Additionally, I had to raise vCore (1.36 range?) and may even have tweaked vDIMM just a bit (1.84 range?). In my case, the limiting factor was the vFSB, which I seem to recall had to be cranked way up (somewhere in the 1.45 range?). Trying to push above 3.8Ghz causes the CPU regulator to kick in and drop the multiplier down in order to reduce temps (in other words, I've hit the ceiling). 100% load under Intel Burn Test sends the CPU temps into the low 80's. On this second build, the best I've been able to get on air was 3.8Ghz, stable through 10 iterations of Intel Burn Test. I use the box as a server, so it's only got a cheap $25 video card with a 40mm fan. On the northbridge I've got Thermalright HR-05 with 70mm(?) fan. On each CPU I've got Thermalright Ultra 120 Extreme with 120mm fan. I've since put togeather a second Skulltrail with the same BIOS, PSU, CPUs, FB-DIMMs and Dominator as what you've got. When I first answered this thread, I was dealing with overclocked Xeon e5420's. If it fails on Intel Burn Test (minutes), it's just a matter of time (hours and hours and hours.) before it fails with the other tools. The beauty of the tool is that instabilities are triggered much faster, as you've seen (very often within 3 iterations when using max memory on 64bit OS, but sometimes as late as 6 to 8 iterations). Adding active cooling to NB keeps temps to max 63c and no more BSOD. Įarlier OC attempts showed NB temps rising to 70c immediately followed by BSOD. I maxed all my voltages to their published limits (except the MCH which I read causes memory problems at 1.4v and above), still no joy.Įdit: Following link would suggest that since small FFT was without issue and Memtest86+ was without issue, then likely candidate is the CPU/RAM interface (i.e. Seems we're both using the same motherboard, BIOS, memory and hard drive. Xeon e5420 x2 (locked multiplier), FSB at all ranges from 400-425 and at stock 333.ĤGB Kingston Hyper X DDR2 800 FB-DIMM (4 x 1GB), Tried at 4-4-4-12 and 5-5-5-18Īctive cooling over RAM and northbridge, all temps under 60 for all components. Most overclock combinations resulted in a BSOD, one combination gave a rounding error, the rest are always "Prime95 has encountered an error. Yet, running the blend test always crashes after many hours (I've always been in bed). In my case, running the small FFT goes no problem at all for 24hrs. But, I'm getting the exact same behavior, albeit with different specs. Thank God someone else is seeing this too! I can't say whether or not I'm failing at the same place (because I haven't checked, though I will now).
EDIT - Tried manually testing 4096 FFT, it crashes in the same way, basically anything above 3072K results in a crash to desktop, or so it seems at present.
Tried 2 different PSUs (Silverstone ZM1200M and Ultra x3 1600w) Loosened timings from optimal 4-4-4-12 to default 5-5-5-18 Tested each pair of memory modules individually (with the other two removed from the board). Increased northbridge voltage from 1.2 to 1.4 (and all values in between), northbridge also has active cooling and never exceeds 64 degrees during the test - NB on this board is known to be hotter than usual.ĭisabling palette snoop/enabling palette snoop RAM has active cooling (temperature never exceeds 65 degrees on any of the 4 modules) Upping Vcore on both CPUs from 1.21 (default) as far as 1.35 (including all values in between).Ĭhecked temps (Core temp 99.3 reports that the hottest core never exceeds 51 degrees during this particular test) In fact even if I begin the custom test at 3072K, it still seems to crash before successfully completing even a single round of tests on any individual core. To try and isolate the problem, I've run a custom test (now about 20 times) in Prime 95 beginning at 2560K FFT - which is the test immediately preceding 3072K - and on every single occasion, the program has crashed to desktop seemingly at exactly the same point.Īs soon as all 8 cores have switched from processing 2560k to 3072k, it never completes even the first batch of iterations of 3072K, it just crashes. Note that there is no rounding error, the progam simply crashes to desktop with the usual microsoft error report box. Over the last 2 days, I've run two Prime95 (v25.6) blend tests on my system and on both occasions, the application has crashed upon reaching the 3072K FFT test (after approximately 9 hours).
0 kommentar(er)
