H3nrik V! wrote on 2023-12-28, 14:13:

Oh, darn it. The SL2KA isn't with ECC L2 cache, which it looks to be necessary for transplanting P3 cache to it ...

That's a shame. L2 cache definitely seems to be the limiting factor. My CPUs will run happily at 450MHz (4.5 x 100) with L2 disabled in BIOS, although they don't get past the POST screen at any 500MHz configuration (5 x 100, 4.5 x 112 or 4 x 124).

EDIT- 466MHz (3.5 x 133) is also stable with L2 cache disabled. Seems like all the best overclocking options for this unlocked CPU are at the stock multiplier!

sfryers wrote on 2023-12-29, 09:32:

That's a shame. L2 cache definitely seems to be the limiting factor. My CPUs will run happily at 450MHz (4.5 x 100) with L2 disabled in BIOS, although they don't get past the POST screen at any 500MHz configuration (5 x 100, 4.5 x 112 or 4 x 124).

You could try out the tool "WCPUL2CacheLatency". Try higher latency settings than the default one out and see if this can help you.
You can change the L2 Cache Latency under Windows and there is also a DOS version (some mainboards,like the Abit BX mainboards with the Softmenu allows the L2Cache Latency setting in the bios):

http://falconfly.3dfx.pl/info/readme-wcpul2.txt

Edit: Also a higher VIO Voltage could help for a better stability, because all external L2 Cache versions of the PII/III Katmai are working with 3,3V VIO. Maybe need a better cooling at ~3,5V VIO.

sfryers wrote on 2023-12-28, 12:22:

I acquired a 350MHz Deschutes PII (SL2S6, 1998 week 43, Ireland) a few days ago and was very surprised to find that it has an unlocked multiplier. It's not a particularly remarkable overclocker- it runs stable at 400MHz (4 x 100), but crashes at 450MHz (4.5 x 100) shortly after loading the Windows 98 desktop.

Somewhat ironically, its fastest stable speed appears to be achievable using the stock multiplier, at 3.5 x 124 = 434MHz.

Other than the unusually late production date for an unlocked PII, this CPU performs much the same as my earlier unlocked Deschutes, a 333MHz version (SL2KA, 1998 week 01, Philippines).

Maybe, it will need more juice. Try it on motherboard that can raise voltage to 2.2v or 2.3v. Maybe at that voltage, it will run 450 Mhz.

I have also interesting finding with Pentium II 333. (week 36 of '98) It was locked on some slot 1 motherboards, that I've tested. I wanted it to be unlocked downwards, so it can be set to 100Mhz FSB. Interestingly enough, on Abit VA6, multiplayer could be changed downwards. It worked even 4x100 = 400Mhz. Not sure why only on that mobo.

W.x. wrote on 2023-12-29, 12:43:

Maybe, it will need more juice. Try it on motherboard that can raise voltage to 2.2v or 2.3v. Maybe at that voltage, it will run 450 Mhz.

Only the CPU core runs with 2.0V stock on the Pentium II Deschutes and PIII Katmai (2.8V for the PII Klamath) , the external L2 cache runs with 3.3V, so rasing the VIO to 3.4V-3.5V is the way to go for a better L2 Cache stability, and also playing with the L2 Cache latency, as I posted before.

The Celeron Mendocino Slot 1/S370 for example has a on-Die L2 Cache of 128kb and operates with 2V VCore, because the L2 Cache is build in the CPU core.

Some Abit motherboards had a BIOS option to force #sel66/100 signal (AFAIR that's the name of it) to make a 100MHz fsb chip think it ran at 66. I guess the idea was that some chips then could use other multipliers ...

The Serpent Rider wrote on 2023-09-12, 14:07:

Well, not sure about all Klamath, but PII 233 I have (1998 week 16) use identical L2 chips to PII 300 and can be overclocked slightly better. I think it's safe to say that all Klamath from 1998 are identical.

I think you're lucky with this Klamath 233. I just tried my unlocked PII Klamath 266 at 3.5 * 100 MHz and the system got frozen even before the CPU got hot. At 3 * 100 MHz it runs rock stable.

I have 300Mhz klamath on asus p2b motherboard that I can slow down to either 133Mhz or 166Mhz, but I have not really sure what that is good for. I dont think I have found any single game or software that would not work with 300Mhz but would work when slowed down with multipliers.

