How can I optimize this to complete projects faster?
11 Sep 2012 16:12:35 UTC
Topic 196517
(moderation:
I just started back with the Einstein@home. I would like to do as much as I can with my computer.
Are the benchmarks results where they should be for what I have? How can I change my settings to improve performance?
CPU type
GenuineIntel
Intel(R) Core(TM) i7-3820 CPU @ 3.60GHz [Family 6 Model 45 Stepping 7]
Number of processors
8
Coprocessors
NVIDIA GeForce GTX 670 (4095MB) driver: 30479
Operating System
Microsoft Windows 7
Professional x64 Edition, Service Pack 1, (06.01.7601.00)
BOINC client version
7.0.28
Memory
16360.37 MB
Cache
256 KB
Measured floating point speed
3525.62 million ops/sec
Measured integer speed
12415.46 million ops/sec
Average upload rate
24.37 KB/sec
Average download rate
984.58 KB/sec
Average turnaround time
0.15 days
Maximum daily WU quota per CPU
32/day
How can I optimize this to complete projects faster?
)
First of all you should run more than one workunit at a time on your powerful GPU. Set this up on this web page:http://einstein.phys.uwm.edu/prefs.php?subset=project. Set the "GPU utilization factor of BRP apps" to 0.5 for 2 simultaneous WUs, 0.33 for 3 and 0.25 for 4. You will have to experiment to find which works best for you.
Also make sure you have enough CPU power reserved to feed your GPU. I have 8 processors (actually 4 cores with hyper threading). I reserve 3 of the "processors" to feed the 3 GPU WUs I run simultaneously (and to make the machine somewhat usable). You set up this in the BOINC manager menu "Tools - Computing preferences", in the "On multiprocessor systems, use at most x % ..." field. Again, experiment to find the level that suits you.
RE: First of all you should
)
In most cases even on High End GPUs running more then 2 makes little
sense, not much gain. Besides they'll also run hotter, GPU load is higher.
And when running 2 you'll have 'free' 1 CPU core(or 2 threads;HT).
Einstein does need 0.5CPU + 1 or 0.5/0.33(etc.) GPU.
When 2 or 4 E@h WUs are running on my ATI(5870s), 1 CPU core is fully used
by EINSTEIN, regardless of the CPU-% setting in BOINC 'feeding' the GPU(s).
(SETI needs 1 cores (out of 4)atleast on ATI GPU, cause processing Time_Limit_Exceeded(a SERVER-side setting) stopped the computation, resulting in a failliar).
In other, shorter words, I'd run 2. (On GTX480 & 470 and 2 HD5780 GPUs)
On all FERMIs/KEPPLERs with >12 Compute Units/Cores. (x32 gives # of CUDA cores;High End FERMIs, IMHO)
Thanks guys
)
Thanks guys
RE: Would you run more than
)
No, not even on a GTX480/580/680 I would not run more then 2 WUs!
Running more gives more room for errors, GPU gets hotter, more CPU FLOPS
are needed to load and unload(writing output file{s}checkpointing,etc.).
I did notice increase in power consumption resulting in extra heat and
, also more constant and high GPU load! (~95%). PCIe bus traffic also increases.
(@ SETI@home people compained about the drop in performence using the latest
drivers, > 300.xx. 290.xx series with the 'sleep-bug'. I still use 275.33).