3 ms·
Curious what will be the answer from Nvidia.
by riku_iki 8y ago
Curious what will be the answer from Nvidia.
- eanzenberg 8y agoTheir existing cards?
- ianhowson 8y ago> the largest Cloud TPU Pod configuration tested (256 chips) delivers a 200X speedup over an individual V100 GPU So a Cloud TPU chip is slower than a V100 chip? Seems like that's the answer.
- KenoFischer 8y agoWell, you lose some performance to distributed memory overhead of course, plus the TPU measure for that number is a previous generation TPU. The new ones are a decent bit faster. Plus you get four chips per "Cloud TPU". Putting all that together, TPUs are certainly a very competitive offering.
- wskinner 8y agoThey seem roughly comprable. If you can get 85% scale out efficiency, based on the numbers in this post, 256 V100s would be at parity with 256 TPUs. In practice, you can get 90% scale out on Resnet with last year’s GPU and networking hardware. Network bandwidth is the bottleneck, and with better network hardware that could be improved. Maybe that’s why Google didn’t publish a benchmark for 256 V100s. I’m also curious what the is the performance of the network interconnects in the TPU pod. I couldn’t find it documented, though I didn’t look super hard. [1] https://github.com/uber/horovod/blob/master/docs/benchmarks.md https://github.com/uber/horovod/blob/master/docs/benchmarks....
- david-gpu 8y agoThis is what NVidia says [0]. Notice how Google did not submit benchmark results for several tests. [0] https://blogs.nvidia.com/blog/2018/12/12/record-breaking-mlperf-ai-benchmarks/ https://blogs.nvidia.com/blog/2018/12/12/record-breaking-mlp...
- scottlegrand2 8y agoIndeed, all of this is but more examples of lies, damned lies, and benchmarks...
- riku_iki 8y agoNVidia won RESNET-50 benchmark with 80 nodes of DGX-1, curious if it is accessible in the cloud, and how much it will cost to rent it..