Agree with rasz_pl .
On this type of motherboard a P3 should run at fsb 66,100 and 133 with no problem.

Besides the mentioned quality of the motherboard there is also the quality of the slotket to consider. Slotkets of this generation were not built to run at 133; some were even unstable at 100.

And re. the multiplier: it is very unlikely that this motherboard is unable to support multipliers up to 6x only. The typical setup for this type of cpu (and all possible jumper settings that are not mentioned in most of the manuals of early generation motherboards) you can find at the bottom of this page:
https://www.pchardwarelinks.com/cpuspeed.htm

multipliers jumpers are usually hardwired to cpu pins and for CPU consumption only, P3 CPUs ignore them completely

Well, let me get this out of the way first. I checked all my modifications to the slotket with the multimeter. I also checked surrounding pins for shorts, traces for continuity etc. There is no way I failed to modify it. In fact I fully expect the next p3 coppermine cpu to work in it. I'm not modifying routing of high frequency signals here. Those are pretty much static logic signals. I can send high resolution microscope photos to prove it.

The way I see it there are only two options.

1 - CPU is broken. However I did buy it from a reputable seller. I don't know for a fact they test them, but I would be surprised if they didn't. They claim everything they sell is good.
2 - The motherboard refuses to run when the multipler set doesn't match what the CPU reports to run at. From what I read in this thread this sounds like blasphemy to some people, but it is perfectly possible and I found another thread on this very board where another user refers to a motherboard that refuses to run if the clock multiplier set with the jumpers doesn't match what the cpu reports. Here is why it is possible and why it likely happens even before any bios activity starts:

This is page 7-7 from Intel's "Pentium 2 Processor Developer's Manual" and it describes first 8 (FSB) clock cycles of startup for the cpu. "System Bus Multiplier" are the 4 lines that determine the multiplier during reset, tell the motherboard what the cpu multiplier actually is shortly therafter, and have different roles from then on. One can clearly see that during the reset signal the motherboard sends its requested multiplier to the CPU (which the CPU can ignore), then once CRESET signal stops after one clock cycle the multiplier pins are set to "Compatibility" - this is not explained anywhere in the manual, but it is hinted at that the motherboard can at this time READ the multiplier state to determine what the CPU has actually been set to. This happens well before any BIOS activity starts as the BIOS runs on the CPU. If indeed this is what happens it happens in the chipset an no BIOS mods will change that (unless there is a register in the chipset that can switch this crap off).
Then, I'm observing exact that behavior with my P2 350. If I set it to a wrong multipier it fails to start. Yes the CPU has its multiplier unlocked for lower multipliers, but it should run at 4 while I get the same black screen if I set it at 4 (and no activity with PC analyzer etc).

If I was really desperate to get this 866mhz CPU running I would take this whole machine to my workshop, hook it up to logic analyzers and see exactly what is happening on those multiplier selection lines during startup to know once and for all, but I already ordered another coppermine cpu. One with multiplier of 6 and FSB of 100. Both should be supported by the motherboard. So I expect it to run with my modified sloket.

As for "crappy sloket" I'll say this is 370sp rev 2.0 which has its own jumper for setting the fsb to 100. Also some people modified it for 133mhz operation. Forthermore if indeed this was a case of crappy signal routing etc. I would expect it to crash immediately, but perhaps at least show c1 as a post code on some attempt. This is not what's happening to me so I doubt this, but I'm also borrowing a gigabyte sloket that has jumpers for 66/100/133 fsb's and vcore as low as 1.3v just to test.

Now to comment on few specific things.

rasz_pl wrote on 2022-11-20, 21:03:

cpu dont care, set jumpers for 66MHz on the slotket try forcing 66MHz with BSEL mod and try that P3 again I have a feeling its […]
Show full quote

flynth wrote on 2022-11-20, 10:55:

I could modify the slotket to add a jumper so the cpu sees it as 1 and mobo as 0.

cpu dont care, set jumpers for 66MHz on the slotket try forcing 66MHz with BSEL mod and try that P3 again
I have a feeling its not the BSEL that is the problem
-most likely you failed modifying your 370SP rev 2.0 slotket to coppermine
-maybe cpu is bad
-maybe motherboard is just low quality low effort. You cant expect quality from something that has this in manual:
"The CPU External (Bus) Frequency has something wrong with 66/100Mhz. "
the f does that mean? 😀 I dont trust low end brands with auto detection of anything. Abit? sure. Asus? yes please.

I don't need to modify the sloket for 66mhz fsb. It used to ground BSEL1 before my mod, and BSEL0 was controlled by jumper to ground or to cpu(which sets 1). I have tried 66MHz fsb with no luck before. Regarding the motherboard being "low quality effort" despite issues running the cpu which it clearly was never designed to support (and chinglish in the manual you point out) I have no problem with it(performance is another factor). So far I've run memtest for 12+h non stop on it, I installed Win98, drivers etc. Everything is rock solid. If indeed it locks up when cpu reports a multiplier it doesn't like it is a strange design choice, but it is very far from being "low quality effort". Back in the day when those motherboards and P2 CPus were new I worked in a PC shop/service center for a gap year. I don't remember every detail from that time, but I do remember we built hundreds of PCs with various VIA chipsets and they were all pretty solid (unless made by ECS. Everything made by ESC was crap back then). Oh, and SiS, that was real piece of ***. No one will catch me using a motherboard with a SiS chipset.

So what is it that I'm trying to say? I'm saying it is possible I have a bad unit, but those motherboards IMO are not "low quality effort". Yes, VIa 693 is a budget option and it lacks behind 440BX in performance by the previously mentioned 30%. I can live with that by putting a faster CPU and GPU in. I have yet to find any game or a piece of software that would run too slow on let's say P2 350 (GPU is MX440) and require Windows 98. All things I know of that could benefit from more performance will happily run on Windows XP (on a much faster hardware).

