Issue running tf_cnn_benchmark on Xeon Phi

soumith / convnet-benchmarks

Easy benchmarking of all publicly accessible implementations of convnets

MIT License

2.68k stars 578 forks source link

Issue running tf_cnn_benchmark on Xeon Phi #131

Closed ilovechai closed 6 years ago

ilovechai commented 7 years ago

I have installed tensorflow as per Intel's instruction, built it from scratch with MKL library. The Xeon Phi (knights landing with 288 logical cores) does not have Graphics Card. Does it need to have a graphic card to run the benchmarks or is CPU only enough?

No. of images/sec that it gave: 4images/sec.

Does it really need GPU?