Use all six cpu`s?
28 Sep 2011 13:35:50 UTC
Topic 195984
(moderation:
Hello, this is my first post here. I`m taking a break from Seti due to their recent problems.
The question is this:
On my particular setup would it be more efficient to keep one cpu core idle to assist cuda apps?
Regards,
Andy
http://einsteinathome.org/host/4174144
Use all six cpu`s?
)
Rule of thumb is you always want at least one CPU core idle for each running GPU core.
When your computer is processing a GPU work unit, it still needs a CPU core to transfer the data back and forth.
Einstein and most GPU projects don't heavily task the CPU core but it's still wise to keep one ununsed CPU core by BOINC or computer lockups or freezes can occur.
Some projects (like DNET) actually can completely use up 2 CPU cores just to feed one fast GPU.
Don't believe what the line in the boinc manager tells you about how much CPU core it will use (like 0.05CPU + 1.00 GPU). These values are not going to be anywhere near correct on most computers.
On the other hand, I allow
)
On the other hand, I allow all CPUs to run other projects, while yielding the small amount of CPU time needed to support a GPU task under standard operating system multitasking control. I don't suffer lockups, or notice significant performance degradation.
The amount of CPU support needed by a GPU application depends on a huge number of variables:
Project coding skills and choices
Project GPU platform choice (native mode or OpenCL)
Operating system
Operating system version (Windows 7 typically uses more CPU than Windows XP)
GPU driver version (later drivers have a tendency to use more CPU than earlier ones)
Hardware balance between the power of your CPU and your GPU
Memory subsystems performance (both system RAM and GPU RAM)
Whatever else the computer is used for
and so on.
I wouldn't begin to give general advice, except 'give it a try both ways, and see which works best for you and your machine'.
I'm running my I7 machines on
)
I'm running my I7 machines on an 24/7 basis. To avoid long term high temperatures, but create a significant amount of output, I've adjusted my systems like
use 5 CPU with 80% usage
run 2 parallel GPU tasks
That creates ~18.000 - 23.000 credits per day and machine
I can work very comfortably with the remaining system capacities, even when playing games. Temperatures are ~ 65C for CPUs and 78C on GPU
RE: Don't believe what the
)
Have to say that mine runs 5% cpu load for each GPU WU running.
dunx
For running two CUDA tasks,
)
For running two CUDA tasks, one free CPU core should be sufficient. On your 2GB 460 cards, you have enough memory to run three or more CUDA tasks per GPU in which case you will probably want to have two free CPU cores. There may not be any performance benefit running more than 3 at once per GPU. I am currently seeing between 1 and 2 core usage (14-16%) while running 6 CUDA tasks.
The following day it now says
)
The following day it now says 0.20 cpu's per GPU !
That's the wonder of BOINC !
dunx
Is the SSE the fastest I can
)
Is the SSE the fastest I can use from Cruncher, I wonder. I have a six core AMD machine. I also run Boinc.
Thank you.
Actually it is the SSE2.
)
Actually it is the SSE2.
Your SETI@home app_info.xml
)
Your SETI@home app_info.xml works only for SETI@home, not for any other project. (That's why it's positioned in the projects\setiathome.berkeley.edu folder. ;-)
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: I also run Boinc. I
)
I should hope so, otherwise you wouldn't be able to attach to eithier the Seti or Einstein projects, or any other Boinc projects ;-)
Claggy