I experienced the idea controller in weird state in several occasions too.
I think your best option is for active cooling of the cpu, but if you are comfy with 180mhz and no further effort - totally get you.

I think over 200 you will need hard peltier 12v cooling and that will require massive insulation to prevent condensation. Can be a cool experiment but without space tech applied it will be a techbench experiment, even the rest of the system can take it. I ran 2x80 and while cool, it requires significant wait states which compromises the point.

I happened upon a QFP Am5x86 that runs well in Windows NT4/w9x at only 3.15 V and 180 MHz with heatsink/fan. If I gradually lower the voltage, it crashes at about 2.9 V. This testbed is using an M919 v3.4, 1024K cache module, 64 MB TSOP EDO memory, Matrox G200, YMF719-E based sound card, and 3C509B ethernet. Timings at 2-1-2 and 0/0 ws. For 0/0 ws and 64 MB, an 8 ns SRAM module was needed, otherwise I must use 1/0 ws for 10-12 ns modules.

The curious part of the experiment is that switching to an ARK1000VL VLB graphics card ruins everything. With the ARK installed, it cannot run at 180 MHz, even with adding wait-states or increasing the voltage up to 4.4 V. The same CPU runs fine at 180 MHz on other systems w/G200.

200 MHz is not stable, but as you noted in the thread earlier, this may just be the motherboard and not the CPU.

I am using two 8 GB CF cards and I cannot soft reset at any frequency.

pshipkov, I am curious if you run your M919 with identical hardware, do you also witness a roadblock with the ARK1000VL at 180 MHz [in Win9x]?

180 at 3.15V and air cooling?
sweet!

I still didn't test the M919. Heading for another weeklong business trip, so hobby will have to wait some more. This is an interesting subject, so i will follow through here.

Update: previous hard drive issue on my system turned out to be the hard drive dying - indeed, it had been the last of 3 Seagate barracudas that came from some store computer, the other 2 drives had died on me in the last year in other projects, so I suppose it was it's time.

Replaced HDD and had been running stable at 200 with 3.85volts, increased size fan in the front of the case running at full speed (previously I had it slowed down), about 20c ambient. So, increased cooling of the whole shebang was the answer in my case. HOWEVER- at the fans full speed I find the system too loud for my liking, so, I find myself running it at 180/3.65v again anyway, as at this setting it is rock solid without the extra noise

I haven't run a Seagate Barracuda in over a decade. I still have one that powers up, but it sounds like a rusty old bicycle chain being peddled at full throttle by a snotty nosed 10 year old attempting to chase down the icecream man.

Do you need 20 C ambient for 200 MHz to be stable? I think you said you are running without a peltier. I've been experimenting with some 23x23 and 25x25 TEC's and found some which will drop the temperature by a delta of only 5-8 C. I use this on one of my Am5x86 chips which cannot quite do air cooling at 180 MHz.

At 180 MHz / 3.65 V with air cooling, does using a VLB graphics card in Windows alter the stability at all? You are running 2-1-2, 1/0 ws?

Vlb cards are definitely more prone to fail at 60/66mhz compared to pci ones.

pshipkov wrote on 2025-11-06, 07:48:

Vlb cards are definitely more prone to fail at 60/66mhz compared to pci ones.

Can't argue with that. But does it make sense that increasing the CPU voltage (180 MHz) makes the VLB card stable?

Is a butterfly’s wingbeat in Beijing influencing the weather in New York?
: )

From my experience, getting out of specification with the old hardware components in question results in entanglements that are not easy to explain or define.
You feel subtle dependencies during the process, often illogical and bewildering, but sharing or even stating the outcome sounds unconvincing or silly.
Because of that i quickly forget about such things and even looking at old notes i am like "really?! no way! i was probably wrong."
So, the answer to your question is - tentatively possible, but don't have the facts to support that.

pshipkov wrote on 2025-11-06, 18:36:

Is a butterfly’s wingbeat in Beijing influencing the weather in New York?

What a well placed analogy. The answer is clearly, "no, it shouldn't be, but in this case, I think it is."

For my particular case, it wasn't just that I had to up the CPU voltage to 4.4 V (when the CPU can normally run fine at 3.8 V), but I also need to use a particular 1024K cache stick, one not necessarily with the fastest response time.

