Any Speedsys score of less than 75 on an IBM/Cyrix 5x86-150 is probably related to the employed L2 cache settings and DRAM wait states as setup in the BIOS. If your L2 cache settings and DRAM wait states are fast, like 2-1-2 and 1ws/0ws (or 0/0) and you still get a score of 68, then use a DOS clock frequency testing app. to determine if you are really at 3x50 MHz. maxspeed.exe comes to mind (note, it is the irregularly occuring speed measurement which seems to be correct). You can also use chkcpu16.exe and cachechk version 4 to determine if the correct FSB and multiplier are being used.

You are surely not running the CPU at 2x50 MHz as you'd get a speedsys score of around 55. A score of 68 is more like a Cyrix 5x86-120, so you'll want to confirm which FSB is being used (with the above mentioned 3 programs).

If it was lack of voltage/stability, I think the computer would freeze-up or you'd get erroneous results.

EDIT: What hypothetical voltage did you apply to this hypothetical 5x86-150? From some of my earlier work, I know 5V will not be long-term stable. I also know that 4.1V is insufficient (at least for the CPUs I tried). That said, I would advise 4.5V as a good starting point for this hypothetical configuration.

I have found that a fairly decent stability test is to play 256kbps mp3s on repeat in Winamp 2.95. Win98SE and W2K seem to be a lot more crashable than WinNT 4.0. Then ensure that Ziff-Davis Winbench99 - FPU WinMark will complete. This method has proven to be a better stability test than CPU Burn-In and CPU Stability Test.

A sigh of relief. According to maxspeed.exe, the operating frequency was 120 (the screen kept flashing 121 and 138 so I'm pretty sure it was at 121).
I haven't run any other tests so far aside from 3dbench2 and that yielded a score of 70.5 and a speedsys score of 68.67.

When I changed the timings to 2-1-2 or 2-2-2 and the ws to 0/0 it locked up on the POST. At least I have something to go on now. By the way, the tests were run with 3-3-3 timings and 3/3 ws.

What hypothetical voltage did you apply to this hypothetical 5x86-150?

Highest voltage that I would test it with was 4v. Anything lower would sometimes not boot. Of course we now know that your suspicions were correct in that it was in fact 120 MHz and not 150.

Do not rely on only 1 test. You'll want to test with MaxSpeed, ChkCpu16, and Cachechk v4.0. Maxspeed should never be trusted on its own right, but only used when ChkCpu16 and Cachechk v4.0 yield strange results. Do not use Cachechk v7.0 for Cyrix 486 processors.

I have found that when MaxSpeed flashes between 121 and 138 on my Cyrix 5x86-133, that Maxspeed is only good to a value between 121 - 138 MHz. EDIT: For example, please refer to the images on page 3 of this post,
The World's Fastest 486

I am pretty sure you cannot realize 150 MHz at 4.1V or less. 5 V will work for about 3 minutes, but it is thermally baking the CPU. To get a Cyrix 5x86 working at 150 MHz, you'd need to add voltage adjustment between 4.0 and 5.0 V and get crazy with cooling.

It also helps if you freeze the CPU for 20 minutes prior to testing. I'd really like to get just one Cyrix 5x86-150 Speedsys result. I have a PLL with frequency variation on it which will be coming in, so I might be able to try intermediate speeds between 133 -150 MHz. Or it might be a total flop.

Uploaded with ImageShack.us
I hope you find this at least mildly interesting. I couldn't get any readable pictures of speedsys due to being stuck with my crappy old camera for now, but speedsys identified the CPU as 120MHz and the memory speed as ~30MB/s. The RAM is 16MB and as far as I can tell, EDO. The speedsys CPU benchmark was ~69.
Maxspeed kept almost solid at 138 MHz but very infrequently flashed 121. Also any tests associated with the memory seemed to hang for a few seconds before completing. I tend to think this is due to some problem with the memory, but I'm not expert on these vintage beauties.
Any thoughts would be appreciated. 😀

P.S. I had the cache timings set to auto, which was 3-2-3, WS set to 2, and pci divider set to 1:1/2.

EDIT: Has anyone found the undocumented 60 MHz FSB setting for this motherboard?