So why am I trying to get coppermine CPU running? Because I think it is cool have an option of running faster CPUs on those old (especially AT) boards.

rasz_pl wrote on 2022-11-20, 21:03:

flynth wrote on 2022-11-20, 10:55:

Edit: Indeed Intel say "individual processors will only operate at their specified front side bus (FSB) frequency, either 100 MHz or 133 MHz, not both." so it is likely it might refuse to run

no

Can you elaborate a bit more that just "no" 😀 The bit in quotes I say comes from intel's own developer's manual. Do you mean you've run a 133fsb coppermine CPU on a 100MHZ FSB with BSEL1 grounded? If yes, then indeed "no" might be a correct answer and I would like to know. If 100MHz fsb was used with no knowledge what BSEL1 was set to this doesn't mean that much. A CPU can't measure its FSB, but it can be designed to refuse to run on a motherboard that reports it doesn't support 133MHZ FSB when it is a 133MHZ FSB CPU. The motherboard reports it doesn't support 133MHZ fsb by grounding BSEL1.

rasz_pl wrote on 2022-11-20, 21:03:

flynth wrote on 2022-11-20, 10:55:

The P2 indeed has an unlocked multiplier. I can set it to lower and go as low as 200mhz. But I didn't manage to run it with 66mhz fsb with p2 as I didn't want to modify what is a nice vintage slot1 cpu.

how do you think changing FSB in bios menu would modify a cpu?

There is no option to set FSB lower than 100MHZ in BIOS on CPUs that sets BSEL0 to 1 on this motherboard. It only goes up to 133MHZ (to overclock). There is no BIOS option to underclock a CPU. The only way to do it is to modify the CPU's BSEL0 line. This is a slot1, encased CPU. Therefore my previous answer.

rasz_pl wrote on 2022-11-20, 21:03:

flynth wrote on 2022-11-20, 10:55:

Edit: I looked at the MB suggested above (Asus p2b) , but... It is an atx motherboard. I'm using a very nice AT case for it.

you have a very nice case for a bad motherboard, I dont know what to tell you

That's like your opinion man.... Also, I haven't soldered that motherboard into the case. If a better AT motherboard shows up that meets my requirements I will swap it. However, when I don't care for the 30% difference in performance this motherboard still looks quite attractive to me. As for the choice of AT. Retro computing it an entire package to me. A nice AT case is a big part of it. I never was a big fan of those beige ATX cases that were so popular from the beginning of the ATX era until other colours became more popular(other colours just look modern). I quite like slim/small desktop ATX cases that require low height cards, but there is no way I'm fitting my ISA cards in there. At the same time I appreciate for others different things matter and maybe me running this Acorp VIA motherboard to you is like someone else's would be running an ESC board with a SiS chipset to me. So I understand where you're coming from.

rasz_pl wrote on 2022-11-20, 21:03:

flynth wrote on 2022-11-20, 10:55:

Is there a common AT MB that would support coppermine cpus?

on slot1 its just a matter of good slotket able to force fsb and voltage, maybe bios mod.

That I already have (and another slotket and CPU coming). So, we'll find out soon enough.

flynth wrote on 2022-11-22, 00:14:

I can send high resolution microscope photos to prove it.

best way would be trying another coppermine in it. Allegro has 1$ untested Intel Celeron 800 MHz with $2 shipping buy now, and plenty of $4 tested ones.

flynth wrote on 2022-11-22, 00:14:

2 - The motherboard refuses to run when the multipler set doesn't match what the CPU reports to run at. From what I read in this thread this sounds like blasphemy

its blasphemy, trace those multiplier jumpers on your board, you will find they only go to CPU socket and only thru resistors weakly pulling up/down. CPU with unlocked multiplier will read those pins, CPU with locked has those pins laser cut.

flynth wrote on 2022-11-22, 00:14:

If I was really desperate to get this 866mhz CPU running I would take this whole machine to my workshop, hook it up to logic analyzers and see exactly what is happening on those multiplier selection lines during startup to know once and for all,

do it purely for educational value 😀

flynth wrote on 2022-11-22, 00:14:

but I already ordered another coppermine cpu. One with multiplier of 6 and FSB of 100. Both should be supported by the motherboard. So I expect it to run with my modified sloket.

and conveniently you will be able to claim its all about multiplier 😜

flynth wrote on 2022-11-22, 00:14:

As for "crappy sloket" I'll say this is 370sp rev 2.0 which has its own jumper for setting the fsb to 100. Also some people modified it for 133mhz operation. Forthermore if indeed this was a case of crappy signal routing etc. I would expect it to crash immediately,

I think it was meant crappy as in early Non coppermine one, as opposed to brand name coppermine ready ones from MSI or ASUS. Personally I had good luck modding crappy cheap slotkets, its one wire plus ripping out one pin receptacle from the socket.

flynth wrote on 2022-11-22, 00:14:

Regarding the motherboard being "low quality effort" despite issues running the cpu which it clearly was never designed to support (and chinglish in the manual you point out) I have no problem with it(performance is another factor). So far I've run memtest for 12+h non stop on it, I installed Win98, drivers etc. Everything is rock solid. If indeed it locks up when cpu reports a multiplier it doesn't like

CPU has no facility to report its multiplier other than thru MSR (cpuid) while already running. FID pins are one way to the CPU, and locked CPUs just ignore them.

flynth wrote on 2022-11-22, 00:14:

it is very far from being "low quality effort"

No need to rationalize ownership of that board. I get it, its yours and its great 😀

flynth wrote on 2022-11-22, 00:14:

. Back in the day when those motherboards and P2 CPus were new I worked in a PC shop/service center for a gap year.

what city? maybe we know each other? 😀

flynth wrote on 2022-11-22, 00:14:

I don't remember every detail from that time, but I do remember we built hundreds of PCs with various VIA chipsets

condolences 😀

flynth wrote on 2022-11-22, 00:14:

Can you elaborate a bit more that just "no" 😀 The bit in quotes I say comes from intel's own developer's manual. Do you mean you've run a 133fsb coppermine CPU on a 100MHZ FSB with BSEL1 grounded? If yes, then indeed "no" might be a correct answer and I would like to know.

yes 😜, or more precisely you need to make sure both BSEL pins have correct level for 100MHz. Unlike FID jumpers that are only for CPU, BSEL jumpers are only for motherboard and CPU doesnt care about them.

flynth wrote on 2022-11-22, 00:14:

