inference/research
verifiedhigh · 0.85updated 2026-10-11

llama.cpp issue #28107: llama-bench Metal OOM failure when model exceeds recommendedMaxWorkingSetSize on Apple Silicon

repo

Fields

id
sources/src-github-llamacpp-issue-28107
URL
https://github.com/ggml-org/llama.cpp/issues/28107
Publisher
ggml-org / community
Kind
repo
Published
2026-09-15
Accessed
2026-10-11
notes
Documents kIOGPUCommandBufferCallbackErrorOutOfMemory failure on Apple Silicon Metal when allocating 112 GiB against a 107.5 GiB recommendedMaxWorkingSetSize ceiling on a 128 GB unified memory machine.

Notes

Documents kIOGPUCommandBufferCallbackErrorOutOfMemory failure on Apple Silicon Metal when allocating 112 GiB against a 107.5 GiB recommendedMaxWorkingSetSize ceiling on a 128 GB unified memory machine.

Cross-references

This record references no other records.