trodas wrote:

You are welcome. I was somewhat surprised, that with the caches on, even deadly slow 10.7MHz Pentium perform about 33MHz i386. IPC gains, I assume and the 512kBy of soldered L2 on the board. Off they go and 1.6 is there.
I was badly surprised that memory settings did not played any role. I man... I set the memory settings to slowest (forget 2-2-2 burst, try 4-4-4 ... forget CAS 2, try CAS3 ... etc) and there was absolutely no difference. I hoped for 1.4 or 1.2... but no. Suxx.

If you haven't tried already, besides memory latencies, I'd try going through the BIOS and disable anything that sounds remotely 'speedy' - like shadowing BIOS and Video, PCI waitstates on so on... This might shave a few more off your benches.

BTW, I figured out how to use LandMark 2.0 without input:

... Use "LM20 /b"

trodas wrote:

The K5-75 is classic 50x1.5 CPU. It can be overclocked to 60x1.5 but that it is. 66x1.5 fail to post... What is even worser, at 7.14MHz FSB it also fail to post completely. And I already masex out the Vcore (3.5V), so there is nothing I can do. No way K5 post at such low FSB settings 🙁

It could just be this particular CPU. If you have others, it maybe worth trying those out, especially if you can get 1x multi working.

trodas wrote:

Maybe K6 can work at such low FSB clocks? Who knows... but I definitively will try it. You know, K5 is not even a x86 CPU, but rather experiemental superscalar AMD cpu, that get the x86 instruction decoder in front of it and that it is... Maybe that is where it fails?

Another possibility will the the AMD K6-2, witch is not mentioned in the supported CPU, but I quess that it can work, or the bios can be updated to make it work... It does work in Asus TXP4 mobo anyway: http://hwbot.org/submission/2373822_

It would be interesting to get results for any CPU that POSTs, especially K6-II+/ k6-III+ CPUs (you can use Setmul on those to change multi and cache options). My results only go to XT speeds for K6-III+ CPUs, where the Turbo function literally destroys its performance in a good way. Presumably due to its on-die L2 cache.

trodas wrote:

So there IS a chance for AMD cpu to get into superlow clocks. But I would want CPU, that can be forced to low multipliers. I heard that Pentium Overdrive can be switched to x1 multiplier by disconnecting the fan. Then it switch to x1 in order to prevent overheating... Never tried it myself, but sounds reasonable and people used this there:
http://www.winhistory.de/more/386/xpmini_eng.htm

The thought of installing an OS like XP did occur to me, then I came to my senses. The only reason I can see to use speeds this slow is to run speed sensitive apps/ games... I can see the challenge though.

trodas wrote:

Another way is mainboard, that have good frequency control. The Asus TXP4-X use as clock generator the ICS 9169 PLL. That chip, however, cannot be controlled by software, because "do not have a System Management Bus connection or the FSB could not be set via System Management Bus. So there is and will be no program that supports these PLL's!!! There is absolutely no chance!"
http://www.cpufsb.de/FSB.HTM

This is a nice DOS utility to change FSB, but it has very limited PLL compatibility:

SMB.ZIP ver. 2.09: http://rayer.g6.cz/programm/programe.htm

trodas wrote:

But for some, 2MHz is possible on CPU too: http://www.youtube.com/watch?v=y3wkhDfzqlo (FIC 486-GAC-2 mainboard)

It would be interesting to see this setup benched relative to an XT 4.77MHz...

Tertz wrote:

I'd call speed close in +-15% difference. More difference can be noticed clear, hence is not good for games linked to the timer.
I like the idea to find a single machine for retro games. The ideal result should support games from 1981-2001 times: full DOS/Win9x period, XT 4.77 - P3 600. I see philscomputerlab's researches in this direction. You do similar research. Some parts may be changed in the using of "retro machine", besides MB, drives, case.

What you are looking for in a single build is kind of a Holy Grail to find. I know a build from XT 4.77MHz - K6-III+ 450MHz is possible. And on paper, a build from XT 4.77MHz - P3 1050/ 1400MHz is also possible, but finding working parts is next to impossible...

j^aws -

I was badly surprised that memory settings did not played any role.

If you haven't tried already, besides memory latencies, I'd try going through the BIOS and disable anything that sounds remotely 'speedy' - like shadowing BIOS and Video, PCI waitstates on so on... This might shave a few more off your benches.

Looks like a good idea. It would take some time, tough, as I'm in the middle of SuperPi 32M benches and I hit few bumps and have to redo the calculation and we are talking about a whole day there with K5 ... So when the job is done, then I will try it.
With Pentium 90 @ 112.5MHz it took 17h http://hwbot.org/submission/2962892_
With Pentium 90 @ 99MHz it took 19h http://hwbot.org/submission/2965531_
...and with K5 75 at 90 and 75MHz it took over day to finish 🙁 (and then crash on saving the file, because of not enought memory...!)

Oh, well, hard time to bench so old systems 😊

BTW, I figured out how to use LandMark 2.0 without input: ... Use "LM20 /b"

Thanks! Will do!

The K5-75 at 7.14MHz FSB it fail to post completely

It could just be this particular CPU. If you have others, it maybe worth trying those out, especially if you can get 1x multi working.

Could be, but no. I have only one AMD K5 75 🙁

Another possibility will the the AMD K6-2, witch is not mentioned in the supported CPU

It would be interesting to get results for any CPU that POSTs, especially K6-II+/ k6-III+ CPUs (you can use Setmul on those to change multi and cache options). My results only go to XT speeds for K6-III+ CPUs, where the Turbo function literally destroys its performance in a good way. Presumably due to its on-die L2 cache.

