# Increase VRAM on Apple Silicon for Local LLMs
Canonical: https://social-archive.org/hanshelder/mCB5U2gAYN
Original URL: https://medium.com/@se.mehmet.baykar/increase-vram-on-apple-silicon-for-local-llms-1b35c453b165
Author: Mehmet Baykar
Platform: web
Share mode: full
## Content
*Override the default VRAM cap on Apple Silicon to allocate more unified memory to GPU tasks like running local LLMs.* Apple Silicon’s unified memory architecture shares the same high-speed RAM between CPU and GPU — no discrete GPU required. This is why you can run large language models locally on a MacBook without a separate graphics card. ## The Default VRAM Cap macOS doesn’t expose 100% of your RAM as VRAM. On a 64GB machine, the Metal GPU driver typically caps VRAM at ~75–80% of total RAM — roughly 51GB out of 64GB: ![Image](https://miro.medium.com/v2/resize:fit:640/format:webp/1*oVnlgQaKTxGR9af_lZenTA.png) That ~12GB gap is reserved for macOS and background processes. Reasonable for general use — but if your machine is dedicated to LLM inference, that’s 12GB of wasted context window or a smaller quantization than necessary. ## Override the VRAM Limit One `sysctl` command adjusts the GPU wired memory limit at runtime: ``` sudo sysctl iogpu.wired_limit_mb=61440 ``` The value is in megabytes. Calculate it as `desired_GB * 1024`: ``` # 60GB → 60 * 1024 = 61440 sudo sysctl iogpu.wired_limit_mb=61440 # Verify the change sysctl iogpu.wired_limit_mb > iogpu.wired_limit_mb: 61440 ``` No reboot required — takes effect immediately. Restart LM Studio (or whichever app reads VRAM) to pick up the new limit: ![Image](https://miro.medium.com/v2/resize:fit:640/format:webp/1*4ohgLILS2PfRuyNxOZkqTA.png) ## How Much to Allocate Leave **at least 4–8GB** for macOS. Allocating 100% will cause system lockups, beachballs, or a hard reset. ## ☕ Support My Work If you found this post helpful and want to support more content like this, you can [buy me a coffee](https://buymeacoffee.com/mehmetbaykar)! Your support helps me continue creating useful articles and tips for fellow developers. Thank you! 🙏 ## Persist Across Reboots The `sysctl` change resets on every reboot. Create a Launch Daemon to apply it automatically. Create the plist file: ``` sudo tee /Library/LaunchDaemons/com.local.iogpu-wired-limit.plist Label com.local.iogpu-wired-limit ProgramArguments /usr/sbin/sysctl iogpu.wired_limit_mb=61440 RunAtLoad EOF ``` Replace `61440` with your desired value from the table above. Load the daemon: ``` sudo launchctl load /Library/LaunchDaemons/com.local.iogpu-wired-limit.plist ``` Verify it’s registered: ``` sudo launchctl list | grep iogpu > - 0 com.local.iogpu-wired-limit ``` The VRAM override now applies on every boot — no manual intervention. ## Remove the Override To revert to default macOS behavior: ``` # Unload the daemon sudo launchctl unload /Library/LaunchDaemons/com.local.iogpu-wired-limit.plist # Delete the plist sudo rm /Library/LaunchDaemons/com.local.iogpu-wired-limit.plist # Reset to default (0 = system default) sudo sysctl iogpu.wired_limit_mb=0 ``` ## Important Notes - This works on **macOS Sonoma (14.x) and later** with Apple Silicon - The `iogpu.wired_limit_mb` parameter replaced the older `debug.iogpu.wired_limit` key - No SIP disabling required — this is a runtime kernel parameter, not a protected config - Setting the value to `0` restores default behavior - Monitor Activity Monitor’s Memory tab after increasing — if memory pressure stays red, reduce the allocation
## Media
1. image: https://social-archiver-api.social-archive.org/media/archives/hanshelder/3HHjrLmdIw/media/0.png