If 100MHz fsb was used with no knowledge what BSEL1 was set to this doesn't mean that much. A CPU can't measure its FSB, but it can be designed to refuse to run on a motherboard that reports it doesn't support 133MHZ FSB when it is a 133MHZ FSB CPU. The motherboard reports it doesn't support 133MHZ fsb by grounding BSEL1.

nothing reports anything to each other, CPU has pull ups/downs on BSEL pins. Motherboard connects those, either directly or thru its own jumper block or softmenu soft controlled gpios, to clock generator chip. If clock gen doesnt support signaled frequency it doesnt start. Its all very primitive.

flynth wrote on 2022-11-20, 10:55:

There is no option to set FSB lower than 100MHZ in BIOS on CPUs that sets BSEL0 to 1 on this motherboard. It only goes up to 133MHZ (to overclock). There is no BIOS option to underclock a CPU. The only way to do it is to modify the CPU's BSEL0 line. This is a slot1, encased CPU. Therefore my previous answer.

masking tape on CPU slot1 edge connector BSEL pins, resistors on SLOT1 BSEL pins on the back of the motherboard. Thats how we used to do it back in the day 😀
Celeron 300A was one sliver of masking tape on B21 and that was enough to force 66->100 without any soldering. Two more pieces of tape on Vcore pins to bump voltage and you were in business even with stubborn CPUs.

flynth wrote on 2022-11-20, 10:55:

That's like your opinion man....

I mean, it doesnt work 😜

flynth wrote on 2022-11-20, 10:55:

That I already have (and another slotket and CPU coming). So, we'll find out soon enough.

My bet is 1 slotket 2 motherboard 3 cpu in that order. Sorry for shitting on your motherboard, but as you are experiencing it right now its not without merit. Even if you manage to overcome this P3 coppermine hurdle it will still be a low end board unable to control fsb, vcore, with slow chipset and crash prone AGP support. Not a fan.

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-22, 00:14:

I can send high resolution microscope photos to prove it.

best way would be trying another coppermine in it. Allegro has 1$ untested Intel Celeron 800 MHz with $2 shipping buy now, and plenty of $4 tested ones.

$2 for untested and $4 for tested ones? How are you finding them? I mean what search terms are you using? When I search for "pentium 3 370" I get one seller that sells them for $10 ($15 including postage as he has no cheap postage options). If I search for "pentium 3 800" I get 600 hits for lots of p4s, cases, etc. Eventually I found one P3 800/133 at a fairly good price of $8.5 (including $2.5 of shipping). Then searching for just "pentium" setting the category to cpus only I found one for $3 (seller doesn't post, in person pickup only 250km away from me). Then other ones come to $8.5 (including cheapest postage).
Contrary to popular opinion $1 USD is not 7PLN (yet). 😀

It is not a fortune by any means, especially that almost all those sellers accept returns "for no reason". So I'll probably get another one of 866/133/6.5x (sl4cb) just to exclude the possibility of a broken cpu, a 800/133/6x (sl4mb) to test the "board doesn't like multipliers over 6x" theory, a 933/133/7x just because I found one quite cheap($6 with postage) and it is another one to test with. All the above except 933 I can return with free return postage. Also I'm already waiting for a 600/100/6x to arrive to test if really the cpu doesn't care about the MB's 133mhz fsb capability.

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-22, 00:14:

2 - The motherboard refuses to run when the multipler set doesn't match what the CPU reports to run at. From what I read in this thread this sounds like blasphemy

its blasphemy, trace those multiplier jumpers on your board, you will find they only go to CPU socket and only thru resistors weakly pulling up/down. CPU with unlocked multiplier will read those pins, CPU with locked has those pins laser cut.

The multiplier pins have their role to set the multiplier only during reset. From then on they have other roles. The below is from intel's own docs:
"During active RESET#, each processor begins sampling the A20M#, IGNNE# , and LINT[1:0] values to determine the ratio of core-clock frequency to bus-clock frequency. (See Table 7-1.) On the active-to-inactive transition of RESET#, each processor latches these signals and freezes the frequency ratio internally. System logic must then release these signals for normal operation; see Figure 7-4 for an example implementation of this logic. " (emphasis added).

So those pins cannot, not be connected to anything else. More about it below.

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-22, 00:14:

If I was really desperate to get this 866mhz CPU running I would take this whole machine to my workshop, hook it up to logic analyzers and see exactly what is happening on those multiplier selection lines during startup to know once and for all,

do it purely for educational value 😀

If I have time I might.

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-22, 00:14:

but I already ordered another coppermine cpu. One with multiplier of 6 and FSB of 100. Both should be supported by the motherboard. So I expect it to run with my modified sloket.

and conveniently you will be able to claim its all about multiplier 😜

Of course I'm trying to maximise the chances of success when ordering. But I've also ordered other cpus to put all those claims/theories to the test. I even have a "high quality" gigabyte slotket on loan coming in the mail, but with the simplicity of the slotket modification there is no way in hell I messed that up. All I did to that slotket was remove pin am2, Bridge reset and reset2 (then after lots of testing) cut Bsel1 and connect it to vpp via 1k on the cpu side, left grounded on MB side (this is not necessary, but it was to test a theory)

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-22, 00:14:

Regarding the motherboard being "low quality effort" despite issues running the cpu which it clearly was never designed to support (and chinglish in the manual you point out) I have no problem with it(performance is another factor). So far I've run memtest for 12+h non stop on it, I installed Win98, drivers etc. Everything is rock solid. If indeed it locks up when cpu reports a multiplier it doesn't like

CPU has no facility to report its multiplier other than thru MSR (cpuid) while already running. FID pins are one way to the CPU, and locked CPUs just ignore them.

Now we get to the interesting part 😀 You say that, and I agree "while already running", but what about during startup, just after CRESET is lifted (you saw the intel's chart I posted)?

They don't sey it out right, but to me it sure looks like the motherboard might be reading the values of those pins at the "compatibility" phase. Intel even suggests to tie one Bsel to reset on mobos that don't support certain frequencies to stop cpus from running. (I'm not sure if any board designers went that far - perhaps I should check for this too).

You say "those pins are cut" I'm not sure what do you mean by FID pins. The multiplier is set during reset by LINT0, LINT1, A20m and IGNNE. All those pins have other important roles afterwards. But even if it was true. If those pins were cut. This still allows the mobo to first set them as told, then to read them back to ensure it gets back the same thing. Still, this is all assumptions so I have various cpus on the way to test.