NostalgicAslinger wrote on 2023-12-29, 12:25:

You could try out the tool "WCPUL2CacheLatency". Try higher latency settings than the default one out and see if this can help y […]
Show full quote

sfryers wrote on 2023-12-29, 09:32:

That's a shame. L2 cache definitely seems to be the limiting factor. My CPUs will run happily at 450MHz (4.5 x 100) with L2 disabled in BIOS, although they don't get past the POST screen at any 500MHz configuration (5 x 100, 4.5 x 112 or 4 x 124).

You could try out the tool "WCPUL2CacheLatency". Try higher latency settings than the default one out and see if this can help you.
You can change the L2 Cache Latency under Windows and there is also a DOS version (some mainboards,like the Abit BX mainboards with the Softmenu allows the L2Cache Latency setting in the bios):

http://falconfly.3dfx.pl/info/readme-wcpul2.txt

Edit: Also a higher VIO Voltage could help for a better stability, because all external L2 Cache versions of the PII/III Katmai are working with 3,3V VIO. Maybe need a better cooling at ~3,5V VIO.

I dug out my Abit BX6 to experiment with the L2 cache settings- my previous testing was carried out using a Gigabyte BX2000. Interestingly (and unfortunately), the 350MHz Deschutes acts as though it has a locked multiplier on the BX6; all Softmenu settings result in a 3.5x multiplier.

Nevertheless, by FSB overclocking alone, the CPU works fine on the BX6 at up to 452MHz (129 x 3.5) with L2 cache disabled. However, it appears that no cache latency settings are stable at this speed. Setting latency to 6, 7 or 8 (default seems to be 5) allows the Windows desktop to load but crashes shortly after starting 3DMark 2000. Setting latency to 9 or above appears to disable L2 cache entirely, with CPU-Z recognising it as a Celeron.

sfryers wrote on 2023-12-30, 11:05:

However, it appears that no cache latency settings are stable at this speed. Setting latency to 6, 7 or 8 (default seems to be 5) allows the Windows desktop to load but crashes shortly after starting 3DMark 2000. Setting latency to 9 or above appears to disable L2 cache entirely, with CPU-Z recognising it as a Celeron.

Interesting, so it does change anything (a little bit more stable), but not really helpful,... so the last option would be to change the VIO Voltage to 3.4-3.5V.

NostalgicAslinger wrote on 2023-12-30, 12:09:

sfryers wrote on 2023-12-30, 11:05:

However, it appears that no cache latency settings are stable at this speed. Setting latency to 6, 7 or 8 (default seems to be 5) allows the Windows desktop to load but crashes shortly after starting 3DMark 2000. Setting latency to 9 or above appears to disable L2 cache entirely, with CPU-Z recognising it as a Celeron.

Interesting, so it does change anything (a little bit more stable), but not really helpful,... so the last option would be to change the VIO Voltage to 3.4-3.5V.

Unfortunately the motherboard doesn't appear to have an option to change VIO, and I'm not intending to perform any hardware mods. I think this is as far as I'll be able to push these Pentium IIs.

I remember my PII-333 (that one from this thread), with 4ns cache was able to boot at 4 x 100, but had no luck at 4.5 x 100 - crashes early during POST at stock VCore, and hangs at WinXP desktop at 2.4V. My board (Lucky Star 6ABX2V) can't tweak the VIO voltage, I might try that later with better cooling (had to use an AM2 heatsink + 92mm fan, since I didn't have a proper Slot1 cooler :)

NostalgicAslinger wrote on 2023-12-29, 12:25:

You could try out the tool "WCPUL2CacheLatency". Try higher latency settings than the default one out and see if this can help you.
You can change the L2 Cache Latency under Windows and there is also a DOS version (some mainboards,like the Abit BX mainboards with the Softmenu allows the L2Cache Latency setting in the bios)

shameless plug: P6Cache can also change the L2 cache latency, besides making it work at low multipliers :)

Now I'm wondering, what if I try 2.5 * 133 MHz (333 MHz) with my Klamath 266 oldie? Its L2 cache chips run at half of the CPU's speed, so this will be better for them than 3.5 * 100 MHz (350 MHz). But maybe in this case the CPU core will not sustain the FSB speed increase from original 66 MHz without any serious overvolting...