Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
cubefox
5 months ago
|
parent
|
context
|
favorite
| on:
Apple M3 Ultra
> Transformers are typically memory-bandwidth bound during decoding.
Not in case of language models, which are typically bound by memory size rather than bandwidth.
whimsicalism
5 months ago
[–]
nope
cubefox
5 months ago
|
parent
[–]
I assume even this one won't run on an RTX 5090 due to constrained memory size:
https://news.ycombinator.com/item?id=43270843
whimsicalism
5 months ago
|
root
|
parent
[–]
sure on consumer GPUs but that is not what is constraining the model inference in most actual industry setups. technically even then, you are CPU-GPU memory bandwidth bound more than just GPU memory, although that is maybe splitting hairs
cubefox
5 months ago
|
root
|
parent
[–]
Why are industry setups considered actual while others are not?
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
Not in case of language models, which are typically bound by memory size rather than bandwidth.