I have to admit that it was the Setmul that draw my attention to this forum. I would love to be able to set 2x multi (that would give 14.3MHz), but I need / must be able to do that in Windows for the CPU-Z validation. No validation, no score. Simple as that... And I chatted a bit by PM with gerwin (Setmul autor) about it and he say it is possible to do it and it should not be (maybe) even that hard, so I waiting since this day for ANY beta to test 😀 Nothing yet...

But you say that I can use just regular AMD K6-II+ CPU like this one:
http://www.ebay.com/itm/AMD-K6-2-550-MHz-SOCK … =item3cfcf64b09
...and I can change the multi to 2x, so if it post at 7.14MHz FSB, then I have 14.3MHz CPU? That would be nead. Pls confirm... so I can order the CPU right now 😲

I want CPU, that can be forced to low multipliers. I heard that Pentium Overdrive can be switched to x1 multiplier by disconnecting the fan.

The thought of installing an OS like XP did occur to me, then I came to my senses. The only reason I can see to use speeds this slow is to run speed sensitive apps/ games... I can see the challenge though.

I understand. It is a painfully slow process to be sure. However it is possible, I done it on P90 as well, as on AMD K5 😀 And as I say, all I care there is the possibility to a) produce 12+h SuperPi 32 runs (for some reasons, killing caches on other 100% stable mainboards produce crashes instead of 12+h results, usually around loop 6-7 for AMD, loop 10 for P4... more in separate theread there: Crashes when caches are disabled - WTF?! ) b) CPU-Z validation on super-low clocks (req. Win2k or WinXP, not W98 🙁 ).

Asus TXP4-X use as clock generator the ICS 9169 PLL. That chip, however, cannot be controlled by software...

This is a nice DOS utility to change FSB, but it has very limited PLL compatibility:
SMB.ZIP ver. 2.09: http://rayer.g6.cz/programm/programe.htm

I know the autor Rayer 😀 However that did not change the fact, that there is no way to control this ICS 9169 by software, as it is claimed. SMB also support southbridge Intel 82371 (PIIX4), so it might be a good idea to try it 😀

But for some, 2MHz is possible on CPU too: http://www.youtube.com/watch?v=y3wkhDfzqlo (FIC 486-GAC-2 mainboard)

It would be interesting to see this setup benched relative to an XT 4.77MHz...

Me too 😀 But who will buy it? 😀 It require special mainboard and special Socket 3 Pentium Overdrive... and other that todays normal ATX PSU, PS/2 keyboaard, etc. Sounds pretty complicated to me, but the challenge to 2MHz is fantastic...! Maybe there is a way how to convert and use normal devices in such maniboard... somehow.

trodas wrote:

... But you say that I can use just regular AMD K6-II+ CPU like this one: http://www.ebay.com/itm/AMD-K6-2-550-MHz-SOCK … =item3 […]
Show full quote

...
But you say that I can use just regular AMD K6-II+ CPU like this one:
http://www.ebay.com/itm/AMD-K6-2-550-MHz-SOCK … =item3cfcf64b09
...and I can change the multi to 2x, so if it post at 7.14MHz FSB, then I have 14.3MHz CPU? That would be nead. Pls confirm... so I can order the CPU right now 😲

Yep, Setmul can change multipliers for K6-II+/K6-III+ CPUs, as noted in this thread, where various other CPUs are listed: SetMul - Multiplier control for VIA C3 / AMD K6+7+8 Mobile / Cyrix 5x86

I've only tested K6-III+ CPUs as low as 2x multi (don't have any K6-II+ CPUs). Of course, this assumes your motherboard can POST with such a CPU...

trodas wrote:

I understand. It is a painfully slow process to be sure. However it is possible, I done it on P90 as well, as on AMD K5 😀 And as I say, all I care there is the possibility to a) produce 12+h SuperPi 32 runs (for some reasons, killing caches on other 100% stable mainboards produce crashes instead of 12+h results, usually around loop 6-7 for AMD, loop 10 for P4... more in separate theread there: Crashes when caches are disabled - WTF?! ) b) CPU-Z validation on super-low clocks (req. Win2k or WinXP, not W98 🙁 ).

I've never tried SuperPi 32 to benchmark slow CPUs. When I get some time, I'll try it out around XT speeds.

trodas wrote:

I know the autor Rayer 😀 However that did not change the fact, that there is no way to control this ICS 9169 by software, as it is claimed. SMB also support southbridge Intel 82371 (PIIX4), so it might be a good idea to try it 😀

I'll check out PIIX4 support. It's nice to use software to control FSB, especially in DOS. However, I'm sure you are aware, it's trivial to setup switches connected to relevant jumpers to achieve similar results.

trodas wrote:

Me too 😀 But who will buy it? 😀 It require special mainboard and special Socket 3 Pentium Overdrive... and other that todays normal ATX PSU, PS/2 keyboaard, etc. Sounds pretty complicated to me, but the challenge to 2MHz is fantastic...! Maybe there is a way how to convert and use normal devices in such maniboard... somehow.

You could probably track down the PLL for that board and hack it to work with more modern boards...

IIRC, some forum members here have slowed that board down to those speeds, but not sure if they used a POD to do this. I have all the parts besides the board, so if ever I come by a similar board, I'll bench it relative to an XT.