search

http403 / pyrit

Automatically exported from code.google.com/p/pyrit
0 stars 0 forks source link

Question about power balance CPU-GPU #75

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago 
This ins not a issue but a question.

by the way, If this isn not the right corner to place question, please tell
me the right one.

I saw issue 70 (where it is wrote 9400M in not powerfull enought to give
vantages versuns core 2 duo).

I am afraid of vicevesa.

Because of it is my intention to buy in near future a new NVIDIA GPU
codename FERMI (GT300 with 512 core), the question is: there is risk that
CPU will be not fast enough to fill GPU with data, so the GPU will not
works at its maximum rateo?
NVIDIA report Fermi SHOULD do 4x compared to a G200: based on this teorical
data, what should be the minimum CPU able to "fill" GPU at 99%-100%?
So, I wish to be sure the GPU power will be not wasted by a poor CPU :-)

Original issue reported on code.google.com by pyrit.lo...@gmail.com on 26 Nov 2009 at 3:26

GoogleCodeExporter commented 9 years ago 
General questions can be asked at http://groups.google.com/group/pyrit

There is no real danger of a CPU not being able to keep up with the GPU as long 
as
you don't have other (expensive) tasks running at the same time. This will also 
be
true for the GT300.

The major problem with the CPU is not computing-bandwidth but latency due to
process/thread-scheduling. If another cpu-intensive task (e.g. video-encoding)
demands timeslices from the OS, it will be allowed to run for a certain amount 
of
time and Pyrit won't be able to push work to the GPU. As the GPU usually 
computes
like ten times faster than the CPU, a latency of 1ms in scheduling equals 10ms 
of
lost computation-time. The scheduling algorithms in the OS are not designed to 
deal
with such asynchronous scheduling tasks.

In the end: A recent multi-core CPU is able to keep all current and most future 
GPU
busy as long as you don't have other GPU- or CPU-intensive tasks running at the 
same
time.

Original comment by lukas.l...@gmail.com on 26 Nov 2009 at 3:41

GoogleCodeExporter commented 9 years ago 
oh, nice to know.
By the way, I decided that when I run pyrit I don't run other tasks, to 
dedicate all
C/GPU power to calculation.

My dream is (when money avaiable) to have 4 GT300 in same PC (ask to google 
about
FASTRA).
Your answer confort my future money effort.

Original comment by pyrit.lo...@gmail.com on 26 Nov 2009 at 3:53