Just in time for Christmas…the postman handed me a package containing an Athlon 3000+

It posed straight away at 133x17 (2266MHz)

Started some stability tests. So far it looks more stable than the 2600+ in so far as it can actually run software and it doesn’t immediately crash/reboot at this speed.

FullYes wrote on 2025-12-24, 19:02:

Just in time for Christmas…the postman handed me a package containing an Athlon 3000+

It posed straight away at 133x17 (2266MHz)

Started some stability tests. So far it looks more stable than the 2600+ in so far as it can actually run software and it doesn’t immediately crash/reboot at this speed.

Is it one with unlocked multiplier?

^Yes. It’s build week is 16 of 2003 so it’s unlocked. I’ve only been looking for unlocked CPUs. I would take a locked AXDA2800DKV4C to see how that behaved with its factory 16x multi, but I’ve never seen one for sale.

I left the machine running overnight at 133x17 (2275MHz according to CPU-z ) with memtest86. 13 passes with no errors

Unfortunately I’m seeing some issues with stability in prime95. A bit unusual. Instead of reporting an error, the system hard locks. Sometime after a few mins, sometimes after longer. I guess it depends on the test. It also does it at 124 x 17 which is LOWER than the CPUs rated frequency. If anything, this setting is even more unstable than 133 x 17

So, new theory time. This magic 17x133 multiplier/fsb setting that posts, isn’t stable.

I’ve been running 140 x 16 and that seems stable. No errors in prime 95 after 8 hours. This is (according to CPU-Z) 2248MHz. I’ve bumped the FSB again now to 142MHz which gives me 2280MHz and I’m seeing how stable this is.

My oscilloscope and post code analysers have arrived so I can do some more fundamental tests now on the buses and the voltage rails to see if any other issues show up

Nope. I was wrong. 142 x 16 hard locks the machine again on prime95 after a while, so it seems I might be at the limit of this CPU. Damn!

I can’t explain what was up at 124 x 17 though. 🤷🏼‍♂️

Now trying 141x16 which should be 2256MHz but in CPU-Z is reporting 2266MHz.

I might get time tomorrow to check some base freqencies. I don’t remember having such a variance between actual freqency and target freqency on any motherboard. Ok it’s only 0.5% or so, but I’m used to stuff being much closer. Maybe I can change the crystal or the clock generator

I wouldn't run it at that frequency without a voltage bump. Prime95 may not test the CPU fully, it is not a replacement of manufacturer's validation tests. Typically one would find the highest stable frequency and then decrease the frequency by 100Mhz to account for lack of comprehensive tests. I ran my Athlon XP at 2.3Ghz at the time, max frequency was 2.4Ghz.

Am I correct in thinking that you are intending to have 3 retro PCs? The Pentium MMX for DOS, the Athlon XP for late DOS and Win98, and an i7 PC for XP?

If so, then I am not sure if you need to have a CPU as fast as the Barton in this Socket A system. The 3DMark scores you posted are extremely high for Win98 gaming.

I get that it's fun playing around with hardware, so keep at it whilst it's enjoyable. But don't feel bad if if you need to use a slower CPU. It will still be a killer Win98 system.

Yeah I’m already with you on that. I’ve gone down a bit of a Rabbit hole just trying to get this setup to work the way I thought it would. Since it’s not really playing ball, I am going to reverse the multiplier mod and go back to stock I think. As you say, I have a faster machine if I need it.

It hard locked after about 18 hours running at 2266MHz.

I’ve been mainly gaming on the Pentium system due to the issues with the athlon!