I have mentioned this elsewhere in reply to a number of threads. I have an Athlon 64 Winchester core 3000. This chip overclocks like crazy using stock cooling and OCZ EL ram on an MSI Neo 2 Platinum motherboard. I have run it as high as 2.93 mhz. At this speed, my WU downloads are listed in the 8 hour completion range. As I downclock the processor, WU times (est completion time in Einstein 4.79 under Boinc 4.19) actually decrease. 6:35 at 2.45 ghz, 5:35 at 2.21. This is only after retuning the processor. Repeat, there is no web access during this process. Given this set of circumstances if I down clock to 400mhz or so I should be doing WU's in no time at all! Am I missing some hidden "feature" of Einstein/Boinc? Intuition says the faster the processer, the shorter the completion times. Why am I seeing exactly the opposite? Guess I will be hitting garage sales looking for the trusty Celeron slowbee and some pc100 memory. Comments welcome.
Copyright © 2024 Einstein@Home. All rights reserved.
My Turn, CPU Throttling for High Speed CPU's
)
> I have mentioned this elsewhere in reply to a number of threads. I have an
> Athlon 64 Winchester core 3000. This chip overclocks like crazy using stock
> cooling and OCZ EL ram on an MSI Neo 2 Platinum motherboard. I have run it
> as high as 2.93 mhz. At this speed, my WU downloads are listed in the 8 hour
> completion range. As I downclock the processor, WU times (est completion time
> in Einstein 4.79 under Boinc 4.19) actually decrease. 6:35 at 2.45 ghz, 5:35
> at 2.21. This is only after retuning the processor. Repeat, there is no web
> access during this process. Given this set of circumstances if I down clock to
> 400mhz or so I should be doing WU's in no time at all! Am I missing some
> hidden "feature" of Einstein/Boinc? Intuition says the faster the processer,
> the shorter the completion times. Why am I seeing exactly the opposite?
> Guess I will be hitting garage sales looking for the trusty Celeron slowbee
> and some pc100 memory. Comments welcome.
I don't know nothin' here, but it would be extremely odd if reducing the clock speed also reduced the time required for one unit to finish.
But I'd like to be sure what stats you are reporting. After a unit is completed, we get "CPU time (sec)" stats & those stats I have seen look very reliable so far.
The latest BOINC I have tried also gives some times while a unit is still being processed. As I recall, it reports the time the unit has run so far, & an estimate of how much ADDITIONAL time will be required to finish the unit. To get an estimate of the total time for the unit, these numbers must be added, as far as I can make out. Are we talking about the same numbers here?
ADDMP
I have an Opteron 146 running
)
I have an Opteron 146 running at 2.0 gHz. The completion time for an Einstein WU is 6:15. Your Athlon 64 running at 2.21 gHz should be in the 5:45 range, so 5:35@2.21 looks about right. At 2.93 gHz you should be easily breaking the 5 hour mark.
What temperatures are you getting when running full load at 2.93 gHz? I would suspect that the CPU is automatically throttling back when hot.
Also what are your memory
)
Also what are your memory latencies? All I know is Intel systems but, from what I have read Athlon 64 are much more sensitive latencies.