Did you make a list of VLB cards which generally run well at 60 and 66 MHz?

pshipkov wrote on 2025-11-06, 18:36:

Is a butterfly’s wingbeat in Beijing influencing the weather in New York? : ) […]
Show full quote

Is a butterfly’s wingbeat in Beijing influencing the weather in New York?
: )

From my experience, getting out of specification with the old hardware components in question results in entanglements that are not easy to explain or define.
You feel subtle dependencies during the process, often illogical and bewildering, but sharing or even stating the outcome sounds unconvincing or silly.
Because of that i quickly forget about such things and even looking at old notes i am like "really?! no way! i was probably wrong."
So, the answer to your question is - tentatively possible, but don't have the facts to support that.

Isn't the VLB directly connected to the cpu, or does it go through the chipset? If direct, more voltage on the cpu could help signal integrity on the VLB signals?

I felt poetic for a moment so the analogy came out right.

Well, you have a strong case in support of your question.
Ark1000vl has an attitude - does not like all mobos and and does not overclock well in general.
I know that it often requires inflated wait states for l2 cache and/or ram which is confusing as by itself as vlb interface is supposed to provide direct mem mapping, independent from l2 caching. So, the dependency is a bit of a mystery to me.
You are at the next level with the increased cpu voltage requirement but i am not surprised at all.

I spent good amount of time with basically all vlb chipsets.
When it comes to overclocking the best by large extent is s3 trio64 - the diamond cards, then everybody else far behind.
If you can get ark1000vl without needing to increase mem or cache wait states - this is the best, but quite difficult to achieve.
If you have to do it, you are better off with diamond s3 trio64 as it is easy on the wait states which results in better perf than the inflated ark1000vl.
Most other chipsets are hit or miss, but the point is that they are slower than these two at dos graphics which is kind of the major factor.

I recall that very few VLB cards work well at 180 MHz on the M919 with working sound. I agree that the Diamond Trio64 VLB is a universally well liked and fairly well performing card, but I could not overcome these crackling sounds through the soundcard when the Trio64 was installed. I would have tried ES1868, YMF719E, and a few other sound cards.

The graphics selection which worked with sound at this frequency was the Diamond Stealth 32 (ET4000/w32p), ARK1000VL, and maybe Madao's S3 968 4 MB. A reasonable person would have given up on VLB for this board and settled with PCI graphics, resulting in 3.1 V, 180 MHz, 2-1-2, 0/0 ws. Instead, I have insisted on VLB and now must run a different Am5x86 CPU, ultimately at 4.4 V, still 180 MHz, 2-1-2, and 1/0 ws. It had been running at 4.0 V until I installed SSF2T, which revealed some instabilities. No issues with other DOS games, just with SSF2T. There's something about this game which upsets the ARK card. Needing to run the CPU at 4.4 V hasn't been sitting well with me. It feels like I should install a small 25x25 mm peltier just to increase the CPU's lifespan.

If the ARK1000VL doesn't tolerate 60 MHz too happily, then I can sort of see why some extremes, like extra voltage, were needed. I'd be curious if you run into similar nasties. I did not re-run all the stability tests with the Stealth 32 VLB to determine if the 4.4 V and 1 ws adjustments were still needed. This would offer a broader picture - is the issue generally VLB on the M919, or is it just the ARK.

feipoa wrote on 2025-11-05, 04:06:

I haven't run a Seagate Barracuda in over a decade. I still have one that powers up, but it sounds like a rusty old bicycle chain being peddled at full throttle by a snotty nosed 10 year old attempting to chase down the icecream man.

Do you need 20 C ambient for 200 MHz to be stable? I think you said you are running without a peltier. I've been experimenting with some 23x23 and 25x25 TEC's and found some which will drop the temperature by a delta of only 5-8 C. I use this on one of my Am5x86 chips which cannot quite do air cooling at 180 MHz.

At 180 MHz / 3.65 V with air cooling, does using a VLB graphics card in Windows alter the stability at all? You are running 2-1-2, 1/0 ws?

Yes I'm running 2-1-2 1/0

I sadly don't own any VLB cards to try

The 20 degrees ambient is roughly what i have in the room consistently. My guess is that the temp can be somewhat higher, since without running the fan on full I could still get 15-20 minutes at 200 /3.85 before I could trigger a crash with the case closed. It could very well be the 919 or the cache (the g200s memory is right next to it and it gets pretty warm)

It is the ark1000vl card. Let's see how wrong i am, but quite confident in what i am saying.
M919 possesses above average overclocking qualities, especially with capable L2 cache chips.
At the same time, the same cannot be said about Ark1000, so ...

