LinkBack URL
About LinkBacks
Okeydoke, I'll give it a go when machine has some free time. Just checked my notes from the mem test (apologies to anyone who thought this was about graphics cards!):
CPU measured at stock as clocking at 1.937GHz, RAM set to 2-3-2-5 1M SuperPi @ 44.1s
(C&Q turned off, HT set to x3)
I then set the more obscure memory parameters and CPU voltage etc as in the xtremesystems post above (still 2-3-2-5 though).
CPU came up as 2.009GHz. 43.9s SuperPI (running on single core - this isn't a soak stress test - ran several runs to get a stable average).
Bumping clock to 2.050 gave 43.0s.
I then dropped the CPU voltage 1.42v rather than whatever was recommended above, with 2.78v on the memory. I must have done something else, because it then SuperPI'ed at 41.7s
Upped the speed to 2.100 GHz, dropped the CPU to 1.39v, memory to 2.68v to stay within warranty spec, SuperPI gave 40.8s, a 1-off 3D Mark 05 gave 7180.
I then tried setting the HyperTransport multiplier to x4 instead of x3, which gave identical timings, just to prove what I had read in other people's tests.
@ 2.200GHz. No other changes. 7302 3dmarks. 38.9s SuperPi.
- Started getting unpredictable behaviour, so upped the Cas Latency to 2.5.
- 7332 3dmarks, 38.7s. 3d'05 *CPU* score of 6793. Seems to be running faster than CL2.
@ 2.250GHz. Wouldn't boot at CL2.5 - Disappointed to say the least!
- Bumped it to 3.0. Maybe a voltage bump would have helped?
@ 2.250GHz, 38.0s to SuperPi. So measurably faster despite the CL bump. Skipped the 3dmarks.
@ 2.300GHz 7343 3dm. CPU score 6510. *lower* than 2.2GHz. SuperPi still improving though. 37.1s
@ 2.350GHz 3dm CPU score 6549 (still lower than @ 2.200) SuperPI 37.1
@ 2.400GHz SuperPI 35.6s.
At this time it was approximately 5am and I was happy with a 20% speed bump. Max CPU temp was only 0.7C above the temp at 2.1GHz when I dropped the voltages. I should probably investigate how come 3dMark 05 dropped the cpu score, but this is now my main machine and if it's turned on it tends to be working (or playing).
Despite the memory latency increase from 2 to 3 cycles, SuperPI has scaled almost exactly, so presumably even at 1M digits the 512kB cache doesn't get many misses and the process is fairly simply reflecting the CPU speed. Then again I was using it more as a crude short term reliability test than anything else.
Subsequently running Folding@Home, with UD or a long SuperPI (and HalfLife 2 on several occasions - I usually stop F@H when I play a game, for heat reasons rather than any game slowdown) hasn't given any nasty reboots or bluescreens. F@H on 1 core, with not much on the other is giving a CPU temp of 35C, but it's getting a tad cold in here!
Amazing chips though, not far off the pace of the expensive stuff but for only a few quid more than the real cheapo stuff, I'm surprised Anandtech didn't really take the price into account. For anyone following these things, CPU is retail box X2 3800, CDBHE 0543 RPMW.
Mike
Reply With Quote
Originally Posted by Workaholic
