Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => Topic started by: KWSN - Sir Brian on 28 Mar 2007, 09:13:09 am
-
Hi Simon
as the intel quad core cpus are effectively dual dual cores I think I've noticed that cache coherency is affecting the performance of the optimised apps quite considerably (I'll post some examples later tonight as I'm at work at the moment and they block seti@home! >:( >:( )
basically I have a Core2 6300 (dual core), a dual proc 5120 (dual core) machine and a dual proc 5320 (quad core). I was expect similar performance per WU from each machine but the 5120 is about 15% slower than the 6300 and the 5320 is 15% slower than the 5120!
I've started using crunch3r's cpu_affinity capable boinc client but it appears to only of had a small performance improvement. any idea's? do you see a similar issue with your woodcrest 5160 machine when comparing it to Core2 E6700 machines?
also it'd be really nice to add a feature to the benchmarking tool to select to utilise all available cores when bechmarking rather than just one, as this issue doesn't surface with the benchmarking tool.
Cheers
Keith
-
OK here's an example of what I'm talking about...
Core2 6300
time taken 6004 seconds
http://setiathome.berkeley.edu/result.php?resultid=506464510
Xeon 5120
time taken 7696 seconds
http://setiathome.berkeley.edu/result.php?resultid=506464278
Xeon 5320
time taken 8740 seconds
http://setiathome.berkeley.edu/result.php?resultid=506464656
Ni!
-
better quad core base for cache use apps is a 4x4 by AMD : http://www.tomshardware.com/2006/11/30/brute_force_quad_cores/
-
From what I read here...
You have 3 systems with 3 different processors (maybe more, 3 u listed).
6300 = 1 CPU * 2 cores = 2 crunchers
5120 = 2 CPU * 2 cores = 4 crunchers
5320 = 2 CPU * 4 cores = 8 crunchers
You have not told us the memory speed for each system or how the motherboards are setup
(FSB, RAM timings, etc.)
But from a simple Guess I would think that the optmized cruncher is simply running into a memory bandwidth problem. Trying to crunch faster than RAM can be delivered to however many cores there are.
Each added cruncher wants RAM at a certain speed, but the systems are simply being overwhelmed.
-
Standard Quad core memory bottleneck ... yesterday i have test seti cruncher on 4x4 (my friend have it :o) :
standard SEE2 1 cruncher : 7032 sec
standard SSE2 4 crunchers : 7216 sec
8)
sorry i forgot links ... take times informative only