未验证 提交 1d434a9d 编写于 作者: T Tao Luo 提交者: GitHub

Merge pull request #15291 from wojtuss/wojtuss/fix-performance-drop

Fix performance drop when with MKL-DNN
......@@ -35,20 +35,8 @@ limitations under the License. */
DEFINE_double(fraction_of_cpu_memory_to_use, 1,
"Default use 100% of CPU memory for PaddlePaddle,"
"reserve the rest for page tables, etc");
#if !defined(_WIN32)
DEFINE_uint64(initial_cpu_memory_in_mb,
#ifdef PADDLE_WITH_MKLDNN
/* Aligned with mozga-intel, MKLDNN need at least 5000 MB
* to obtain the best performance*/
5000ul,
#else
500ul,
#endif
"Initial CPU memory for PaddlePaddle, in MD unit.");
#else
DEFINE_uint64(initial_cpu_memory_in_mb, 500ul,
"Initial CPU memory for PaddlePaddle, in MD unit.");
#endif // !defined(_WIN32)
DEFINE_double(
fraction_of_cuda_pinned_memory_to_use, 0.5,
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册