Pages: [1]
Sid2
 
Forum moderator - BOINCstats SOFA member
BAM!ID: 28578
Joined: 2007-06-13
Posts: 7336
Credits: 593,058,630
World-rank: 3,055

2008-07-02 14:01:48


If there isn't, there are some quads o/ced to 8Gigs.


Dr Who Fan
BAM!ID: 1075
Joined: 2006-05-31
Posts: 942
Credits: 145,179,783
World-rank: 8,279

2008-07-05 19:20:44


If there isn't, there are some quads o/ced to 8Gigs.


Sid, I have never seen any and as far as I know there are no optimized apps/clients available for SIMAP.

Sid2
 
Forum moderator - BOINCstats SOFA member
BAM!ID: 28578
Joined: 2007-06-13
Posts: 7336
Credits: 593,058,630
World-rank: 3,055

2008-07-05 20:13:26


If there isn't, there are some quads o/ced to 8Gigs.

Sid, I have never seen any



DWF:

My stock Q6600 produces these numbers:

Measured floating point speed 2338.27 million ops/sec
Measured integer speed 5395.12 million ops/sec


While this computer [by no means the highest numbers I have found, just an anonymous one] returns these results:

Measured floating point speed 3239.2 million ops/sec
Measured integer speed 10069.45 million ops/sec


I have heard of Q6600 being o/ced to 3.4Gig, at least from the MIS this is much higher.








Rakarin
 
BAM!ID: 1019
Joined: 2006-05-30
Posts: 92
Credits: 0
World-rank: 0

2008-07-06 15:35:10


Measured floating point speed 3239.2 million ops/sec
Measured integer speed 10069.45 million ops/sec


I have heard of Q6600 being o/ced to 3.4Gig, at least from the MIS this is much higher.


There was an optimized client long ago, by akosf (who did some wonderful optimizations for SZTAKI and Einstein, that I know of), but that one didn't work well.

http://boinc.bio.wzw.tum.de/boincsimap/forum/viewtopic.php?t=96

It's possible someone "revived" the code.

However, as I understand it, the cobblestones is generated and reported by the BOINC client itself, not the project app. My guess is that either the user has a dual processor system (even with dual, quad, and now trial core systems, multi processors have not gone away), or is using one of the "optimized" BOINC clients.

<Side_Rant> Personally, I have no idea why someone thinks they need a BOINC client that will use SSE instruction sets. It loads and unloads programs. You'll never be loading or unloading so many programs at once in parallel that SSE will give you improved performance. </Side_Rant>
Rakarin
 
BAM!ID: 1019
Joined: 2006-05-30
Posts: 92
Credits: 0
World-rank: 0

2008-07-06 15:38:44


Measured floating point speed 3239.2 million ops/sec
Measured integer speed 10069.45 million ops/sec


I have heard of Q6600 being o/ced to 3.4Gig, at least from the MIS this is much higher.


http://www.amdusers.com/forum/archive/index.php?t-1638.html

Actually, check that link. Scroll down a bit, and you will find people reporting much higher cobblestones with "optimized" BOINC clients.
Sid2
 
Forum moderator - BOINCstats SOFA member
BAM!ID: 28578
Joined: 2007-06-13
Posts: 7336
Credits: 593,058,630
World-rank: 3,055

2008-07-06 21:26:27


Thanks for the link, it gives me some background for understanding.

I got involved in BOINC a little over a year ago, and back then optimized aps was a real storm.

I am settling into a participation mode, I'll let the big dogs lead.



zombie67
BAM!ID: 1560
Joined: 2006-06-04
Posts: 606
Credits: 178,107,527,132
World-rank: 45

2008-07-07 13:51:36

There is a difference between "app" and "client".

The BOINC client does no actual work. The only "optimization" it can do, is to mess with the benchmarks, so that it claims more or less upon completion of a task. This has no impact on projects that use such things as server-side calculated credits, fixed credits or transaction counting.

The project app does the actual crunching. Each project has its own application(s). Some projects have optimized applications, that take advantage of special instruction sets in more modern CPUs (SSE*). These optimized applications actually do the work faster, which also generates higher credits than the standard application.

