Google claims that its supercomputer is almost twice as fast as a comparable device based on the Nvidia A100 GPU in artificial intelligence training tasks.
The search giant published a scientific article describing its supercomputer based on specialized TPU v4 chips. In one system, Google bundled 4,000 of these processors.
As a result, this solution is 1.7 times faster and 1.9 times more energy efficient than a system based on the Nvidia A100. At the same time, Nvidia has already released a much more productive H100 accelerator, but the fact is that Google has started using the current generation of TPU since 2020, so the comparison with the A100 is more than correct. At the same time, Google hints that it is already working on a new generation of TPUs that can compete with the H100, but there are no details on this yet.