What are potential bottlenecks to AI
development right now?
There is an acute shortage of GPUs, almost all of which are made
by Nvidia, for supercomputers used to train LLMs. Lead times are at least nine months and
more usually 18 months for the new H100 chip. The OpenAI supercomputer used to train GPT-4 reportedly uses 10,000 of the older Nvidia A100 GPUs. H100s have a retail price of
around $10k but can trade on the grey market for as much as $40k
Beitrag empfehlen 
Warren Buffett(3290)
ChatGPT - What are potential bottlenecks to AI development right now? 