NVIDIA DGX-1的對深度學習效能如何？
How effective for DGX-1 with 8 new Tesla P100 GPUs?
A few pilot runs. Our training with a computation-intensive neural network model for a new image learning problem (not fully optimized yet):
With multi-GPU (here 2) configuration, Tesla P100 on DGX-1 is 6 times faster than that with K80 (also 2 K40 units).
We can gain another 1.67x speed up with 4 GPUs (Tesla P100).
Love our DGX-1 (Thanks to NVIDIA, NVIDIA Taiwan, and Eric Chang)
