|I've had another go and have made some progress with fixing the production of -2%/0Mpts results. I also fixed the fact that is has been very slow to send (reparsing) results using previous 4.44 clients.|
|One manual send and O bugs: Looking good.|
|Ditto that. Running well on all 4 machines... good manual send.|
|I'm kind of waiting for Xanathorn to turn up here and comment on the relative speed of 4.44c. I'd thought I'd somehow messed up the threaded performance a bit because I noticed 4.44b in particular only loaded my quad core CPU to 75-80%.|
However, I just benchmarked 4.44c using the same test problem as I did for 4.43d and 4.44 and got some surprising results. 4.44c completed the result in 329.25 seconds as compared to 358.687 seconds for 4.44! Have I done something wrong in the testing, or is 4.44c really that fast?
|What's even stranger is that the CPU usage is still at 75-80%, but the program is still faster! So the improvement must have been something that isn't thread balancing (I did make one or two improvements earlier on in 4.44b that might now be showing through).|
|Had to wait for the program run for a while, since I had to work till late yesterday . 4.44c is faster but still not as fast as 4.43d/4.44a (currently 840 kpts/sec/core measured over 12 hours). I don't know what causes it though, cpu utilization seems fine. Maybe it's because of the reruns that take so long that messes up benchmarking a little bit, or it really takes up more computing time than 4.43d somehow.|
|It certainly will take up more time for a simulation than 4.43d because it does a more accurate calculation in one part. However I increased the Mpts scoring in 4.44x by about an eighth to reflect this (which made 4.44 and 4.43d equivalent in terms of Mpts rate on my computer).|
|Stephen.. what type of computer do you have?|
|Or more accurately, what type did you test it on?|
|Well I let it run for another 24 hours and speed is definitely lower on my rig than with v4.43d, 800 kpts/sec/core instead of 960 kpts (on an AMD Athlon X2 4400+). Mpts output is also lower by the same amount (20%), it's now about 40000 mps instead of ~47000 mpts. I haven't installed anything else than the new client since so it must be the client itself that's somehow slower. Anyone else have a benchmark comparison on an Athlon X2?|
|I tested it on a Kentsfield (quad-core Xeon X5355).|
|I think I'll let it run for 24 hours straight without flushing and see then on how many mpts the clients are at that point, might be more accurate I think.|
|stephen i think you deleted to muts on the home page, de full version of 4.44c are not avalible anny more|
|Patch to 4.44c from 4.42+ (RAR file, 165KB; Zip file, 232K. How to use...|
Use WinRAR to open RAR files, WinZip for ZIP files.
Version 4.44c Muon1 for Windows (RAR file, 2177KB; Zip file, 2781K.
|All those links work, and they're on the main page. Try refreshing.|
|sorry it all works, maybe it was time to go to bed for me|
|Are results from pre v4.44 clients still being accepted?|
I just wondered because our team is seeing active cruncher numbers plummet over the past week or so ,but it may just be a coincedence.
Btw Stephen ,as I mentioned in The benchmarking thread ,DPADs multi threaded performance isn't great anyway (75-80% sounds familiar) ,so I think the 75-80% loading you noticed is about the same as v4.43d. Hence I ran multiple clients .
|Odd... I got a few results out of 4.44c but then it crashed and now it crashes as soon as I start it up. Graphical version seems to work fine.|
|Assimilator, the only pre-4.44 optimisation allowed is PhaseRotEb1, so 4.43x clients should end up working exclusively on that. However, these new optimisations have shorter results, so it takes longer to get to 100KB! That might explain why fewer appear to be active - it's just they're sending results in less frequently. It might be worth me tweaking the client a bit some time so a send can be triggered by a number of results (e.g. 50) as well as hitting 100KB.|
Nexus, try a clean(er) install? On January 2nd I'll be back and can send out a debug version in case this keeps happening.
|I have a new installation, but it also crashes after some time. And the background process sometimes just hangs and does nothing. I have set CPU throotle to 20%...|
|I had that too on my AMD Ahtlon 1 Ghz and the background client, but turning off autosave in the config file fixed that so far. Only downside is when u turn off the pc the progress for the current run is lost.|
|Also when u do that don't forget to delete the autosave file in the client's folder to be sure.|
|Thanks for the info Stephen ,I'll watch this space.|
Happy new year btw
I'm curious ,why on earth would you want to throttle it to as low as 20%? (laptop?)
|--[And the background process sometimes just hangs and does nothing.]--|
I just had a thought - these hangs aren't immediately after putting the laptop onto mains power after the client has been sleeping on batteries (or on wake from hibernation) are they? The throttle relies on a timer, and I think I might not have realised that on a laptop that's hibernating, the timer might occasionally give very large time periods!
For the other bug e.g. Xanathorn I'll have a debug version around I guess.
|I didn't use the laptop on batteries for some time. But Hibernation may be. I check it...|
|I just hibernated, and the muon1.exe process is frozen... but i switched it on almost immediately after...|
It's 20% because laptop gets quite hot, when i don't do it. And it uses not more power than without it.
|Nexus if you send me an e-mail I can forward you the debug version of Muon1 that I just sent to Xanathorn.|
|I am suffering the odd client crash but ONLY on 5a runs and what i have noticed now as this has happened 4 or 5 times in the last 24 hrs is, it allways happens at the same timestamps on these 5a runs at 196.11ns i run the clients in cmdline so can find this out. You can restart the run but as soon as it hits 196.11 each time it fails.|
Let me know if you want one of the autocli.sav collected so you can see what this is possibly upto Stephen.
|i have the version from here http://stephenbrooks.org/muon1/v444c.zip but it keeps crashing on my xp|
then i delete all the files , extract all and it works ok for a few days
running XP 5.1 build 2600 SP2
|btw i saved the whole folder so let me know if u want any of the files|