2 ms·
Well, 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
by KenoFischer 8y ago
Well, 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.