GNU/Linux S5R3 App 4.12 available for Beta test
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250553468
RAC: 34510
17 Oct 2007 14:17:18 UTC
Topic 193233
(moderation:
A new Linux App is available from our Beta Test page.
New over 4.09:
- Time-critical stuff has been compiled with gcc-4.0.3, which in our case (somewhat surprisingly) gave faster code than the previously used gcc-4.1.2.
- New signal handling: In 4.09 some times the generic BOINC signal handler caught some signals where we actually wanted our own to catch the signals. This has been fixed, so we hope to get more useful debugging information in case of an error.
- A change to the routine that writes the final output should fix the problem that occurred with a few workunits and the new checkpointing code.
Switching from results in progress of the 4.09 App should work (still showing the 4.09 number in the manager), tasks began with the 4.02 should (automatically) be finished with the original 4.02 App, new tasks should get assigned to 4.12.
Please test & report!
BM
BM
GNU/Linux S5R3 App 4.12 available for Beta test
)
I switched from 4.09 to 4.12 and it did not cancel the work done, so it is OK for me.
Tullio
Linux users should get this
)
Linux users should get this ASAP, so far it looks like the improvement is very respectable also over 4.09, compared to 4.02 its screaming fast. I finished 2 results with 4.12, both were started with and almost 50% done on 4.09, and still finished with a nice performance gain. The first "pure" 4.12 results are still in progress, a rough estimate clearly indicates that these will be the fastest i ever had on S5R3, and from what i can see im not on the fastest WUs in my que yet.
Team Philippines
The new app is worth
)
The new app is worth installing ASAP for the various improvements Bernd mentioned, but I'm not at all sure you will see a "drastic" performance increase (yet). The results I see lie within the natural "fluctuation" of runtime for S5R3. I just want to caution that you should not expect a huge perforance increase over 4.09. By "huge" I mean anything > 10%.
CU
Bikeman
The fastest 4.09 result i
)
The fastest 4.09 result i have is 23,919 sec for 220.74 credits. I have enough 4.09 results to say that this rig couldnt do it much faster than that. Check this graph, all the 4.09 results my Xeon3060 has finished so far:
4.09 results
That means, im still on the downward slope towards faster WUs and one result in progress is looking to break that 10% barrier with a good margin already now. VERY promising indeed, i wouldnt be surpirsed to see 15% faster WUs already tomorrow (if im really on the downhill slope). So 15% on core 2 duo is my bet, wanna bet against BM? :)
Will be interesting to see how my Pentium E2140 does with 4.12, it should return some results before morning, and i look forward to see some X2/Opteron results as well.
Team Philippines
RE: The fastest 4.09 result
)
Wow, indeed, the results look pretty fast!! It seems that gcc 4.0 happened to like the app much more than gcc 4.1 .
CU
H-BE
Help please! Is anyone
)
Help please! Is anyone running BOINC under Ubuntu and uses the client from the official repositories (not the one from the BOINC download page but the aptitude packet)? My file structure looks rather different and I don't want to break anything...
Annika, The Debian package
)
Annika, The Debian package installs boinc under /var/lib/boinc-client. So you want to put the beta app files in "/var/lib/boinc-client/projects/einstein.phys.uwm.edu". Since thats are not in your home folder you need to use "sudo cp" to copy the files there, or start nautilius with "sudo nautilius" to use the GUI. If its nautilius the explorer is called in Gnome, im KDE only.
And btw, you can stop boinc with "sudo /etc/init.d/boinc-client stop" before copying the files, then start it again with "sudo /etc/init.d/boinc-client start".
Team Philippines
Thanks, I hadn't noticed the
)
Thanks, I hadn't noticed the "projects" folder at first... sorry, I had a few rather busy nights doing stuff for Uni ;-). Really good you could help me out. Now the transition should be no problem. I'm looking forward to the performance increase, should be quite a bit since I'm still running 4.02 atm.
EDIT: Just switched, seems to be running fine. Thanks again. Really handy you included the star/stop command, didn't know that one before. Logical though, if you know it ;-)
Great, my x3060 broke the 20k
)
Great, my x3060 broke the 20k barrier, Chuck Yeager style, Bernd is pushing the envelope with 4.12. =)
19,835.65 - 236.19 - pending
19,957.82 - 236.19 - 236.19
(220.74 at old credit level)
Not 15% speedup, more like 17! =) Looks like im reaching the bottom now, will not be much faster than this. Linux is now the fastest OS for E@H i think, i dont see any Windows rigs in the Top20 that can match this credit/core per GHz (im at 3.6GHz), but then im comparing a single dualcore to dual quads, not really fair if the current app is limited by memory on 8-16 core systems.
Annika, you will have to finish those 2 results in progress on 4.02? Will be some time before we see what 4.12 does for you i think? Looking forward to see your Yonah results, just too bad theres no base for comparing it to 4.09.
Would still be nice to see some AMD64/X2/Opteron results with 4.12. Reason is, so far my Pentium E2140 @2.8GHz hasnt impressed me much (thats a Core2-based Pentium, same arch, smaller cache with lower associativity), 4.12 might love the big fast cache of the higher model Intels, the speedup for AMD and cheap Intels could be smaller.
Team Philippines
Hi! Just to make sure this
)
Hi!
Just to make sure this is "for real": Bernd has told me that the new beta is linked with a more recent version of the BOINC client lib, which features changes in the "CPU-seconds accounting".
Sooo...there is a possibility that the observed performance increase is due to a lower CPU seconds count only. You do not, by chance, have data on the "real time" (wall clock) time of the results? It would be most helpful to find out whether wall clock time decreased in proportion to the alleged CPU time decrease on hosts that do nothing or little else than crunching for E@H.
If you are only now installing the new beta, you might be able to lookup the real-time spent on the last result from the message tab if your host ran it uninterrupted.
Cheers
Bikeman