A solid 138 MHz on maxspeed indictes to me that you are indeed running at 150 MHz. This is fantastic. Maxspeed always undershoots like this at frequencies above 100 MHz.

Can you use 60 ns RAM instead of 70 ns? If you plop in a 60 ns stick (preferable FPM over EDO), you should be able to run your cache at 2-1-x (or 3-1-x) and your RAM at 1ws/0ws (read/write). 3-2-3 is awefully slow. 3ws/3ws is increadibly slow too, even for 70 ns RAM. I'd try to hone in on the lowest possible stable numbers and re-run the tests. You'll also want to set your FSB-to-PCI divisor to 1:2/3 instead of 1:1/2, that is, if the HOT-433 has a 1:2/3 setting. I don't recall off the top of my head.

I beleive that this combination of cache and RAM timings is why your benchmark scores are so slow. Is write-back L1 cache enabled for the CPU? Are you using L2 write-back cache mode as well?

Once you fine-tune the cache/RAM settings, there are a few other Cyrix enhancement features which you probably do not have enabled that will greatly effect your SpeedSys score. I'll have a report out shortly which will identify the percent boosts of each feature, but for now I can tell you that LSSER, FP_FAST, L1 WRITE-BACK, and BTB have an impact on SpeedSys.

SpeedSys
LSSER=3%
FP_FAST=14%
WRITE-BACK=6% (as opposed to write-thru)

BTB = 0%, but if used in combination with LOOP and RSTK, =4%
(note, such a setting is not very stable, and therefore not recommend)

The fact that you got this running at 4V instead of 5V is really good news.

You do not need to take a digicam shot of your monitor for Speedsys, it has a built-in report generator. The option will be presented once you've finished all the memory and harddisk tests. Press R to create the PCX image report (screenshot). It will save this file to the speedsys folder. Then just convert it to a PNG file once in Windows. You should get something like the enclosed file below.

This is of an IBM 5x86C-133 with all Cyrix enhancements turned on, including the not-so-stable BTB+LOOP+RSTK and is the fastest 486 SpeedSys score I've ever seen (76.43). My hope is that your IBM 5x86C-150 will clobber the IBM 5x86C-133. A stable score for the IBM 5x86C-133 is 75.4 and a score with enhancements off is 60.63.

So if you are getting a score of 69.00 with enhancements off, expect to increase that to about 85.00 with enhancements on. Now if you improve those cache/RAM timings, I'd hope that you'd near 90.00. Keep us posted.

A solid 138 MHz on maxspeed indictes to me that you are indeed running at 150 MHz. This is fantastic. Maxspeed always undershoots like this at frequencies above 100 MHz.

I'm very happy to read this. 😀
I never thought I'd actually be running an IBM 100HF @ 150MHz (although I have a LOT of tuning to do before I get an impressive score).

I managed to get it to boot (not very stable) at 2-2-2 timing and 1ws/0ws with the pci clock at an unrecommendable 1:1. 3D Bench yielded a score of 100.0 before crashing immediatly after. But my speedsys score is still stuck at ~69 mark. (btw I ran the benchmarks on 4x 60ns ram sticks 4MB each, which should hold me over until I can test my 50ns sticks in there).

You'll also want to set your FSB-to-PCI divisor to 1:2/3 instead of 1:1/2, that is, if the HOT-433 has a 1:2/3 setting. I don't recall off the top of my head.

There is no setting for that, only 1:1 or 1:1/2. WB L1 and L2 are being used.
EDIT: I still don't know how to use the 5x86 utilities, so I'm going to leave that alone until I learn how to use it so I don't break anything 🤣 .

Your score must be heavily related to the Cyrix enhancements then.

Start with setting,
LSSER = 0
RSTK = 1
LOOP = 1
FP_FAST = 1
MEM_BYP = 1
DTE = 1

Good luck.

With some of the stable cyrix enhancements on (finally started to figure it out), my computer yielded a paltry score of 82.66. I couldn't get any more performance out of it using ram/cache tuning in BIOS, but I guess I'm lucky to have it even POST at only 4.0 volts.
I'm not sure I'll do further testing at 150MHz any time soon as I'm concerned with how much damage I'm doing to the components (maybe after I get a peltier and/or a PLL to play with 4.1v-4.5v).