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-22, 00:14:

Can you elaborate a bit more that just "no" 😀 The bit in quotes I say comes from intel's own developer's manual. Do you mean you've run a 133fsb coppermine CPU on a 100MHZ FSB with BSEL1 grounded? If yes, then indeed "no" might be a correct answer and I would like to know.

yes 😜, or more precisely you need to make sure both BSEL pins have correct level for 100MHz. Unlike FID jumpers that are only for CPU, BSEL jumpers are only for motherboard and CPU doesnt care about them.

Well hopefully be able to prove/disprove that. There is no shortage of Intel saying one thing in a datasheet and doing another in hardware. I'm definitely not going to claim their documentation as gospel. I imagine you say that with all this certainty based on running many(more than one) of 133mhz coppermine CPUs on 100mhz fsb only boards? Is that right?

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-22, 00:14:

If 100MHz fsb was used with no knowledge what BSEL1 was set to this doesn't mean that much. A CPU can't measure its FSB, but it can be designed to refuse to run on a motherboard that reports it doesn't support 133MHZ FSB when it is a 133MHZ FSB CPU. The motherboard reports it doesn't support 133MHZ fsb by grounding BSEL1.

nothing reports anything to each other, CPU has pull ups/downs on BSEL pins. Motherboard connects those, either directly or thru its own jumper block or softmenu soft controlled gpios, to clock generator chip. If clock gen doesnt support signaled frequency it doesnt start. Its all very primitive.

I heard that too, but is it definitely true for every motherboard? In theory there can be many boards where this is 100% true, 100% false and anything in between. To be able to make a claim like this with good level of certainty one would have to have lots of experience with various boards back in the day doing something more involved than swapping parts(pcb repair?) . This would allow one to see many different board/cpu combinations in various situations.

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-20, 10:55:

There is no option to set FSB lower than 100MHZ in BIOS on CPUs that sets BSEL0 to 1 on this motherboard. It only goes up to 133MHZ (to overclock). There is no BIOS option to underclock a CPU. The only way to do it is to modify the CPU's BSEL0 line. This is a slot1, encased CPU. Therefore my previous answer.

masking tape on CPU slot1 edge connector BSEL pins, resistors on SLOT1 BSEL pins on the back of the motherboard. Thats how we used to do it back in the day 😀
Celeron 300A was one sliver of masking tape on B21 and that was enough to force 66->100 without any soldering. Two more pieces of tape on Vcore pins to bump voltage and you were in business even with stubborn CPUs.

Indeed it was done back in the day. To make the fsb higher. Not lower. The BSEL truth table was posted on page 1 of this thread. By masking you can make a 0 into 1, but not the other way around (0 is ground, 1 is nc or pull up).

rasz_pl wrote on 2022-11-22, 04:36:

I mean, it doesnt work :P […]
Show full quote

flynth wrote on 2022-11-20, 10:55:

That's like your opinion man....

I mean, it doesnt work 😜

flynth wrote on 2022-11-20, 10:55:

That I already have (and another slotket and CPU coming). So, we'll find out soon enough.

My bet is 1 slotket 2 motherboard 3 cpu in that order. Sorry for shitting on your motherboard, but as you are experiencing it right now its not without merit. Even if you manage to overcome this P3 coppermine hurdle it will still be a low end board unable to control fsb, vcore, with slow chipset and crash prone AGP support. Not a fan.

Well, it is a matter of personal opinion so I'm not looking to convince you, but for anyone else reading this thread I'll say. If a board doesn't work with a cpu it has never been designed for in the first place is it a sign of bad quality? No, if it does work it is a bonus, but one can have a shitty board that works with everything badly, one can also have a rock solid board that only works with certain cpus. Back in the day there were lots of MB manufacturers. Reputatation for flakiness could destroy a company, so I'm not surprised some manufacturers went out of their way to ensure users didn't run "unsupported" cpus and then blamed the mobo when they run like crap. For me the measure of board's quality is how stable it runs So far I haven't seen anything to say this board is shoddy.

Then you mention other things that are not quite right:
- unable to control fsb - not entirely true, the board has fsb control in the BIOS. On a 100mhz cpu you can go as high as 133mhz. On a 66mhz you can go up to 88mhz. You just can't swap 66 for 100 in the BIOS and vice versa. However, there seems to be an unpopulated jumper footprint to be able to force 66mhz (there is already a jumper to force 100mhz. I might just solder it in.
- unable to control vcore - not true, sure not in the BIOS, but jumper wise it goes down to 1.5V.
- slow chipset - yes, that one is true based on what I read, I already mentioned why this 30% doesn't really matter to me. What does (IO) runs fast enough.
- crash prone AGP - I have not seen this at all. If I did that woukd be enough for me to bin it, but as I said I have a 440mx in the slot and so far it has been fine. We'll see down the road. I'll make sure to update this thread in some weeks as it became a sort of unofficial "693 via chipset judgemt day" thread 😀

Edit: BTW, the city I worked back then was Olsztyn.

flynth wrote on 2022-11-22, 00:14:

The way I see it there are only two options. 2 - The motherboard refuses to run when the multipler set doesn't match what the CP […]
Show full quote

The way I see it there are only two options.
2 - The motherboard refuses to run when the multipler set doesn't match what the CPU reports to run at. From what I read in this thread this sounds like blasphemy to some people, but it is perfectly possible and I found another thread on this very board where another user refers to a motherboard that refuses to run if the clock multiplier set with the jumpers doesn't match what the cpu reports. Here is why it is possible and why it likely happens even before any bios activity starts:
Screenshot from 2022-11-22 00-03-30.png
This is page 7-7 from Intel's "Pentium 2 Processor Developer's Manual" and it describes first 8 (FSB) clock cycles of startup for the cpu. "System Bus Multiplier" are the 4 lines that determine the multiplier during reset, tell the motherboard what the cpu multiplier actually is shortly therafter, and have different roles from then on. One can clearly see that during the reset signal the motherboard sends its requested multiplier to the CPU (which the CPU can ignore), then once CRESET signal stops after one clock cycle the multiplier pins are set to "Compatibility" - this is not explained anywhere in the manual, but it is hinted at that the motherboard can at this time READ the multiplier state to determine what the CPU has actually been set to. This happens well before any BIOS activity starts as the BIOS runs on the CPU. If indeed this is what happens it happens in the chipset an no BIOS mods will change that (unless there is a register in the chipset that can switch this crap off).
Then, I'm observing exact that behavior with my P2 350. If I set it to a wrong multipier it fails to start. Yes the CPU has its multiplier unlocked for lower multipliers, but it should run at 4 while I get the same black screen if I set it at 4 (and no activity with PC analyzer etc).

