Maybe cache is disabled ? or Waitstates to kill of memory performance.

Anyway there is no benefit in OC this Machine.
Just grab a faster one ?

I guess cache chips aren't working good at 50Mhz... try lowering the OC to 45mhz or something in between 45-50.

matze79 wrote on 2020-07-03, 09:33:

Maybe cache is disabled ? or Waitstates to kill of memory performance.

Anyway there is no benefit in OC this Machine.
Just grab a faster one ?

Cache was enabled on my first couple attempts. I then disabled it to see if it'd make a difference but the CPU was still extra slow. I already tried both 1 and 3 waitstates for memory but no difference.

I already have several 486 machines ranging from 33-133mhz, but I want to run a strictly 386 system as fast as possible. Others have already achieved 50mhz w/ same board and CPU.

Now that I think of it, it could be an issue with the Am386DX-40's I ordered. Even at 40mhz, I cannot start DOOM: "invalid opcode in DOOM.EXE". Windows 95 won't start either.

Neither DOOM nor 95 are an issue with my Intel 386DX-33 running at 40mhz. Could both Am386DX's I got be defective?

kixs wrote on 2020-07-03, 10:18:

I guess cache chips aren't working good at 50Mhz... try lowering the OC to 45mhz or something in between 45-50.

I'll order a 90mhz crystal tomorrow night if I can't get 50mhz working by then

Would it be worth trying with the cache chips a actually removed from the motherboard rather than just disabled?

You can also try the Ti486SXL2-50 cpu. It will run at 50Mhz by default, maybe even higher...

kixs wrote on 2020-07-03, 11:08:

You can also try the Ti486SXL2-50 cpu. It will run at 50Mhz by default, maybe even higher...

Is the chip double-clocked / clock-doubled by default?
If so, wouldn't it require a 25MHz bus then?

Also, a 486DLC-40 has internal cache (needs to be enabled) and can be used as a faster 386DX40 replacement.
However, it's internal cache could break during overclocking..

I think when the "cache bad" is displayed, then it's disabled like disabling it manually from the BIOS.
I bet if you run the same benchmark with the 80MHz crystal and cache disabled you'll get even lower results

kixs wrote on 2020-07-03, 11:08:

You can also try the Ti486SXL2-50 cpu. It will run at 50Mhz by default, maybe even higher...

Probably not much higher. So far 55MHz is the best anyone has been able to do with the 5V CPGA version of this chip.

SXL2 starts in 1x mode by default. You need to use software to switch it to 2x mode.

I say if you're going to run 50MHz, you'd better make sure you have at least 15ns RAMs, especially the tag. 12ns would be ideal.

I can run my SXL-50 (50Mhz 1X mode) on OPTi495sx reliably, but I seem to recall it only works with the slowest cache settings.

Pretty sure the 20ns cache is not going to cut it at 50Mhz fsb. It definitely doesn't work on 486 boards at that speed from what I have seen.

Just some thinking.

20ns cache should be able to keep up with 50MHz bus speed (absolutely no more than that), but the tag is doubtful. But what if your cache chips can't keep up with its rated speed, even with the extra wait states?
A 40MHz bus only needs 25ns cache to keep up, so that's a bit of space to breath for your chips and may be enough to keep up with a just faster tag.
My 486 DX50 runs 20ns tags without issues, but cachechk tells me it's actually running 19ns? So above rated speed, I got lucky I guess.
45MHz should be ok then, with cache running about 22ns and tag a little faster.
Do you have another system with cache chips that fit into the 386 mobo? Any 15 or 20ns should do. 15ns would certainly stand 50MHz.

Also your DX40 may reach its limits as it was not designed to do 50MHz and can't keep up with itself. So if cache does stand the 50MHz the CPU might give you some errors. Again, 45MHz sounds like a good try. Or 47. Or 42.

Could there be any other factor? Does your ISA bus scale with front side bus? There may be something there, don't know what but it could be...

Also, have you tried at 40MHz without cache to compare? It should be a little below you other results.

Cachechk is your friend 😀

OK, so since the cache chips can be a pain to remove (w/o breaking), I decided to just try my 2 other FX-3000 motherboards, each with 128kb of 20ns cache...

The first one shows the exact same symptoms: "cache bad". The second board acted even more weirdly... on the first attempt, it froze at this weird string after the "wait" screen:

Subsequent attempts on that board led to a freeze with a blinking cursor after "wait". It looks like I didn't get lucky on any of my chips, and that 20ns just won't cut it. Time to look for 15ns chips...

Also just in case the CPU also can't handle it, I'm gonna try some slower crystals, ~90mhz. The place I ordered from last time only has 89.2 and 92.9405 mhz crystals... I'll probably order both.

waterbeesje wrote on 2020-07-03, 13:55:

Could there be any other factor? Does your ISA bus scale with front side bus? There may be something there, don't know what but it could be...

Also, have you tried at 40MHz without cache to compare? It should be a little below you other results.

Cachechk is your friend 😀

My ISA bus is set to 50/5 = 10mhz.

I tried 40mhz with cache disabled, and as I thought, it's still much faster than 50mhz w/o cache... something is definitely off, not just the cache. Cachecheck also reports no errors.

