Squid1

The overclocked cards are probably just more likely to expose the problem in a noticeable fashion. There are sufficient reports of non-overclocked cards to tell that it's not just overclocking.

I don't like this style of layout, with the CPU at the front and the RAM between the back of the board and the CPU. If you have a normal power case with the harddrives mounted in a cage around that area you cannot mount big HSFs at all, and you cannot remove you harddrives even with normal HSFs.

Which P4 connector and which PCIe connector exactly are you using there?

So, does eVGA actually have cards at the same clocks that do not hang? Or are they just sending out new products hoping this one will hold?

I have a similar drive just the 160, but from the same generation. Loud and slow is a good summary. (I think write speed was like my seagates but read speed was noticably slower, might have been the other way round). Trouble with MIR, too.

The 12V rails are actually at about 12.07-12.09 V measured with a DMM using a very demanding rig under full load. My Zippy was a little better at 12.03-12.05 or somesuch but I don't think you can say the OCZ has stability problems. And at least mine does not have +/- 5%.

Mr. Cynical says the most likely explanation for the Infinity to show up on NVidia's list and the SLI-DR/Ultra-D is not is that NVidia is too stupid to tell the DFI board lines from each other and they actually tested a SLI-DR. (Which of course would be rooted in DFI's insufficient naming scheme, the SLI-DR/Ultra-D are practically nameless).

FSP changed the specs on the Epsional rails several times and went through different revisions of the specsheet. The problem is that they couldn't make up their mind at which temperature they should give the load specs and what exactly peak means.

Sorry, distributed deadlocks are not a SMP or single processor machine OS kernel level concept at all. As the name says, they deal with deadlocks between machines, and unless you have a very sophisticated cluster OS that is done in userlevel, not in the kernel.

Uh, that sounds a little off. There is nothing in particular dealing with "deadlocks in a multi-user environment". Unix kernels are either multithreaded with the normal thread synchronization primitives or they are interrupt-driven with an elevator scheme, using pretty much the same synchroniztion primitives. Whether there's different users sitting on the different userland processes entering the kernel has nothing to do with synchronization management.

OK, so this is dual-boot and the internet connection is one of the LAN ports to the cable modem or whatever? I have no problems dual-booting a variety of NForce-GbE boards between Win2K and Linux. However, I pretty much always have the newest drivers. The NForce GbE driver was broken and semi-broken very often.

Describe your network setup, how's your internet connected and how are the PCs connected?

It is only NVidia's broken hardware firewall that causes this. Turn it off and fixed are the CRC errors.

Benchmark 1T versus 2T: http://cracauer-forum.cons.org/forum/viewtopic.php?p=65#65 Again, you don't have as stict requirements for similarity for the pair as you have within the pairs. And "same vendor" doesn't buy you anything at all, it can still be different chips and other differences that ruin the show.

The DFI SLI-DR/Ultra-D won't do three sticks because the two CPU slots closer to the CPU do not do any single-channel memory at all, neither 1 nor 3 sticks. Other boards do that, but of course it's no big deal. There are boards that do dual-channel with one module by splitting that module into two parts. I don't remember whether those were socket 939 and/or whether they do 3-module dual-channel.