6 ms·
See I absolutely dislike this thought that hyperscalars can easily beat nvidia. It is not their domain of expertise. Tpu are not where near GPU in performance.
by ilove196884 2y ago
See I absolutely dislike this thought that hyperscalars can easily beat nvidia. It is not their domain of expertise. Tpu are not where near GPU in performance. People really underestimate Nvidia's expertise and strengths.
- Etheryte 2y agoThey don't need to beat them on performance though. If you get half the performance at third the price you can just make more chips and be fine. It's not like Google is gonna run out of datacenter space.
- ilove196884 2y agoTPU support is very minimal. Also XLA just does not compare to CUDA yet.
- Etheryte 2y agoAll of this may very well be true — but it doesn't matter. Google is getting results very similar to the rest of the pack, probably at a fraction of the cost. The implementation doesn't matter so long as you get the results.
- mupuff1234 2y agoAnd yet Google seems to be doing just fine while spending less on GPUs than other big tech, so reality doesn't seem to align with your comment.
- ilove196884 2y agohttps://www.hpcwire.com/2024/10/30/role-reversal-google-teases-nvidias-blackwell-as-it-softens-tpu-rivalry/ https://www.hpcwire.com/2024/10/30/role-reversal-google-teas... Well they are still buying GPUs.
- mupuff1234 2y agoMostly for external cloud customers so doesn't change my point - TPUs have been more than enough for Google to lead the pack.