Good timing that yall mention the TI486SLX2. I just grabbed a SXL2-66 off Ebay last week, which appeared to have some kind of PGA-168 tp PGA-132 interposer... first time I've seen one. It'll definitely make for some fun benchmarking at 40, 50, even 66mhz if I could find the clock doubling software. Just thinking out loud though... for this build, I'd like to keep it strictly 386 (no SXL, no DLC etc)

Jo22 wrote on 2020-07-03, 11:19:

Also, a 486DLC-40 has internal cache (needs to be enabled) and can be used as a faster 386DX40 replacement.
However, it's internal cache could break during overclocking..

PSA right here ^^. A couple days ago (before I got the Am386's) I tested my DLC-40 at 50mhz... huge mistake. It was giving CPU register errors before the Win95 screen even came up, and even after switching back to 40mhz, Win95 kept giving BSoD fatal exceptions. RIP 👻

My guess is that your 40Mhz AMD 386 just does not like 50Mhz at all.

NScaleTransitModels wrote on 2020-07-04, 00:29:

I tried 40mhz with cache disabled, and as I thought, it's still much faster than 50mhz w/o cache... something is definitely off, not just the cache. Cachecheck also reports no errors.

This is strange. With cache disabled in both cases, you get 37 at 40MHz and 25 at 50MHz?
Long shot, but are you sure high speed/turbo was enabled and working when you got 25? Could you turn it off for example to go even lower?
Also try another benchmark, like norton sysinfo, maybe landmark is wrong (even though I understand your results are confirmed by doom)

Welp... giving up on the overclock for now. The FX-3000 board works fine with the Intel 386DX-33 at 40mhz (don't want to push that past 40), but it won't even take my AMD 386 at 40mhz, even tho that CPU works fine in my PCChips M326 motherboard. The PCChips has no oscillator tho, so it's also 40 tops 😕 I'll probably just save my FX-3000s for spares or future 486 builds, and just do a 386DX-40 build with the PCChips for now.

konc wrote on 2020-07-04, 08:50:

This is strange. With cache disabled in both cases, you get 37 at 40MHz and 25 at 50MHz?
Long shot, but are you sure high speed/turbo was enabled and working when you got 25? Could you turn it off for example to go even lower?
Also try another benchmark, like norton sysinfo, maybe landmark is wrong (even though I understand your results are confirmed by doom)

Yes that is correct 🤔
I had the turbo button disconnected the entire time. I'll probably try connecting it in the morning and see if it makes a difference... maybe the turbo on/off gets reversed at 50mhz? Highly unlikely but just for the LOLs
I did run Sysinfo at 50mhz last night, and also got something ridiculously low. 9-point-something IIRC, didn't write it down. I somehow ended up with the slowest 386 instead of the fastest 😂😂

Anonymous Coward wrote on 2020-07-03, 11:48:

Probably not much higher. So far 55MHz is the best anyone has been able to do with the 5V CPGA version of this chip. […]
Show full quote

kixs wrote on 2020-07-03, 11:08:

You can also try the Ti486SXL2-50 cpu. It will run at 50Mhz by default, maybe even higher...

Probably not much higher. So far 55MHz is the best anyone has been able to do with the 5V CPGA version of this chip.

SXL2 starts in 1x mode by default. You need to use software to switch it to 2x mode.

I say if you're going to run 50MHz, you'd better make sure you have at least 15ns RAMs, especially the tag. 12ns would be ideal.

I can run my SXL-50 (50Mhz 1X mode) on OPTi495sx reliably, but I seem to recall it only works with the slowest cache settings.

amadeus777999 wrote on 2020-07-04, 19:02:

Anonymous Coward wrote on 2020-07-03, 11:48:

Probably not much higher. So far 55MHz is the best anyone has been able to do with the 5V CPGA version of this chip. […]
Show full quote

kixs wrote on 2020-07-03, 11:08:

You can also try the Ti486SXL2-50 cpu. It will run at 50Mhz by default, maybe even higher...

Probably not much higher. So far 55MHz is the best anyone has been able to do with the 5V CPGA version of this chip.

SXL2 starts in 1x mode by default. You need to use software to switch it to 2x mode.

I say if you're going to run 50MHz, you'd better make sure you have at least 15ns RAMs, especially the tag. 12ns would be ideal.

I can run my SXL-50 (50Mhz 1X mode) on OPTi495sx reliably, but I seem to recall it only works with the slowest cache settings.

Cache is already disabled tho, and I doubt that 50mhz w/o cache would be far slower than even 40 w/o cache, unless something else is wrong. I think what Horun said is true... my CPUs can't handle 50 and those that can are likely from lucky batches.

Now the good news is that I tried my FX-3000 board #2 again just now. And I am no longer getting the VMM32.VXD or invalid opcode errors w/ the AMD CPU. So that does warrant a couple more attempts at 44.6 and 46.5mhz; I'll post back when I get the crystals.

Oh and as expected, the turbo button makes it even slower at 50mhz 🤣

You can convert MHz to nanoseconds by taking 1000/MHz. So indeed 1000/50=20ns which means if you have 20ns chips there is absolutely no margin for error.