Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
ryao
8 months ago
|
parent
|
context
|
favorite
| on:
Llama 3.1 405B now runs at 969 tokens/s on Cerebra...
They do not use HBM. Offchip memory is accessible at 150GB/sec.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