VLB is just the better place for DOS and pre-Win95/98 overall.
Also, faster in DOS and more fun than PCI hardware from the mid 90ies.

I would also go with some VLB hotrod for an M919 build - overclocked or not.

@bertrammatrix
How many CPUs you went through to find the unicorn you got?

From the other thread, bertrammatrix mentioned that it took "20-something" CPUs to find his 200 MHz Am5x86. He notes that 180 MHz is stable at 3.7 V and 200 Mhz is also stable 3.7 V but with increased air flow. I'm a little sceptical about the 200 MHz result. He is using a PCI G200, which helps a lot, as does avoiding the built-in IDE controller.

As I continue testing other boards and the years fly by, I'm noticing other curious hiccups when overclocking. These hiccups would normally go unnoticed by narrow testing. As noted earlier, my M919 with Am5x86-180 was running like a champ at 3.95 V with whatever I threw at it. It took my kids bugging me to play Super Street Fighter 2 Turbo (not available on SNES) to reveal that the ARK1000VL wasn't cutting it at 4.4 V in DOS. SSF2T would crash within a few minutes, yet no other DOS game I had tested would crash.

I currently have the UUD on the testbed and have been testing these S1R3 Cx5x86 QFP chips at 133/150 MHz. Many S0R5 and S1R3 Cyrix chips will run at 133 MHz (2x66) with onboard IDE-CF and only 1 ws on EDO read. It wasn't until I tested a game which simultaneously does a lot of CD-ROM reads and CF card read/writes that some issues emerged. The game is Outlaws (Direct3D). The game play itself is perfectly fine, but if I let the CD-ROM intro video play fully during loading (don't bypass it with ESC), there is frantic HDD LED and CD-ROM access for about 10 minutes. Sometimes the system would make it through the intro video, but most of the time it would hang-up hard. If I set 2 ws EDO read, there are no issues. I tried hard to overcome this with my stockpile of EDO/FPM sticks and different cache, but it didn't help. How to get 1 ws EDO read working in this case? I had to ditch the CF card for a traditional, but late model, magnetic platter drive.

My personal conclusion from these anecdotes, and other's I haven't shared, is that we need a very broad variety of tests to establish stability. It will ultimately boil down to 'what do you want to run' and 'what might you want to run in the future'. If you have a narrow spectrum of software, or only do bench-top testing, then it's pretty easy to overclock successfully.

pshipkov wrote on Yesterday, 02:55:

It is the ark1000vl card. Let's see how wrong i am, but quite confident in what i am saying. M919 possesses above average overcl […]
Show full quote

It is the ark1000vl card. Let's see how wrong i am, but quite confident in what i am saying.
M919 possesses above average overclocking qualities, especially with capable L2 cache chips.
At the same time, the same cannot be said about Ark1000, so ...

VLB is just the better place for DOS and pre-Win95/98 overall.
Also, faster in DOS and more fun than PCI hardware from the mid 90ies.

I would also go with some VLB hotrod for an M919 build - overclocked or not.

@bertrammatrix
How many CPUs you went through to find the unicorn you got?

Something over 20. Always just bought the dirt cheapest ones from China disregarding stepping, so most were ADW. Tried to avoid sellers with only 1-2 just to weed out any that may have already been tested for fast operation, however unlikely. The few that weren't ADW showed no real advantage over ADW, but admittedly the sample size of those was low. Also I immediately disqualified anything that wouldn't post at 160 on stock voltage so there's a slight possibility that I missed a chip that could be "ok", however I doubt it, I wouldn't really want to run it at close to 5 volts anyway if that's what it took.

feipoa wrote on Yesterday, 10:40:

From the other thread, bertrammatrix mentioned that it took "20-something" CPUs to find his 200 MHz Am5x86. He notes that 180 M […]
Show full quote

From the other thread, bertrammatrix mentioned that it took "20-something" CPUs to find his 200 MHz Am5x86. He notes that 180 MHz is stable at 3.7 V and 200 Mhz is also stable 3.7 V but with increased air flow. I'm a little sceptical about the 200 MHz result. He is using a PCI G200, which helps a lot, as does avoiding the built-in IDE controller.

