There are 2 CPUs and 8 nvidia GeForce GTX 470 cards in the server.
Position: /cudasdk/C/src/bandwidthtest
File: bandwidthTest.cu
The script I’m using is
[b]
START=5368709120
END=10737418240
INCR=1073741824[/b]
rm -f GPU.[0-9]
for i in 0 1 2 3 ; do
echo GPU : $i
numactl --cpunodebind=0 --membind=0 ../../bin/linux/release/bandwidthTest \
--memory=pinned --mode=range \
--start=$START --end=$END --increment=$INCR --device=$i --htod --noprompt \
2>&1 > GPU.$i &
done
for i in 4 5 6 7 ; do
echo GPU : $i
numactl --cpunodebind=1 --membind=1 ../../bin/linux/release/bandwidthTest \
--memory=pinned --mode=range \
--start=$START --end=$END --increment=$INCR --device=$i --htod --noprompt \
2>&1 > GPU.$i &
done
wait
This is getting me errors like
Device 2: GeForce GTX 470
Range Mode
Illegal argument - start must be greater than zero
Should we go from 32 bit to 64 bit integer arithmetic?
How to change the program?