Have you tried the undocumented setting [off on on off] for 6.5X with your 866 cpu ?

PARKE wrote on 2022-11-22, 11:31:

flynth wrote on 2022-11-22, 00:14:

The way I see it there are only two options. 2 - The motherboard refuses to run when the multipler set doesn't match what the CP […]
Show full quote

The way I see it there are only two options.
2 - The motherboard refuses to run when the multipler set doesn't match what the CPU reports to run at. From what I read in this thread this sounds like blasphemy to some people, but it is perfectly possible and I found another thread on this very board where another user refers to a motherboard that refuses to run if the clock multiplier set with the jumpers doesn't match what the cpu reports. Here is why it is possible and why it likely happens even before any bios activity starts:
Screenshot from 2022-11-22 00-03-30.png
This is page 7-7 from Intel's "Pentium 2 Processor Developer's Manual" and it describes first 8 (FSB) clock cycles of startup for the cpu. "System Bus Multiplier" are the 4 lines that determine the multiplier during reset, tell the motherboard what the cpu multiplier actually is shortly therafter, and have different roles from then on. One can clearly see that during the reset signal the motherboard sends its requested multiplier to the CPU (which the CPU can ignore), then once CRESET signal stops after one clock cycle the multiplier pins are set to "Compatibility" - this is not explained anywhere in the manual, but it is hinted at that the motherboard can at this time READ the multiplier state to determine what the CPU has actually been set to. This happens well before any BIOS activity starts as the BIOS runs on the CPU. If indeed this is what happens it happens in the chipset an no BIOS mods will change that (unless there is a register in the chipset that can switch this crap off).
Then, I'm observing exact that behavior with my P2 350. If I set it to a wrong multipier it fails to start. Yes the CPU has its multiplier unlocked for lower multipliers, but it should run at 4 while I get the same black screen if I set it at 4 (and no activity with PC analyzer etc).

Have you tried the undocumented setting [off on on off] for 6.5X with your 866 cpu ?

Yes. I did. The setting I tried for 6.5 was two middle jumpers populated. Two on the outside off.

I figured out those jumpers are just binary from left to right. Each number down is half more of the multiplier with 15 setting being 2x multiplier. 14 being 2.5. 13 is 3. 14 is 3.5 and so on.

If this is wrong please let me know.

Edit: One more thing I tried that I don't think I mentioned is bios patching. I used the Rom.by patcher. Version 6 alpha. It added various cpu ids including one for the coppermine cpu I had. This didn't make any difference. The more I think of it, the more I become convinced it might be a bad cpu...

It was just a hunch but if it had worked it would prove that your idea that the motherboard 'needs' a correct multiplier setting in order to work is correct. I have not come across this idea before this thread - the general consensus is that locked P3's totally ignore the motherboard settings wich also coincides with my own (limited) experience. We'll have to wait for your new hardware to arrive.

flynth wrote on 2022-11-22, 10:42:

rasz_pl wrote on 2022-11-22, 04:36:

flynth wrote on 2022-11-22, 00:14:

I can send high resolution microscope photos to prove it.

best way would be trying another coppermine in it. Allegro has 1$ untested Intel Celeron 800 MHz with $2 shipping buy now, and plenty of $4 tested ones.

$2 for untested and $4 for tested ones? How are you finding them? I mean what search terms are you using? When I search for "pentium 3 370" I get one seller that sells them for $10 ($15 including postage as he has no cheap postage options). If I search for "pentium 3 800" I get 600 hits for lots of p4s, cases, etc. Eventually I found one P3 800/133 at a fairly good price of $8.5 (including $2.5 of shipping). Then searching for just "pentium" setting the category to cpus only I found one for $3 (seller doesn't post, in person pickup only 250km away from me). Then other ones come to $8.5 (including cheapest postage).
Contrary to popular opinion $1 USD is not 7PLN (yet). 😀

people dont know what 370 is, what coppermine is, they have a green cpu with "celeron" written on it and sell it like that. In this particular case someone with a ton of retro gear still doesnt bother describing it in more detail
https://allegrolokalnie.pl/oferta/intel-celeron-800-mhz-x4z

flynth wrote on 2022-11-22, 10:42:

rasz_pl wrote on 2022-11-22, 04:36:

its blasphemy, trace those multiplier jumpers on your board, you will find they only go to CPU socket and only thru resistors weakly pulling up/down. CPU with unlocked multiplier will read those pins, CPU with locked has those pins laser cut.

The multiplier pins have their role to set the multiplier only during reset. From then on they have other roles. The below is from intel's own docs:
"During active RESET#, each processor begins sampling the A20M#, IGNNE# , and LINT[1:0] values to determine the ratio of core-clock frequency to bus-clock frequency. (See Table 7-1.) On the active-to-inactive transition of RESET#, each processor latches these signals and freezes the frequency ratio internally. System logic must then release these signals for normal operation; see Figure 7-4 for an example implementation of this logic. " (emphasis added).

you are right. sorry, FID is AMD nomenclature for pins setting multiplier on socket A cpus. In AMD case FID[3:0] arent shared with anything.
Intel calls them BF, BF are weak pull ups/downs on the signals you mentioned. Sadly doesnt change anything, on locked CPU bootstrap logic is laser cut and CPU doesnt probe those pins.

flynth wrote on 2022-11-22, 10:42:

Now we get to the interesting part 😀 You say that, and I agree "while already running", but what about during startup, just after CRESET is lifted (you saw the intel's chart I posted)?

They don't sey it out right, but to me it sure looks like the motherboard might be reading the values of those pins at the "compatibility" phase. Intel even suggests to tie one Bsel to reset on mobos that don't support certain frequencies to stop cpus from running. (I'm not sure if any board designers went that far - perhaps I should check for this too).

You say "those pins are cut" I'm not sure what do you mean by FID pins. The multiplier is set during reset by LINT0, LINT1, A20m and IGNNE. All those pins have other important roles afterwards. But even if it was true. If those pins were cut. This still allows the mobo to first set them as told, then to read them back to ensure it gets back the same thing. Still, this is all assumptions so I have various cpus on the way to test.

again: "sorry, FID is AMD nomenclature for pins setting multiplier on socket A cpus."
For motherboard to read anything back it would need a working CPU or embedded microcontroller. It was 20 years ago, there were no secondary CPUs in the chipset. Intel ME/Amd Secure whatever came 10 years later. Everything was quite primitive.

flynth wrote on 2022-11-22, 10:42:

Well hopefully be able to prove/disprove that. There is no shortage of Intel saying one thing in a datasheet and doing another in hardware. I'm definitely not going to claim their documentation as gospel. I imagine you say that with all this certainty based on running many(more than one) of 133mhz coppermine CPUs on 100mhz fsb only boards? Is that right?

no 😀 but selling probably over hundred overclocked Celeron systems in 1999-2001 period. When forcing FSB you insulate offending BSEL pins (ones that need different logic level from what CPU shipped with) between CPU and motherboard and _only_ influence what motherboard is reading. CPU dont care.

flynth wrote on 2022-11-22, 10:42:

I heard that too, but is it definitely true for every motherboard? In theory there can be many boards where this is 100% true, 100% false and anything in between. To be able to make a claim like this with good level of certainty one would have to have lots of experience with various boards back in the day doing something more involved than swapping parts(pcb repair?) . This would allow one to see many different board/cpu combinations in various situations.

It works like that on every motherboard I had in my hands, that would be ~everything sold in Poland at the time from PCChips to Asus/Abit.
Could Acorp implement super clever useless bit of complicated logic? Yes. Would they? Not if it cost more than 10 cents.

flynth wrote on 2022-11-20, 10:55:

Indeed it was done back in the day. To make the fsb higher. Not lower. The BSEL truth table was posted on page 1 of this thread. By masking you can make a 0 into 1, but not the other way around (0 is ground, 1 is nc or pull up).

In case of Core2 you get more combination and need both pullups and pulldowns, works same way https://linustechtips.com/topic/767799-how-to … th-tape/page/2/
https://www.dpreview.com/forums/thread/2326574

flynth wrote on 2022-11-20, 10:55:

Well, it is a matter of personal opinion so I'm not looking to convince you, but for anyone else reading this thread I'll say. If a board doesn't work with a cpu it has never been designed for in the first place is it a sign of bad quality?

chinglish, low end chipset, individual components might not be bad quality, but design and end product is low end

flynth wrote on 2022-11-20, 10:55:

Then you mention other things that are not quite right:
- unable to control fsb - not entirely true, the board has fsb control in the BIOS. On a 100mhz cpu you can go as high as 133mhz. On a 66mhz you can go up to 88mhz. You just can't swap 66 for 100 in the BIOS and vice versa. However, there seems to be an unpopulated jumper footprint to be able to force 66mhz (there is already a jumper to force 100mhz. I might just solder it in.

So they tried, but failed implementing proper FSB control. You can do some things (go up), but are artificially locked from other settings without any reason. All they had to do was solder 4 pins and print simple table, or implement fallback to 66MHz if CPU fails to start.

flynth wrote on 2022-11-20, 10:55:

- unable to control vcore - not true, sure not in the BIOS, but jumper wise it goes down to 1.5V.

what jumpers? https://theretroweb.com/motherboards/s/acorp-6via86p the ones not on the board? 😜

flynth wrote on 2022-11-20, 10:55:

Edit: BTW, the city I worked back then was Olsztyn.

Warsaw here. ~1998-2008. SME at first, then private sector.

Thank you for taking the time to reply to what starts to turn into a massive conversation on multiple topics 😀

While I'm waiting for those cpus to arrive I'll answer just one thing from the previous post.

rasz_pl wrote on 2022-11-22, 20:35:

flynth wrote on 2022-11-20, 10:55:

- unable to control vcore - not true, sure not in the BIOS, but jumper wise it goes down to 1.5V.

what jumpers? https://theretroweb.com/motherboards/s/acorp-6via86p the ones not on the board? 😜

I was talking about the jumpers on the slotket. I have no need to alter voltage for slot1 (when not using the slotket so that's what I was thinking of.).

Also regarding previously mentioned "unstable AGP" on this motherboard. Yesterday I had 3dmark99 on the loop running for the entire day with no crashes, no artifacts etc.(with case closed) My gpu is a mx440 agp card.

Afair the problem with Via Apollo Pro 133 was it either ran horrendously slow like VIA Apollo Pro bad AGP performance? or crash prone if setup for faster operation.

As I found this thread interesting I just want to add someting to @flynth

- a 133MHz Coppermine runs fine even on 100MHz, but the Board and/or Slotkey has to support Coppermine
- every P3 ignores the multiplyer jumpers on those early P2 Boards - if your 866MHz Coppermine would work in your Mobo/Slotkey combination it would just report as P3-650 (100MHz x 6,5 Multi) no matter what multi you set
These jumpers only refer to P2 and early Celeron CPUs.
The mentioned "Pentium 2 Processor Developer's Manual" is for P2 only. The P2 may need the correct Multi set to work but I dont know that because I never had one. As I said before the P3 ignores that completly.

I have one or two of these AT Slot1 Boards myself but back in the days I never got Coppermine to work. Never tried it with a Katmai P3.
I assume all of your newly orderd CPU´s will not work on your Board if they are all Coppermines. I think this is an early P2 only Board. If it officially supports P3 CPU´s it should work with Slot1-only Katmai (450MHz to 600MHz - all with 512KB L2 Cache). If that works there is a chance with the right Slotkey that also Coppermine could work.
Hope I am wrong and you get it running.

I still believe there's a possibility that the microcode of the 866 is unknown by your board. Check this post out Re: PIII 600 bottleneck?

