Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
dannyw
3 days ago
|
parent
|
context
|
favorite
| on:
GLM-5.2 – How to Run Locally
Correct. The main bottleneck with LLM inference is, and have always been, memory bandwidth.
TPS = active weights in GB / your memory bandwidth.
That’s it for decode. That’s all.
help
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
TPS = active weights in GB / your memory bandwidth.
That’s it for decode. That’s all.