asdf_

5 ms·

1. 1.▲
   by asdf_ 9y ago

   Sure, no argument there. For Google, producing and using TPUs, is probably cheaper than buying nvidia hardware.
2. 2.▲
   by asdf_ 9y ago

   The interconnect hardware is present on the dgx-1 board. The point of my previous comment is that it doesn't make sense to compare an entire TPU module (containing 4x TPUs) to a V100 board. The closest comparison to a TPU module (with
3. 3.▲
   by asdf_ 9y ago

   A single V100 board is vastly smaller than a TPU module with 4 TPU chips. The closest comparison in terms of size would be 1 Volta DGX-1 (8x V100s) compared to 2 TPU modules (8x TPU2 chips). Volta DGX-1: https://www.nvidia.com&#x
4. 4.▲
   by asdf_ 9y ago

   A single V100 board is vastly smaller than a TPU module with 4 TPU chips. The closest comparison in terms of size would be 1 Volta DGX-1 (8x V100s) compared to 2 TPU modules (8x TPU2 chips). Volta DGX-1: https://www.nvidia.com&#x
5. 5.▲
   by asdf_ 9y ago

   A single chip is only 45 TFLOPS. It is 180 TFLOPs per module (4x TPU chips): https://arstechnica.com/information-technology/2017/05/googl...
6. 6.▲
   by asdf_ 9y ago

   Google's 180 teraflop number is for a module, which contains 4 TPU chips. Meaning each TPU is 45 teraflops. See: https://arstechnica.com/information-technology/2017/05/googl...
7. 7.▲
   by asdf_ 9y ago

   It is actually 45 teraflops per TPU2 chip and 180 teraflops for a module (4x chips): https://arstechnica.com/information-technology/2017/05/googl... V100 is 120 teraflops of tensor ops per chip: https:/