Uploaded with ImageShack.us

EDIT: I skipped the other tests because I figured the CPU benchmark was really the only important one.

Well that certainly broke the 80 point threshold! It would be nice to let the memory test finish so we could see the L1/L2/Memory speeds on the graph, with the most interest being the L1 speed.

Could you let me know which Cyrix enhancement features were enabled for this result? If it works at 4.0 V, I see no reason to increase the voltage. I'd jump straight into a Win98SE installation and run Quake timedemo1. If you can run the full series of tests found in the U4BC (screenshots of which are in the World's Fastest 486), then I'd say there's a good chance this CPU is stable.

You'll still need a fan/heatsink with the peltier to remove heat from the otherside of the junction. The long-term stability of a peltier is questionable without a sophisticated controller/thermistor and I'm not certain if it will help in this case. The heat is still getting generated inside the CPU core and it must transition through the ceramic encasing.

I used the following setting for my latest SpeedSys post.

CR0
PG = 0
CD = 0
NW = 1
AM = 0
WP = 0
NE = 0
1 = 1
TS = 0
EM = 0
MP = 0
PE = 0

CCR1
MMAC = 0
SMAC = 0
USE SMI = 1

CCR2
USE SUSP = 1
BWRT = 1
WT = 1
SUSP HLT = 0
LOCK NW = 1
USE WBAK = 1

CCR3
MAPEN[3-0] = 0001
SMM MODE = 1
LINBRST = 1
NMI EN = 0
SMI LOCK = 0

CCR4
FP FAST = 1
DTE EN = 1
MEM BYP = 1
IORT[2-0] = 000

PCR0
LSSER = 0
LOOP EN = 1
BTB EN = 1
RSTK EN = 1

SMAR
Address A31-A15 = 0000111000001010
A15-12 = 0000
Size[3-0] = 0101

PMR
HLF CLK = 0
CLK1 = 0
CLK0 = 1

I heard that Peltier coolers "sweat". Wouldn't that be bad to have water droplets forming over your CPU?

Could you let me know which Cyrix enhancement features were enabled for this result?

LSSER = 0
RSTK = 1
LOOP = 1
FP_FAST = 1
MEM_BYP = 1
DTE = 1

You'll still need a fan/heatsink with the peltier to remove heat from the otherside of the junction. The long-term stability of a peltier is questionable without a sophisticated controller/thermistor and I'm not certain if it will help in this case. The heat is still getting generated inside the CPU core and it must transition through the ceramic encasing.

I heard that Peltier coolers "sweat". Wouldn't that be bad to have water droplets forming over your CPU?

Ok, I had thought of still needing a heatsink/fan on the peltier, but you're right about needing much more equipment to handle it... I'm tempted to dig out my Prometiea and use some phase change cooling (yes I have the sealing kit that I could mod for a socket 3) that would cool the whole cpu to ~-40C/F.

I attempted two more tests at 150MHz and I must say, it feels like walking on a thin glass bridge over the Grand Canyon. It wouldn't let me install Windows 98SE so I'll have to consider it unstable at 4v.

Uploaded with ImageShack.us

I noted that the score is .01 lower than before (possibly) because I forgot to crank the RAM timings at all this time.

In other thoughts, I don't have a sound card and I don't want an ISA one while I'm benching this thing. Does anyone know if an Aureal Vortex (version 2 maybe?) PCI based card would work in native DOS?

Well this is certainly a very gratifying image! I think I'll gaze at it for a bit...

The L2 and RAM throughput rates are a bit slower than mine, even slower than my Cyrix 5x86-133 at 33 MHz FSB. This is probably why you didn't get a score higher than 82.7. Some board's northbridge controller just doesn't like to run fast.

What were your symptoms for Windows 98SE install crashing? Can you install DOS and run Quake1 timedemo 1? What about 3Dbench2 and pcpbench? I've found that you can run the DOOM timedemo 3 from a burned cd-rw disc without having DOS installed.

If you're looking for stability and want to modify your motherboard's voltage regulator, i.e. to run at 4.3 Volts, see this article I wrote,
Modifying your motherboard's voltage regulator for overclocking

Grab the PDF at the bottom of the page.