As for SIMAP, there are no optimized applications. Also SIMAP uses some kind of server-side calculated credits, so using an "optimized" BOINC client will have no effect on credits awarded.
Reno, NV
Team: SETI.USA
Sid2
 
Forum moderator - BOINCstats SOFA member
BAM!ID: 28578
Joined: 2007-06-13
Posts: 7336
Credits: 593,058,630
World-rank: 3,055

2008-07-07 15:38:56

Zombie:


Thanks for an elegant and understandable explanation.

Am I correct to assume that BOINC client will be updated to take advantage of improved instruction sets with new generations of processors?

. . . or is that assuming too much?




Saenger
 
Tester - Translator
BAM!ID: 5
Joined: 2006-01-10
Posts: 1735
Credits: 221,151,993
World-rank: 6,229

2008-07-07 15:59:30
last modified: 2008-07-07 16:03:11

Zombie:
Thanks for an elegant and understandable explanation.

Am I correct to assume that BOINC client will be updated to take advantage of improved instruction sets with new generations of processors?

. . . or is that assuming too much?

No need for this, why?
The BOINC client is doing nothing whatsoever, there is absolutely no need to "optimise" it. The current, so-called "optimised" BOINC clients are usually only used to cheat with extremely overblown benchmarks without any justification for them.

The applications from the projects are doing actual crunching work, they need to get optimised.

Edit:
"Optimising" the BOINC client for better performance is like optimising the keyboard drivers for better writing speed. The hands do the speed (or the apps in BOINC), the driver will not make any difference.
Grüße vom Sänger
zombie67
BAM!ID: 1560
Joined: 2006-06-04
Posts: 606
Credits: 178,107,527,132
World-rank: 45

2008-07-07 20:02:57

The BOINC client does no actual work. The only "optimization" it can do, is to mess with the benchmarks, so that it claims more or less upon completion of a task


Let me correct myself. You can also "optimize" the BOINC client to do other things such as "Return Results Immediately" and "CPU Affinity". Things like RRI and CPU affinity are turned off in the standard version available on the download page. FWIW, with 5.10.45, you can turn on RRI via cc_config.xml, and I have not seen any real performance improvement with CPU affinity. The OS seems to already do a pretty good job at that.
Reno, NV
Team: SETI.USA
Sid2
 
Forum moderator - BOINCstats SOFA member
BAM!ID: 28578
Joined: 2007-06-13
Posts: 7336
Credits: 593,058,630
World-rank: 3,055

2008-07-07 21:02:58
last modified: 2008-07-07 21:07:28



Let me correct myself. You can also "optimize" the BOINC client to do other things such as "Return Results Immediately" and "CPU Affinity". Things like RRI and CPU affinity are turned off in the standard version available on the download page.



Is this an option the user can exercise?

Would WU's upload without waiting around for the Update action?

FWIW, with 5.10.45, you can turn on RRI via cc_config.xml, and I have not seen any real performance improvement with CPU affinity. The OS seems to already do a pretty good job at that.



What exactly is CPU affinity?

I have seen the same project run better on a P4 instead of a C2D, is that anything that the user can adjust?

I'm obviously sitting at the feet of a Master.

. . . and it isn't a good day until I learn something new.


zombie67
BAM!ID: 1560
Joined: 2006-06-04
Posts: 606
Credits: 178,107,527,132
World-rank: 45

2008-07-08 00:18:40
last modified: 2008-07-08 00:19:32



Let me correct myself. You can also "optimize" the BOINC client to do other things such as "Return Results Immediately" and "CPU Affinity". Things like RRI and CPU affinity are turned off in the standard version available on the download page.



Is this an option the user can exercise?

Would WU's upload without waiting around for the Update action?


Not directly from the client or manager. The BOINC client needs to be compiled with RRI turned on. An alternative method is to add it to your cc_config.xml file. You can read more about it here:

http://boinc.berkeley.edu/trac/wiki/ClientMessages




What exactly is CPU affinity?


Let's say you have a dual core processor, and running two BOINC tasks. CPU affinity tells it to keep task A on core #1, and task B on core #2, and keeps them from switching back and forth. Theoretically, this helps when the L2 cache for each core is separate. Turns out that the OS is pretty good at managing this without having BOINC tell it to do so explicitly.

Reno, NV
Team: SETI.USA
Pages: [1]

Index :: Optimized clients :: Is there an optimized client for SIMAP?
Reason: