SETI BOINC 4.09 speed increase

[zaph]

This weekend I upgraded all my BOINC crunchers to v4.09, and every single one of my machines has seen a nice speed increase.

All my machines are beating the estimated times by half an hour or so, where previously they were coming in more or less at the estimate.
Most noticable is the athlon64 2800 going from ~4 hrs/wu to just over 3 hours/wu average.
Least improvement, but still noticeable is a athlon classic 1400 goung from 7 to 6.5 hours per wu.

It seems eo me they did more than a bit of code optimizing with the latest release, but is anyone seing this on other platforms ( intel or PPC cpu's?)

 

[imported_Thunder]

I just got 4.09 installed on all 7 of mine today (late-afternoon), so gimme some time to accumulate some WU's and I'll let you know. 🙂

Now that I have BOINCView installed and monitoring both home and work machines, I don't have to worry about catching the WU's before they flush out.... they're all logged! 😀

Edit: I just realized though, on 6 out of the 7, I also switched from the GUI to running as a service, so that's going to affect the times as well, but I'll still report what I see.

 

[Rattledagger]

Since the actual crunching is done by the seti-application that still is v4.03, the cpu-time used for a wu is the same regardless of gui or cli or version-upgrade of the core client.
Of course, the run-time can be different, but if you're not constantly using the "pretty screensaver" or looks on the disk-pie-charts, there shouldn't be a marked difference here either between gui or cli.

That being said, v4.09 fixed many bugs, among others in the benchmark-code so the tests is actually run and not optimized away by the compiler. This gives lower benchmark-score and therefore longer expected crunch-times, but the actual crunch-time isn't changed so an initial look can seem the crunching is now done faster than before. But this isn't the case, so any shorter wu you're currently crunching is most likely due to angle-range.

 

[imported_Thunder]

Well, sadly I can't report anything but bad news. I don't have any perfect way to compare, but so far, all my P3 based machines are taking 15m, 30m or 1hr more (depending on the speed of the CPU) to do the WU's that I've seen.

Of course, the only thing I have to base this on is hand-written notes of approximate times before and now I'm logging the exact info. Also, I'm only basing this on 3-4 WU's on the new client and like Rattledagger said, it could just be a batch of different AR's than those of the WU's that I jotted down the times for before.