Which caps to replace: ECS P4M800Pro-M V1.0A

[Mattz_GT]

the pentium 4 and pentium d cpu design called the netburst microarchitecture is designed to scale with more mhz or ghz, so the faster ram speed is more helpful than tighter timings. try doing a mem bandwidth test or benchmark like in sisoft sandra, aida64 or maxxmem2. u should have overall higher memory reads, writes and copy speeds at 667 5-5-5 than u do at 533 4-4-4.

or u can just try to get low latency ddr2-667 at cl4. most of not all of the low latency ddr2-1066 and ddr2-800 from the good brands like kingston, g-skill and mushkin etc. seem to be all but sold out on junkbay. but i noticed plenty of unwanted low latency ddr2-533 and 667 available on junkbay. if u're lucky, it may even do ddr2-667 speeds at 3-3-3-9 or 3-4-4-10. the tighter timings also result in slightly higher mem bandwidth which results in better performance. hope this helps in squeezing out every inch of performance out of your system which is the point of any overclocker. hehehe

Double trouble with the RAM, I struggled to find the required low density 2R8, anything faster is probably going to be high density, anyway, kit at the ready (hmm, maybe I ought to get some flux... otherwise I know that putting a bit of fresh solder on can help melt the old solder, hopefully this predates RoHS anyway.
Wish me luck, I'm going in, well, tomorrow anyway

[momaka]

^ Flux is a must for me.
Yes, you can (and should) use fresh solder on top of the old leads to improve the heat transfer from your soldering iron to the board/joint. But once you have the cap pulled out, you might find that clearing the hole can become a bit of a "mess" if you just keep adding solder. At some point, just adding flux and heating with your iron is more helpful than adding fresh solder.
As for clearing the holes, I prefer doing it the "manual" way, with a (stainless steel) sewing needle - basically heat each hole with the soldering iron and then push through the hole with the sewing needle from the other side to push the solder out. For boards with really thick copper planes, I find neither desoldering braid nor a vacuum pump / solder sucker to be of any use - some leftover solder always tends to stay in the hole.

yes i think that was the problem. iirc gp caps arent meant for smps use. the high ripple generated by the switching frequency of the smps will easily fry any gp cap. for smps, caps which are said to be low-esr or better should be used.

Exactly.
OST RLG tend to be pretty stable when used in low-stress circuits... at least that's been my experience with them so far.

well, u could swap between the two gpus as needed... depending on if u wanna play on win98 or winxp.

The only problem with that is if you do it often enough, it will eventually wear the contacts in the AGP slot and it might stop detecting cards properly after a while. I'm running into this issue already with the PCI-E slot on my video card test PC - it has about only 10-15% chance it will detect a GPU now... and quite often, not always in full 16x mode. So I've started using PCI-E 1x-to-16x crypto miner risers to test GPUs. Unfortunately, this has the side effect of the card not being able to reach its full potential - even under a single core Athlon 64 4000+. And I can't know if a GPU has problematic PCI-E lanes or not. But at least this I can still test if the rest of the GPU portion (and connected memory) is OK.

[Mattz_GT]

Getting some Nichicon polymer 1000uF 6.3V for the RAM VRM, so the CPU VRM out and in are getting changed, and the RAM VRM, I don't think I'll bother with the general decoupling ones all over the board unless it's still not stable