As I continue testing other boards and the years fly by, I'm noticing other curious hiccups when overclocking. These hiccups would normally go unnoticed by narrow testing. As noted earlier, my M919 with Am5x86-180 was running like a champ at 3.95 V with whatever I threw at it. It took my kids bugging me to play Super Street Fighter 2 Turbo (not available on SNES) to reveal that the ARK1000VL wasn't cutting it at 4.4 V in DOS. SSF2T would crash within a few minutes, yet no other DOS game I had tested would crash.

I currently have the UUD on the testbed and have been testing these S1R3 Cx5x86 QFP chips at 133/150 MHz. Many S0R5 and S1R3 Cyrix chips will run at 133 MHz (2x66) with onboard IDE-CF and only 1 ws on EDO read. It wasn't until I tested a game which simultaneously does a lot of CD-ROM reads and CF card read/writes that some issues emerged. The game is Outlaws (Direct3D). The game play itself is perfectly fine, but if I let the CD-ROM intro video play fully during loading (don't bypass it with ESC), there is frantic HDD LED and CD-ROM access for about 10 minutes. Sometimes the system would make it through the intro video, but most of the time it would hang-up hard. If I set 2 ws EDO read, there are no issues. I tried hard to overcome this with my stockpile of EDO/FPM sticks and different cache, but it didn't help. How to get 1 ws EDO read working in this case? I had to ditch the CF card for a traditional, but late model, magnetic platter drive.

My personal conclusion from these anecdotes, and other's I haven't shared, is that we need a very broad variety of tests to establish stability. It will ultimately boil down to 'what do you want to run' and 'what might you want to run in the future'. If you have a narrow spectrum of software, or only do bench-top testing, then it's pretty easy to overclock successfully.

Avoid using the onboard IDE for anything, even CD rom, unless you want problems like that under windows. Last I used it with a Cyrix and I'd also get HD led flashing for no reason with some games if a CD was present in the drive. I just use it to install the system then immediately switch the cable to the secondary port on the promise ide card.

There is an old note on cutting a trace to a smd capacitor at the lowest ISA slot on these 919 boards, supposedly to rectify some stability issues with secondary IDE and certain ISA cards - I did do this on one of my boards but didn't notice a difference, however, I don't use the onboard IDE anyway (let alone the secondary) so really I can't confirm if this did anything useful or not. I'll post the link since there is other info there I've found useful

http://th2chips.freeservers.com/m919/unoff/m919.html

Thanks. That link is a classic for the M919 and has been around forever. Seems like the issue with the secondary IDE and serial or LPT ports was on the earlier board revisions. I'll keep an eye out for these issues. However, the the IDE-CF and CD-ROM issue I noted above with Outlaws was on a testbed with the MB-8433UUD motherboard.

Like you, I tend to avoid onboard IDE on socket 3/4/5 boards. Unfortunately, DOS CD-ROM access will not work with any PCI IDE controller card I've come across. Many of the later era DOS games either have disc verification tests before a game will run, or the audio is on the disc. Because of these, I have had to use onboard IDE ports for mostly CD-ROM access. An alternative is to use SCSI, which will have a DOS CD-ROM driver. Sometimes there are no-cd fixes for old DOS games. Did you give up on CD-ROM access in DOS?

Unfortunately, on my cased Biostar MB-8433UUD system (IBM 5x86-133/2x), I do not have an available PCI slot to use a PCI SCSI or UDMA contrller. It has 2x Voodoo2 and a Matrox G200. As such, I needed to use the onboard IDE with XT-IDE.

Lately, I had been using CF cards on my testbed for simplicity. This is where the latest issue with onboard IDE and CF cards arose. I might shift back to platter IDE drives for some testbeds.

feipoa wrote on Yesterday, 13:51:

Did you give up on CD-ROM access in DOS?

Yes I did, I never use straight DOS anyway unless it's for installation or just to run some benchmarks (but I usually run those from HDD unless it's just a test bed setup). I just leave an IDE cable plugged in to the primary connector (since it's awkward to get to) and plug it into the CD rom if I need to.

Possibly I will try something other then mechanical HDD one day, but so far my stash of working fairly modern ones has not dried up so I'll stick with them for now.

UMC chipsets have weak IDE controllers. This is a known issue that was discussed multiple times before. They are especially vulnerable when overclocked.

Additionally, CF cards offer convenience and much faster speeds than mechanical HDDs, but highly optimized or overclocked systems can become unstable because of latent bugs in IDE controllers - issues that were less likely to occur in the past because older local storage devices didn’t allow for such stress testing.