Windows S5R3 App 4.25 available for Beta Test
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4330
Credit: 251365395
RAC: 36440
19 Jan 2008 3:20:23 UTC
Topic 193449
(moderation:
A new Windows App is available from our Beta Test Page.
This is our first App that has the graphics in a separate program rather than in a thread, with the advantages already published:
* screensaver should work on Windows Vista (with recent Core Clients)
* graphics problems shouldn't affect the computation
* no additional CPU load when not showing graphics
Also too this app features the fast "linear sin/cos" code. At first glance the App didn't look much faster than the 4.15 to me in total, though.
Happy testing!
BM
BM
Windows S5R3 App 4.25 available for Beta Test
)
does it require a V6 boinc client or will our older ones be fine??
RE: does it require a V6
)
It should run with all Core Clients. I'm not completely sure which version of the Client / screensaver is required to see the graphics. It seems to work fine with 5.8.16.
BM
BM
RE: Also too this app
)
I've loaded the new ap on a Windows XP Q6600. BOINC version 5.10.30
After the first eleven minutes of run time, it appears that the ap is using less power than was 4.15.
Possibly the comparison is flawed, as before switchover I had four in-process WU's, which I've suspended to allow fastest completion of four "pure" new ones.
My power inference is indirect, the observation is that the core temperatures as reported by Speedfan appear to be down by 3 or 4 degrees C.
I'm mentioning it in hope that another tester finds this possibility interesting, and monitors their changeover more carefully than I did, preferably measuring power rather than core temp, preferably using the same work units before/after changeover, and preferably waiting longer than ten minutes for more thorough stabilization.
Even if true, I'd expect the effect to vary with architecture.
Is there anymore S5R2 units
)
Is there anymore S5R2 units left, this because I´m using a app-info that lets me get both S5R2 with 4.33 and S5R3 with latest beta. Should I keep this or can I go to one that´s only for S5R3.
Running fine for me on a P4
)
Running fine for me on a P4 and a Core2Duo. The P4 had a result that was about half finished (using the old beta) when the switch over occurred. The Core2Duo has fresh results.
Kathryn :o)
Einstein@Home Moderator
First task is currently
)
First task is currently faster than the previous tasks with 4.15, however I think I was at the upper end of the runtime variation for the previous tasks. I've stopped processing work for other projects and am only working on Einstein for a few days so as to try to properly test the app...
I installed this on some of
)
I installed this on some of my machines. Does EAH_NO_GRAPHICS in the BOINC directory have any effect on this app?
RE: Does EAH_NO_GRAPHICS in
)
No, none at all.
BM
BM
I've had multiple successful
)
I've had multiple successful completions on my Q6600 Windows XP system, both of "pure" new Work Units, and of Work Units started on 4.15 beta and finished on 4.25 beta.
Most have validated, and those not yet validated are waiting for first returns from quorum partners.
The bad news is that I'm pretty confident that the 4.25 ap requires appreciably more CPU time than the 4.15 ap on my Q6600. By good fortune, I have a recent log of several cycles of 687.nn results from the same system. The four results available for pure 4.25 are all 688.0n, and the one mixed result is 487.90.
While the nn's are not all the same, sadly, the rather orderly behavior of the previous data paints a picture of orderly cyclicity. I think we can reasonably infer that this graph suggests the 4.25 ap requires between 5% and 10% more CPU time on my host than the 4.15 ap.
Confession: my choice of scale excludes about 4 high-time outliers in the 4.15 population of a bit over 230. But I think it unlikely that same anomaly has recurred for all five of the 4.25 points to date.
The unit on my Pent M looks
)
The unit on my Pent M looks like its completion time is going to be 16+ hours, max previously 13.5 hrs. Therefore I agree with archae86 that it looks like 4.25 is slower.