[Edit; actually the post before the one I linked]

OP mentioned having POST card, but we seem to all forgotten about it 😀 outdated microcode will not prevent POST card from showing a few codes.

oLdStuffUser wrote on 2022-11-23, 17:58:

As I found this thread interesting I just want to add someting to @flynth […]
Show full quote

As I found this thread interesting I just want to add someting to @flynth

- a 133MHz Coppermine runs fine even on 100MHz, but the Board and/or Slotkey has to support Coppermine
- every P3 ignores the multiplyer jumpers on those early P2 Boards - if your 866MHz Coppermine would work in your Mobo/Slotkey combination it would just report as P3-650 (100MHz x 6,5 Multi) no matter what multi you set
These jumpers only refer to P2 and early Celeron CPUs.
The mentioned "Pentium 2 Processor Developer's Manual" is for P2 only. The P2 may need the correct Multi set to work but I dont know that because I never had one. As I said before the P3 ignores that completly.

I have one or two of these AT Slot1 Boards myself but back in the days I never got Coppermine to work. Never tried it with a Katmai P3.
I assume all of your newly orderd CPU´s will not work on your Board if they are all Coppermines. I think this is an early P2 only Board. If it officially supports P3 CPU´s it should work with Slot1-only Katmai (450MHz to 600MHz - all with 512KB L2 Cache). If that works there is a chance with the right Slotkey that also Coppermine could work.
Hope I am wrong and you get it running.

Yes that is exactly right, thanks for saying so. I'm also using another Intel pdf for P3. It's title is: "Pentium III Processor for the PGA370 Socket at 500Mhz to 1.3ghz" (a weird title). It is similar to the developer manual but for P3 coppermine. Also there are articles from back in the day about making p2/p3-klamath only MB become electrically compatible with coppermine by using a modified slotket.

Perhaps you're right and there is something extra about those AT boards that prevents it from working. All the videos I saw used atx boards. Also Back in the day I remember lots of AT PCs coming in for repairs, but all we sold new was atx.

Regarding MB software CPU support, my MB happens to run Award bios 4.51 for which there is a patcher tool that let's one add extra microcode and cpu Ids I used it too, just in case, but as rasz_pl says below. If this was it, we would've hoped something /anything would show on the post analyzer card, but there is nothing, not even a C1.

There is one more thing regarding that "electrical compatibility mod" people did back in the day. It usually involves removing one pin from the slotket (am2) - this pin is GND on p2s, and unused on p3s, but thy fail to start if there is GND on it. So it is removed. Then connecting Reset with Reset2 (because those pins are swapped between p2/P3). That's it. However in theory there are at least two more pins that should have additional pull up/pull down resistors added. Original article described those as "required", but the author didn't bother with them and "every P3 he tested worked anyway". So the mod from then on "officially" included two actions only. Perhaps this only works on ATX/newer boards and some old boards do require those extra steps? If indeed none of p3s work I'll try that as a next step.

I really would prefer not to have to carry this machine to my workshop, then try finding enough free desk space to fit it, then solder lots of probes to it for the logic analyzer. It would have great educational value to me to do it, but at those frequencies (100mhz +) one has to use coax for everything, ideally matching the trace impedance etc just getting it properly hooked up sounds like a full Saturday of work .

H3nrik V! wrote on 2022-11-23, 18:02:

I still believe there's a possibility that the microcode of the 866 is unknown by your board. Check this post out Re: PIII 600 bottleneck?

[Edit; actually the post before the one I linked]

Please see above, about me patching the BIOS. The patcher shows me what cpu ids/microcodes bios supports afterwards and 0866 was there too.

rasz_pl wrote on 2022-11-23, 21:29:

OP mentioned having POST card, but we seem to all forgotten about it 😀 outdated microcode will not prevent POST card from showing a few codes.

Indeed that's what I hope. I would expect at least a C1? But who knows for sure?it would be very useful to know if there was someone who had an unsupported/not working cpu and a post card that stopped mid way.

P5 600/100 already arrived. I haven't had a chance to test yet. Hopefully I'll be able to test all of them on the weekend.

Edit: I forgot to mention this, but regarding AGP performance on this MB my 3dmark99 results were somewhere between 3200~3500 (sorry AI don't remember the exact number). The articles I read written back in the day comparing games performance had via 693 running up to 35% slower than 440bx. If anyone knows what a geforce 440mx agp with 64mb ram(on 100mhz fsb with default timings) should score in 3dmark99 please let me know (for games I'm planning to run on it it should be very fast). Also I run two gpu benchmarks from phil's computers lab dos benchmark pack (the first two). One showed 220fps,the other showed 50 (points/fps? Who knows?). Perhaps it was vsynced? So, so far it seems stable and performance seems adequate (until someone comments below it should score 30k points 😉

I think it depends on the game/software and on the cpu. In the past I did some benchmarking with two Apollo Pro 133 boards and the difference in performance between them and BX became larger in relation to increased speed of the cpu's used - meaning the faster cpu's would perform better on BX. With slower cpu's the difference became smaller.
I think 30% difference occurs only with certain types of software and is not representative for the total performance. Here a review from back in the day:

PARKE wrote on 2022-11-24, 12:09:

I think it depends on the game/software and on the cpu. In the past I did some benchmarking with two Apollo Pro 133 boards and the difference in performance between them and BX became larger in relation to increased speed of the cpu's used - meaning the faster cpu's would perform better on BX. With slower cpu's the difference became smaller.
I think 30% difference occurs only with certain types of software and is not representative for the total performance. Here a review from back in the day: 9.doc

Thanks, but this is not my chipset. This is 693A (via Apollo Pro133), my chipset is 693 (via Apollo Pro+). The main difference is that my chipset, although it can run at 133mhz it not fully supported at this fsb. 133 is considered overclocking on it. My figure of up to 30~35% came from a Tom's hardware article backed up here: http://www.thg.ru/mainboard/20000410/print.html
In the article CPU benchmarks are a bit closer. Up to 15% (depending on motherboard, there are certainly some horrible ones). With GPU the figure goes up to 35% on the worst MB (with some MB never finishing the test). So on some motherboards the opinion of "flaky and slow agp" on Apollo Pro133 is certainly true, but I'm not running this chipset.

