A very nice speedup--let's hope it turns out not to have any nasty habits.
What would be really great now is a comparison against Linux 4.27 on the same hardware...
I've paused some results in queue in order to hasten answers across enough phase angle variation to give both parameters required for a decent performance estimate. Those should be available when I wake up about twelve hours from now (sooner for those wishing to dig through my task pages themselves).
Should someone wish to compare to my Q6600 and E6600, I should mention both are mildly overclocked at 3.006 GHz, and both are running rather relaxed memory timings.
An interesting note... not that it matters in the long run: I've found that jobs started on the old 4.26 take LONGER to complete when switched over on the new 4.32 app. The new jobs that start seem considerably faster. One job took nearly an extra hour over the "average" of previous times, when it only had 20% left to do.
One new job in my queue seems to be heading for a ~23K second completion time on my stock speed Q6600 (memory at DDR2-800, 4-4-4-12). Previous jobs took closer to 30K seconds. Go speed racer.
An interesting note... not that it matters in the long run: I've found that jobs started on the old 4.26 take LONGER to complete when switched over on the new 4.32 app.
That is not what I've seen, nor what I expect. You might get a better idea of the expected value by taking into account the cyclic effects using Mike Hewson's Ready Reckoner than by comparison with averages.
I need to fix my settings on the Linux machine. It's currently running a screensaver that takes up half the CPU power... stupid screensaver. It's part of the reason why some of the recent WUs take much longer than the previous ones.
Check back in a few hours, and you'll see the first WU from the Windows machine done. It looks like it's heading towards a finish time of approximately 36K seconds, which matches the 4.27 Linux app clock for clock.
Yep another to say all went smooth when changing from 4.26 to 4.32, did not lose my current 5 WU's, one is 26% done on 4.26 and the rest will be processed with 4.32.
Update. Windows Host WU 91799821 running 4.26 with sequence number 181, very near trough of 187.5, took 35,742 seconds. WU 91938799 started on 4.26 and changed at 26% done to 4.32, with sequence number 109, not near peak or trough but closer to peak, I expected 38,000 to 39,000 seconds but it took 36,272 seconds. WU 91951323 a 4.26 WU but run completely on 4.32 at sequence number 96, again not near peak or trough so I expect around 37,000 to 38,000 seconds, it took 30,765 seconds.
That is the fastest so far on this computer, and faster by almost 14% over my previous lowest, which was near the trough, and I am only half way to the trough on this latest WU.
This Linux Host has the same specs as the above windows host, both AMD Opteron 285 machines with dual cpus, same Tyan motherboards, same amount of Legend ECC memory and neither over clocked, same Antec Titan cases and same Arctic cooler cpu coolers, only 2 differences are the hard drives, 2x 250 GB WD in the Linux machine and 2x 400 GB Samsung drives in the windows machine and slighly different model Nvidia graphic cards (7600 model/linux and 8600 model/Windows). WU 92024886 sequence number 186 (trough 185.4), which took 27,732 seconds.
So very much quicker and probably going to take it up to Linux or at least get very close if not go under.
So very much quicker and probably going to take it up to Linux or at least get very close if not go under.
My viewpoint is that there should be no more than 5% difference between apps running on the same hardware, but a different OS. A difference of 1-3% would be ideal...
Of course, I'm some half-cracked nut who doesn't know how difficult it is to do all this x87/SSE stuff... ;-)
RE: A very nice
)
What would be really great now is a comparison against Linux 4.27 on the same hardware...
RE: RE: A very nice
)
I've paused some results in queue in order to hasten answers across enough phase angle variation to give both parameters required for a decent performance estimate. Those should be available when I wake up about twelve hours from now (sooner for those wishing to dig through my task pages themselves).
Should someone wish to compare to my Q6600 and E6600, I should mention both are mildly overclocked at 3.006 GHz, and both are running rather relaxed memory timings.
An interesting note... not
)
An interesting note... not that it matters in the long run: I've found that jobs started on the old 4.26 take LONGER to complete when switched over on the new 4.32 app. The new jobs that start seem considerably faster. One job took nearly an extra hour over the "average" of previous times, when it only had 20% left to do.
One new job in my queue seems to be heading for a ~23K second completion time on my stock speed Q6600 (memory at DDR2-800, 4-4-4-12). Previous jobs took closer to 30K seconds. Go speed racer.
RE: An interesting note...
)
That is not what I've seen, nor what I expect. You might get a better idea of the expected value by taking into account the cyclic effects using Mike Hewson's Ready Reckoner than by comparison with averages.
I was coming to that
)
I was coming to that conclusion because the credit/time that I was getting on those 4 "halfway jobs" was lower after I made the app switch.
My apps in progress didn't
)
My apps in progress didn't crash either. Let's see how this goes!
RE: What would be really
)
If you're looking my direction....it'll have to wait for March.....I'm running MalariaControl for Dogbytes during the February "remembrance effort".
RE: What would be really
)
You might actually get that sooner than you'd expect. The following two hosts are running more or less the same hardware:
Windows Machine
Linux Machine
One is running Server 2003 SP2 with the new 4.32 app, and the other is running FedoraCore8 with 4.27.
CPU: Intel E2160
Memory: 4GB DDR2-800
Chipset: Intel 965
I need to fix my settings on the Linux machine. It's currently running a screensaver that takes up half the CPU power... stupid screensaver. It's part of the reason why some of the recent WUs take much longer than the previous ones.
Check back in a few hours, and you'll see the first WU from the Windows machine done. It looks like it's heading towards a finish time of approximately 36K seconds, which matches the 4.27 Linux app clock for clock.
RE: Yep another to say all
)
Update. Windows Host
WU 91799821 running 4.26 with sequence number 181, very near trough of 187.5, took 35,742 seconds.
WU 91938799 started on 4.26 and changed at 26% done to 4.32, with sequence number 109, not near peak or trough but closer to peak, I expected 38,000 to 39,000 seconds but it took 36,272 seconds.
WU 91951323 a 4.26 WU but run completely on 4.32 at sequence number 96, again not near peak or trough so I expect around 37,000 to 38,000 seconds, it took 30,765 seconds.
That is the fastest so far on this computer, and faster by almost 14% over my previous lowest, which was near the trough, and I am only half way to the trough on this latest WU.
This Linux Host has the same specs as the above windows host, both AMD Opteron 285 machines with dual cpus, same Tyan motherboards, same amount of Legend ECC memory and neither over clocked, same Antec Titan cases and same Arctic cooler cpu coolers, only 2 differences are the hard drives, 2x 250 GB WD in the Linux machine and 2x 400 GB Samsung drives in the windows machine and slighly different model Nvidia graphic cards (7600 model/linux and 8600 model/Windows).
WU 92024886 sequence number 186 (trough 185.4), which took 27,732 seconds.
So very much quicker and probably going to take it up to Linux or at least get very close if not go under.
RE: So very much quicker
)
My viewpoint is that there should be no more than 5% difference between apps running on the same hardware, but a different OS. A difference of 1-3% would be ideal...
Of course, I'm some half-cracked nut who doesn't know how difficult it is to do all this x87/SSE stuff... ;-)