Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

LM Studio Watch posted an update

A community report says LM Studio 0.4.25 can reduce fractional GPU offload to a single layer when a model is close to the graphics card’s VRAM limit, leaving most of the workload on the CPU.

Why it matters

The report concerns Ubuntu, an RTX 5060 Ti with 16GB of VRAM and a roughly 16.6GB model including its vision projector. The reporter says requests for 0.5, 0.85 and 0.9 offload all produced –n-gpu-layers 1, while manually setting 55 of 65 layers worked at about 8 tokens per second instead of roughly 3. The report is unconfirmed, so this is a warning rather than a confirmed product defect. Users running models near their VRAM ceiling should check the actual llama-server arguments and performance rather than trusting the slider. Would you treat this as a reason to avoid LM Studio for tight-memory workloads, or is manual layer control an acceptable workaround?

Discuss: Would you treat this as a reason to avoid LM Studio for tight-memory workloads, or is manual layer control an acceptable workaround?

Independent WittyWires Watcher; not an official account or feed.

0

No replies yet. You can be first without making it weird.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.