At the time I didn't realise a difference between Apollo Pro133 and Apollo Pro Plus. Despite searching a lot I haven't found any benchmarks at all for motherboards with this chipset, but I searched by the chipset name. Perhaps one would have more luck finding motherboards that use this chip and searching for their benchmarks. I might do that later.

I found this interesting feature comparison with 440bx from 1998 (before 440bx officially supported 133mhz fsb) and Pollo Pro Plus. http://www.thg.ru/mainboard/20000410/print.html

Also, I found a wiki page that claims this chip contrary to Apollo Pro133 doesn't support P3. I think it may not officially support it, but this is because it was released in the beginning of 1998. By the end of that year via had Apollo Pro133 so they started pushing it instead. Still I think p2 and P3 are similar enough a chipset that supports P2/celeron should be able to run P3 with some persuasion.

So I can no longer say with certainty it is "up to 35% slower" it may be better or worse. Until I see some other benchmarks I can only compare my own results. I would love to find some other 3dmark99 results of a geforce mx440 running on intel 440bx at 100mhz fsb. But unfortunately by the time geforce mx440 was released it was 2002. By that time there was another version of 3dmark(2001?). And everyone running a "performance" build had 133mhz fsb. So the numbers can't really be compared. I read somewhere just increasing the fsb from 100 to 133 is "worth 1500 3d mark99 points". Perhaps at some point I'll try overclocking it to 133fsb and running 3dmark2001 (if it is even possible on win98). That would give us some interesting answers.

Edit: I forgot to mention it is very interesting the older of the two (apollo Pro133, apollo pro plus) supports better sdram timings. In fact apollo pro plus supports same timings as 440bx(in mid 1998) while Pro133 is slower. I wouldn't be surprised if Pro133 was the same exact chip, just with validation done for 133mhz and sdram timings dropped to help it at that frequency.

flynth wrote on 2022-11-24, 14:18:

Thanks, but this is not my chipset. This is 693A (via Apollo Pro133), my chipset is 693 (via Apollo Pro+). The main difference is that my chipset, although it can run at 133mhz it not fully supported at this fsb. 133 is considered overclocking on it. My figure of up to 30~35% came from a Tom's hardware article backed up here: http://www.thg.ru/mainboard/20000410/print.html

Ouch.... I totally missed that. Sorry for the confusion.

flynth wrote on 2022-11-24, 09:47:

Also Back in the day I remember lots of AT PCs coming in for repairs, but all we sold new was atx.

at that point in time AT was relegated to upgrades and extreme budget (aka cheapest possible K6 shitboxes)

flynth wrote on 2022-11-24, 09:47:

I would expect at least a C1?

yes

flynth wrote on 2022-11-24, 09:47:

Edit: I forgot to mention this, but regarding AGP performance on this MB my 3dmark99 results were somewhere between 3200~3500 (sorry AI don't remember the exact number). The articles I read written back in the day comparing games performance had via 693 running up to 35% slower than 440bx. If anyone knows what a geforce 440mx agp with 64mb ram(on 100mhz fsb with default timings) should score in 3dmark99 please let me know (for games I'm planning to run on it it should be very fast). Also I run two gpu benchmarks from phil's computers lab dos benchmark pack (the first two). One showed 220fps,the other showed 50 (points/fps? Who knows?). Perhaps it was vsynced? So, so far it seems stable and performance seems adequate (until someone comments below it should score 30k points 😉

Is my performance normal in comparison to the components i have?
440mx ~= Geforce 2 TI
~6000 on 600-700MHz is reasonable

rasz_pl wrote on 2022-11-25, 05:24:

Is my performance normal in comparison to the components i have? 440mx ~= Geforce 2 TI ~6000 on 600-700MHz is reasonable […]
Show full quote

flynth wrote on 2022-11-24, 09:47:

Edit: I forgot to mention this, but regarding AGP performance on this MB my 3dmark99 results were somewhere between 3200~3500 (sorry AI don't remember the exact number). The articles I read written back in the day comparing games performance had via 693 running up to 35% slower than 440bx. If anyone knows what a geforce 440mx agp with 64mb ram(on 100mhz fsb with default timings) should score in 3dmark99 please let me know (for games I'm planning to run on it it should be very fast). Also I run two gpu benchmarks from phil's computers lab dos benchmark pack (the first two). One showed 220fps,the other showed 50 (points/fps? Who knows?). Perhaps it was vsynced? So, so far it seems stable and performance seems adequate (until someone comments below it should score 30k points 😉

Is my performance normal in comparison to the components i have?
440mx ~= Geforce 2 TI
~6000 on 600-700MHz is reasonable

Thanks for the link (I didn't see it before). I wish someone tested Geforce mx440 on slower cpus. Specifically 350mhz p2. Of course the cpu and the card are almost from different eras (4 years apart) so it is unlikely anyone would. If I do get a faster cpu to work it will be interesting to compare.

It appears Geforce mx440 is bottlenecked by the cpu below ~0.8-1GHZ and around 6.2k 3dmark99 points seems a true maximum for the card on 100mhz fsb (at this moment it being bottlenecked by the fsb).

When I was collecting parts for this system I didn't check the year mx440 came out. I seemed to remember it being "roughly right", but clearly the gpu is too new for a period correct system(before 2000,win98). CPU-wise P3 800 is the fastest one released before 2000, but if we take winxp release date as a cutoff even a p4 2.4Ghz would be fine. Then considering many people (including myself) didn't upgrade to winxp until sp1 came out towards the end of 2002 we were well into 2.8ghz p4 time. It is interesting one can build a p2 266 mhz system or a p4 2.6ghz system and both can be correctly described as being from a win98 era 😀

Coming back to GPUs, having found the 3dmark99 megathread my conclusion is that if one is running a cpu below 400mgz on 100mhz fsb (or 66mhz) one is usually seriously bottlenecked by the cpu/fsb. Even cards like Geforce 256 get 2400 points in such system, a result similar to riva tnt 16mb 😁 (I haven't found anyone testing mx440 with p2, geforce 256 is the fastest card I found a benchmark of with a p2 so far) So there is really no point sticking a Geforce Mx440 in such system unless that's what one already has(or got it for free).

Having said that I think I'll try it with passive cooling only. There is no need to replace a fan with a quieter one as planned when it is running at 40% of its max.