Building a 50 Teraflops AMD Vega Deep Learning Box for Under $3K
Carlos E. Perez

Carlos E. Perez

So, you just run the benchmark code which Soumith created. If so, the results are based on only one GPU not two GPUs.

./caffe/build/tools/caffe time — model=./imagenet_winners/alexnet.prototxt — iterations=10 — gpu 0 — logtostderr=1 >>output_alexnet.log 2>&1

in the command it says ‘gpu 0’ which means it is using only GPU #0

So, can you do tests on multi-gpu to see the performance improvements in that case?

Just change ‘-gpu 0' to ‘-gpu 0,1’. Then it will use both GPUs