PyTorch is only part of it. There is still a huge amount of CUDA that isn’t just... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		qeternity 46 days ago \| parent \| context \| favorite \| on: TPUs vs. GPUs and why Google is positioned to win ... PyTorch is only part of it. There is still a huge amount of CUDA that isn’t just wrapped by PyTorch and isn’t easily portable.

svara 46 days ago [–]

... but not in deep learning or am I missing something important here?

qeternity 46 days ago | [–]

Yes, absolutely in deep learning. Custom fused CUDA kernels everywhere.

Scene_Cast2 46 days ago | | [–]

Yep. MoE, FlashAttention, or sparse retrieval architectures for example.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact