S5R4 V6.06 client versus 6.05 optimised
28 Nov 2008 20:57:40 UTC
Topic 194060
(moderation:
Swapped to the S5R4 6.06 client last night on 2 of my rigs.
Now, I've had time for there to be many WUs crunched and reported.
I took the time to look at the new time estimates to finish the typical WU in my half day cache. They are considerably slower than the 6.05 client I was using, by a significant margin (difficult to estimate, but considerable).
I am going back to 6.05 when the current 2 WUs on each machine are crunched, so I don't loose the crunching. I suppose it could mean we are heading to the longer part of the cycle, but the rise is in the order of 20% - 25+%.
Anyone else finding this to be true, or not?
Shih-Tzu are clever, cuddly, playful and rule!! Jack Russell are feisty!
S5R4 V6.06 client versus 6.05 optimised
)
A quick and dirty observation.
http://einsteinathome.org/host/1702377/tasks&offset=60
On this page one can see some times in the 30ishK area. Generally, these are the initial 6.05/6.06 tasks. Then the time has settled down to 26-28K times, which is equal or perhaps slightly more than the times I was getting with the straight 6.05. I have no idea on where they may fall in the "trough" though. This is on a Q6600.
RE: A quick and dirty
)
That page has work from two different regions.
For frequencies near 795, which that host was returning on November 25, the estimated cycle length is 187 sequence numbers, so the results returned near 467 are in the center of the valley. For frequencies near 551, which the host was returning on November 28, the estimate cycle length is 92, so some of the results returned are quite far up a peak, and some well into valley.
RE: RE: A quick and dirty
)
Thanks for the heads up. Unfortunately, I can not point to results from before Nov. 17/23, as they have cleared from the results shown database. So what I say here, everyone can take with a grain of salt. But my average times since starting the 6.05 in early Oct. have been everywhere in the 25K-27/28K, assuredly running through many peaks and valleys. I'm going to let the 6.06 run for a bit (dependent on successive successful validations of course) and see about a more "full" range of results for a time range.
Hi ! Because the runtime
)
Hi !
Because the runtime of the individual workunits vary so much, it is impossible to judge app performance just by observing a few results.
As 6.06 is more concerned with bug-fixing than with performance increase, no significant differences in speed are expected. An easier way to check performance is by benchmarking the apps with a "standardized" reference workunit so that timing results are more comparable. This was done here (the whole thread is discussing ways to measure app performance with confidence).
Happy crunching
Bikeman