The peltier might give you some insight into if the CPU is crashing for thermal reasons, or you can also put the CPU in the freezer for 20 minutes and see if you get a little further. What happens at 5V? Even a Cyrix 5x86-100 upclocked to 133 MHz was pretty stable at only 3.7 V, so I'd think that 4.0-4.30V would be sufficient for 150 MHz. You definately need to get a variable voltage regulator onto your motherboard (see above link).

I'll look to see if I can get my hands on a 8433uud motherboard sometime, probably (if I don't already have one buried somewhere in here) to test if it's the board's northbridge or whatever else it could be.

It looks like it'll be a while before I can afford to mod the voltage regulator, but thank you very much for the post. I'll need that info when I get started.

I ran a series of tests and it stayed together for all of them except that the results weren't consistent.
3dbench2 resulted in a range of 70.6-100.0 with no apparent reason. I tested 3dbench1 and got a score of 74.6 (didn't bother testing 3dbench after that). Speedsys was the most consistent with a score around 82.

I attempt to run Quake, but it wouldn't play from the hard drive so I ran it from CD instead of bothering to figure out why not. Sadly it froze at 150MHz so I ran it at 100MHz for reference. 10.6 FPS @ 100MHz IBM 5x86 with LSSER = 0 RSTK = 1 LOOP = 1 FP_FAST = 1 MEM_BYP = 1 DTE = 1.

As for the Win98SE problems, I'm going to try to install again after I take a break (the problem was that it would simply freeze up). I'll keep you posted.

P.S. As a good sign is that it's letting me do a lot more now before freezing up (versus when I first started the build). So I'm slowly ironing out the kinks.

It could still be an issue with the HOT-433. I've never had a rev 1-3 that I would classify as reliable, and that's 6 for 6 (albiet my standards are a bit high). Try installing Win98SE at 100 MHz with the same settings (4V). If it fails, then drop the voltage down to 3.45 V and 100 MHz, if still fails I'd start to suspect the HOT-433. If the CPU works fine at 4V and 100 MHz, then the issue is not thermal (that is, it is not due to over voltage).

I was unsuccessful at 150 MHz on my MB8433-UUD, but this is likely a factor of my CPU not being up to the 150 MHz challenge.

I figure all the money you need to dish out for the voltage regulator modification is $3 for a variable 5K resistor. Everything else is kinda standard at home lab equipment.

I was unsuccessful at 150 MHz on my MB8433-UUD, but this is likely a factor of my CPU not being up to the 150 MHz challenge.

Were you able to install DOS at 150MHz (I was able to do that without much trouble), or did you just try WIN98SE?

I figure all the money you need to dish out for the voltage regulator modification is $3 for a variable 5K resistor. Everything else is kinda standard at home lab equipment.

Actually I think I might already have the 5k resistor, but I haven't had a "home lab" setup in quite a while due to lack of space. Therefore, I'm missing most of the hardware needed to safely do this venture (by "afford" I ment the all-encompassing definition, not just monitary values).
I'm in no rush right now anyway. I've learned that patience is the key to getting things right.

nemesis wrote:

Were you able to install DOS at 150MHz (I was able to do that without much trouble), or did you just try WIN98SE?

If I recall, I couldn't even POST.

feipoa wrote:

nemesis wrote:

Were you able to install DOS at 150MHz (I was able to do that without much trouble), or did you just try WIN98SE?

If I recall, I couldn't even POST.

Ouch. What CPU were you using? I ask this because I managed to POST a cyrix 5x86 100GP at 150MHz (although I didn't actually test it any further) on the Shuttle HOT-433 Rev. 4.

Update: I retested the 100GP at 150MHz and it refused to go past DOS startup with normal settings. With these chips being as rare/expensive as they are in my area, I decided to save any further testing on that particular CPU til a later date.

I was using an IBM 5x86C-100HF on my crappy HOT-433 rev 1-3 boards. Where did you find a rev 4 HOT-433? To confirm, the Rev. 4 has only single bank cache (4 cache sockets + 1 TAG), and this is what you have? If so, I'd say this is the board to be testing overclocked CPUs on. I've always had shoty luck with my rev 1-3 boards.

Could you let me know if the PS/2 mouse feature of the rev 4 actually works? it never worked on the rev 1-3 boards, as far as I could tell. Thanks.