4 Sep
2008
4 Sep
'08
5:33 p.m.
On Thu, 2008-09-04 at 18:41 +0200, elthariel wrote:
Hello,
Thank you for updating this very interresting post.
Do you have any website where you update the status of your work,
publish some test codes ?
The test code used is the (unmodified) bandwitdhTest provided with the
cuda SDK and run with the following arguments:
$ ./bandwidthTest --memory=pinned --htod --mode=range --start=1024\
--end=10240 --increment=1024
Range Mode
Host to Device Bandwidth for Pinned memory
..........
Transfer Size (Bytes) Bandwidth(MB/s)
1024 44.4
2048 92.1
3072 50.7
4096 184.3
5120 207.8
6144 258.1
7168 299.8
8192 188.7
9216 374.0
10240 406.9
The call used in the test is blocking and busy spinning. It is possible
- actually quite easy - to do the equivalence in a non-blocking fashion,
say to let the Intel chip immediately do some touch-up and output the
result from the previous run, as well as start gathering and prepare
data for the next run.
Cheers
/j