Which version of LM Studio?
Example: LM Studio 0.3.15 B11
Which operating system?
What is your operating system? MacOS
What is the bug?
Even after setting the VRAM limit to 124 GB using sudo sysctl iogpu.wired_limit_mb=126976, LM Studio does not appear to utilize the full available GPU memory. It seems to cap usage around 64 GB. Models larger than ~70 GB are marked as "Likely too large" and, when loaded, cause high CPU usage (even exceeding 1000%) while the GPU remains mostly idle.
Smaller models (under 64 GB) do load entirely on the GPU, as confirmed by activity monitoring, so it seems the system is capable, but LM Studio is not making use of the full available VRAM.
Screenshots
System Info
Options
Slightly Larger Model Activity
Smaller Model Activity
To Reproduce
Steps to reproduce the behavior:
- Run sudo sysctl iogpu.wired_limit_mb=126976
- Open LM Studio
- Attempt to load a model larger than 64 GB
- Observe memory usage and system behavior (high CPU load, low GPU utilization)
Additional info
Smaller models below 64 GB do load completely into GPU memory with no issue.
I’m not entirely sure if I’m misunderstanding how VRAM limits work on Apple Silicon, or if there’s a configuration step I’m missing. I’d appreciate help understanding what’s going on or how I might adjust my setup. Thanks!
Which version of LM Studio?
Example: LM Studio 0.3.15 B11
Which operating system?
What is your operating system? MacOS
What is the bug?
Even after setting the VRAM limit to 124 GB using sudo sysctl iogpu.wired_limit_mb=126976, LM Studio does not appear to utilize the full available GPU memory. It seems to cap usage around 64 GB. Models larger than ~70 GB are marked as "Likely too large" and, when loaded, cause high CPU usage (even exceeding 1000%) while the GPU remains mostly idle.
Smaller models (under 64 GB) do load entirely on the GPU, as confirmed by activity monitoring, so it seems the system is capable, but LM Studio is not making use of the full available VRAM.
Screenshots
System Info
Options
Slightly Larger Model Activity
Smaller Model Activity
To Reproduce
Steps to reproduce the behavior:
Additional info
Smaller models below 64 GB do load completely into GPU memory with no issue.
I’m not entirely sure if I’m misunderstanding how VRAM limits work on Apple Silicon, or if there’s a configuration step I’m missing. I’d appreciate help understanding what’s going on or how I might adjust my setup. Thanks!