| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
| World Community Grid Forums
|
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 4
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I just did a minor system update on my desktop. As I finished that, I realized I had enough hardware lying around to make myself a fileserver that should do fairly well on the grid, too. So I set everything up, get things working, and finally go to set up BOINC. When the first benchmarks came through, I was getting a little over half the speed I got off the same CPU on my desktop per core. (an X2 3800, now benching at 1.059GFlops per core) The early times for the WCG work units (crunching two HDC units now at 5-6 hours per) seem to bear out this estimation. The CPU seems to be running just fine, at 2000mhz, BOINC is using both cores as expected, top indicates the processes are getting 99% of the CPU or so...
So since the hardware is basically the same (RAM, CPU, etc) as it was when it was in my Windows box, only the system board being different, why am I losing so much speed in Linux? The only hardware-based conclusion I've been able to come up with is that the nvidia 6150 based board is hogging so much memory bandwidth for video that it's slowing the system down that much, but that doesn't seem likely to me. Anyone have any ideas? |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
One of my machines is configured to dual-boot Linux/Windows. I have BOINC on both OS's. Note that the hardware is therefore the same for both OS's. I have also noticed that BOINC's benchmarks under Linux are considerably lower but the WUs seem to complete in the same time regardless of the OS.
Other crunchers have noticed the same effect under Linux and that is one of the reasons a few have created "optimised" BOINC clients. They adjust the benchmarks upward so your credits will be more in line with what you would get under Windows on the same hardware. See TruXoft BOINC at http://boinc.truxoft.com. TruXoft calls his a "calibrating" BOINC client, whatever. It's based on BOINC 5.2.13, has other bells and whistles you might find useful and it works fine here. Give it a dozen or so WUs to calibrate itself, it takes a bit of time. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
This sounds like the Linux gcc compiler. It is infamous for poor optimization. But there are other compilers for Linux now. I suspect that Rick Alther has picked another compiler for our application programs.
Just guessing, Lawrence |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I may have indeed spoken too soon. I posted this after the first set had completed and I was projecting from the CPU time and percent complete from the next set, but now things seem to have normalized.
So it *would* seem to just be the benchmarks. I know app performance has nothing to do with client performance, but it was just a shock. I may try the calibrating client, as well. BoincView seems to have adjusted its expectations based on results, too. |
||
|
|
|