this post was submitted on 25 Apr 2026
394 points (97.8% liked)
Technology
84143 readers
2451 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
LLMs are limited by memory bandwidth much more then calculating power. You need HBM. Dedicated accelerators only lower power usage.
This is commonly cited, but not strictly true.
Prompt processing is completely compute limited. And at high batch sizes, where the weights are read once for many tokens generated in parallel, token generation is also quite compute limited. Obviously you want enough bandwidth to match the compute, but its very compute heavy.
You can see this for yourself. Try ~10 prompts in parallel on a CPU in llama.cpp, and it will slow to a crawl, while a GPU with a narrow bus won't slow down much.
Training is a bit more complicated, but that's not doable on CPUs anyway.
Now, local inference (aka a batch size of 1), past prompt processing, is heavily bandwidth limited. This is why hybrid inference works alright on CPUs. But this doesn't really apply to servers, which process many users in parallel with each "pass".