Just another successful mod, which I did on an overpriced AMD-K6-2+/570ACZ (570 MHz, 2.0 V originally). Paid way too much, 75 €, for the CPU thanks to the eBay's bad search system. It would have been great to have stumbled upon this thread before buying the CPU.
Relevant test system specs:
- Asus P5A-B Rev 1.04 (new polymer electrolytic capacitors, a.k.a. solid state capacitors), BIOS: 1011 Beta 005
- 384 MB PC-133 @ 100, CAS 3
- Windows 98 SE, with unofficial SP3-LITE 01-30-2019
Unfortunately, I did not do any before takes, so only the results are briefly mentioned here. After the mod, I have been running the CPU at 600 MHz (6x100 MHz) and 2.0 V, seems to be fine with that. Stability tested running Prime 95 Version 21.3.1 for couple of hours with K6 optimizations enabled. Also have run 3dmark 99 max and 3dmark 2000 several times after the mod, played briefly some games and done other random stuff. No crashes so far (after fixing some initial hickups mentioned later). See the cpu-z vintage and 3dmark99 max pro results in the attachment.
Couple of issues I encountered after I first did the mod:
- The system was unstable, any 3dmark run was guaranteed to fail. Also other crashes were happening, and even windows 98 start files got corrupted... Did not encounter these issues when running the CPU as basic K6-2+ before the mod.
+ This was fixed by fully recapping the motherboard with X-CON 6.3V 1000 uF, 7 mOhm ESR, polymer capacitors. (NOTE: DO NOT DO THE SAME! Requires additional modifications to the regulator circuits for the board to be fully stable with solid polymer capacitors.)
- There was RAM errors that were not encountered when the CPU was running as K6-2+.
+ For some reason, CAS setting had to be changed from 2 to 3 to get the RAM to work stable. RAM was running at 100 MHz frequency in both cases.
Edit / later addition: I have modified the Chipset & RAM, and CPU regulator controllers feedback compensations circuits since the original post, as required for the solid-polymer capacitors with very low ESR, and it seems I am now able to even run the RAM at the tightest CAS 2 setting again (2-2-2 / 143 MHz RAM timing setting in bios). Note that in the bencmark picture, CAS 3 setting was used, which caused about 2 % performance hit at least in 3DMark.
- On another note, not related to this mod; It seems the P5A-B is unstable in any configuration if PCI Delayed Transaction is enabled and there is a DMA capable ISA-card present. I think this might be quite a bummer, especially if using PCI-video cards. Now I am using Geforce 4 MX 440 on AGP and Voodoo 2 on PCI. Have not yet tested and compared how Voodoo 2 performance is affected with PCI Delayed Transaction disabled.
For mod testing, I just cleaned the cpu die, and the silicone at the corners of the IHS, and the corresponding spots on the ceramic casing. What remained of the silicone mold in the middle, nicely slotted the die with the heat spreader during testing so that it was in the correct spot. Just added a very small drop of Noctua's heat sink compound there. The CPU and the heat sink can also be seen in the attachment, during the phase when I had to remove the CPU while I was recapping the motherboard.
Edit: Since the modded CPU has been working pretty well, I also permanently glued the IHS from the corners back to the CPU with heat conductive silicone glue. I would not recommend thermal silicone glue for the die though, as I think that it is not an appropriate application for this. I only put some noctua's thermal paste between the CPU die and the IHS. Based on my experiences, good heat sink compound / paste beats most heat conductive glues, with the added bonus in this case that it wont risk the CPU die, if the CPU needs to be delidded ever again...