3 ms·
A large part of accelerating AI loads in H/w lives in the software stack, not in the chip itself. If people can run TensorFlow loads directly on the chip via q
by ur-whale 8y ago
A large part of accelerating AI loads in H/w lives in the software stack, not in the chip itself.
If people can run TensorFlow loads directly on the chip via qroq's s/w stack, who cares about Cuda.
That will be the differentiator for this company, not their h/w manufacturing prowess.
- sanxiyn 8y agoApparently they have a person who was involved in TensorFlow TPU port.