2 ms·
These results were achieved on dual-socket Xeon E5 2699v4 (the architecture is 5 years old and has no AVX512, they optimized for AVX2) and on Xeon Phi 7250 (tha
by Const-me 6y ago
These results were achieved on dual-socket Xeon E5 2699v4 (the architecture is 5 years old and has no AVX512, they optimized for AVX2) and on Xeon Phi 7250 (that thing does have AVX512 but that’s not a processor, a specialized accelerator with 68 cores).
Also Tensorflow is awesome fit for GPGPUs and is usually way faster on them.