honest, I am not (and never was) a big PC-gamer in generally - it's simply not my thing. So my collecion of old PC-games is really not big. Maybe I will try to install and test Wolf 3D, if I can found it (btw, can it be downloaded somewhere here on vogons?).

The next days I will test the HOT-433 ver. 4 with 256 KB L2 at DOS as you suggest, but for the same frequency the L2-timings will be still 3-2-3 (as explained above).

One more question - is there a database/comparison of results for Quake II for the old systems in Win98/ME?
The Quake II result of 16,5 FPS was the best one 486-result for me until now, but maybe they are still much better ones.

Missed the 5V to CPU note in your previous posts. Thanks for clarifying it.
How long it took you to find the magical processor ?
Since you didn't mention Peltier cooling i assume you are on fan+heatsink.
There is a chance this causes you subtle problems.
I had issues like that. The CPU is not fully stable but the L2 cache appears to be misbehaving.
Swapping with a better chip, or Peltier-ing it suddenly fixes the cache, etc.
If you can - slap Peltier on top to establish a stable base and then walk back from there.

Yeah, DIP-28/32 chips don't handle well such overclocks.
256Kb L2 cache is possible with tight wait states, but so far was not able to achieve anything notable with 512 and 1024 Kb sized buffers.

First time hearing about mouse related issues in 3DS for DOS.
Apologize for the obvious question, but do you have a decent mouse driver loaded with the mouse working in other programs ?

Wolf3D test is included in this set of benchmarks.
I got it from another Vogon user, but cannot remember who.

1cd mrk/bench/3d/wolf3d
2test

Not aware of Quake II tests database that includes results from 486 class hardware.
Even Quake 1 and GLQuake are too much for the poor 486es.

Gonzo,

I did compare some faster 486's as part of my 686 benchmark run. However, Quake2 in OpenGL mode is run with a much slower Matrox G200. IBM 5x86c-133/2x and Am5x86-133 results are present. Re: The Ultimate 686 Benchmark Comparison

For GLQuake (not Quake 2) scores on a Voodoo with various 486 processors, are shown here: Re: Voodoo 1 vs. Voodoo 2 on a 486 scroll down.

I should re-run the 486 benchmarks, adding more high-end systems like the Am5x86-180, more games, and 386 systems upgraded to 486 systems. However, other things interest me more at the moment.

Concerning your EDO comment. I have two version 1 HOT-433 boards and I know that at least one of them, with a BCA suffix on the Northbridge, runs fine with EDO sticks. I think I used to run 4x32 MB. What are the suffixes on your HOT-433 v1 boards? My other v1 board has BTS.

I concur with pshipkov concerning operation at 180 MHz. If you think you are reaching stability at 180 MHz and 5 V with a single 32 MB stick, it will unlikely be stable when you add more RAM sticks (and keep tight timings) and place the motherboard inside an AT case. I have tried at least a dozen ceramic topped Am5x86 chips, even those suspected to be stable at 180 MHz without peltier, but in the end, I found a late QFP variant to be more reliable if you are not using a peltier.

Thank you for your answers, pshipkov and feipoa.

As I can see, the Quake II results of the HOT-system above are very good.

Attached they are pictures from the other 2 boards HOT-433 rev.1 I have here.

I am absolutely sure, I do not have any overheating-problems with my AMD 486-CPUs at 4/5 Volt and 180/200 MHz.
Attached you can see how I cool them (3 pieces of 180 MHz - 2 @ 5,0 V and 1 @ 4,0 V; 1 piece of 200 MHz @ 5,0 V) .
The cooler-dimensions are similar for the 4 CPUs, and the fans for the 180 MHz are similar, too (dismounted from some Mini-ITX-systems with Pentium M many years ago, and very fast rotating).
For the CPU @ 200 MHz I use a fan-BOMB (see pics 9 to 11), dismounted from the back of some old server-racks long time ago (extremly fast rotating and very loud, but who cares...).

All of the coolers stay "normal" warm (slightly warmer than my fingers touching them) all the working-time of the CPUs. Switching the system off and immediately dismounting the CPUs, the bottom of the CPUs is "normal" warm, too (so there is in reverse logic a good contact between the CPU and the cooler).
And, finding once the appropriate boundary conditions (L2, RAM, CPU-voltage, etc.), all the CPUs run absolutely STABLE, using exactly the same cooling conditions.

So at this single point I just do not see any reason for using a "laborious" peltier...

Now I risk to be outrageous, and ask you, if there is possible to increase the CPU-voltage slightly over 5 Volt...If I am lucky to have such CPUs here, maybe the one piece running at 200 now can do more than 200 MHz...??
Or, maybe the CPUs running at 180 MHz can do 190 (or anything else between 180 and 200)??
To make it a little easier, we can stay by the HOT-433 rev. 1. Maybe you have an idea how to increase the voltage, and how to change the FSB at unusual values (e.g. to 55 or 70 MHz, or similar crazy values).

next pics

and the last pic

Ok, so you have a magical CPU if it can run at 180 with the tiny heatsink and fan. Congrats.

Can you supply more than 5V to CPU ?
Draw from the 12V line, add resistors to lower to 5++V, but they will get HOT, so cooling will be needed.
You will need to insert a PGA169 socket and cut some of the pins and route them to the 5++V line you will create.
If you are eyeing 2 x 83V - this usually requires more relaxed wait states to get it under control, if at all, but Shuttle HOT-433 V4 can do decent job at it actually, so ...

Pshipkov, would you maybe do a drawing (a sketch) of your explanations above about to how to increase the 5 Volt? That would be really very helpful.

Another idea is to generate a voltage between 4,0 and 5,0 Volt for the one CPU here doing 180 MHz @ 4,0 V, but not booting at 5,0 Volt (to test if it can do more than 180 MHz by adding of + 0,xyz Volt)

Today the test of the HOT-433 rev. 4 with 256 KB L2 (timings of 3-2-3), GF 2 MX-400 and 32 MB EDO (all items shown above) is over.
The results are:
DOOM: 61,4 FP (1216 realticks)
Quake (DOS): 18,8 FPS
PNPBench /vgamode: 26,8
3dbench 1.0c: 93,3

And, I measured the temperature of the cooler during the DOOM-test (the measuring surface for the red spots of the device is maybe not very optimal, but the "tendency of heat" can be seen anyway, I think).

and the rest of the pics

gonzo, I think pshipkov is referring to using a cheap buck regulator, like that shown here: Re: Custom interposer module for TI486SXL2-66 PGA168 to PGA132 - HELP! , to run the CPU off the 12 V power supply, then have the voltage stepped-down using the buck regulator. You can see what I did in the above link to run my SXL2 at 5.25 V instead of 5.0 V.

In short, you will want to find an unused PGA-168 socket. Cut off all the pins on that socket which correspond to the incoming voltage (Vcc) from the motherboard. There will be about 23 pins. You are cutting them off so that you no longer power your CPU from the motherboard's CPU voltage regulator. Instead, you are powering it from the 12V line via the switching (buck) regulator. Connect the input of your switching regulator to the 12 V rail on the PSU (via molex), then connect the output of your switching regulator to the cut-off Vcc pins on your PGA-168 socket. Use the trim pot on your switching regulator module to adjust the voltage, anywhere from about 3 V to 11 V. I'd suggest setting it for 5 V to start. If you need greater precision, you could install one of those blue Bourns trimmers like I show in the above link. Sometimes the trim pots that come on these cheap buck regulators don't work well at all. Mine was faulty.

You should never try to regulate a voltage with resistors alone, it will vary with current few m draw, so in idle you might be looking at close to 12V on the supply, if you run to 5.xx under full load