A GitHub repo called PS5LM ports Llama.cpp to jailbroken PlayStation 5 consoles, allowing users to run any GGUF-format model directly on the hardware. The project first gained traction in a Hacker News thread that reached 165 points and 44 comments.
The port compiles Llama.cpp for the PS5's AMD-based system and exposes the same GGUF loader used on Linux and macOS. No additional quantization steps are required beyond standard GGUF files.
How the Port Works
The repo provides a custom build script that targets the PS5's FreeBSD-derived OS after jailbreak. Users load models through the existing Llama.cpp command-line interface with no PS5-specific API changes.
Inference runs on the console's CPU and integrated graphics. The setup reuses the same GGUF files that run on PCs, so model compatibility stays identical to desktop Llama.cpp.
Performance Numbers Reported
Early testers on the thread report running 7B and 13B GGUF models at usable speeds. Larger 70B models fit in memory but drop below interactive token rates.
No official VRAM or token-per-second figures appear in the repo. Community comments note that 7B Q4_K_M models produce roughly 8-12 tokens per second on the PS5 hardware.
How to Try It
The GitHub repository at https://github.com/cobanov/PS5LM contains the build instructions and required patches. Users must already have a jailbroken PS5 running the supported firmware.
After cloning the repo, the build script compiles Llama.cpp against the PS5 toolchain. Model files are copied to the console's storage and loaded with the standard ./llama-cli binary.
Pros and Cons
- Runs entirely on console hardware with no cloud dependency
- Supports every GGUF model already available
- Requires a jailbroken PS5, limiting the user base
- Performance remains lower than dedicated GPUs for larger models
Alternatives and Comparisons
| Feature | PS5LM + Llama.cpp | llama.cpp on PC | MLX on Apple Silicon |
|---|---|---|---|
| Hardware | Jailbroken PS5 | Any x86/ARM | M-series Mac |
| GGUF support | Full | Full | Partial |
| Token speed (7B) | 8-12 t/s | 30-60 t/s | 40-80 t/s |
| Setup difficulty | High | Low | Low |
Other local options include standard Llama.cpp builds on Linux or Windows PCs and Apple's MLX framework on Mac hardware.
Who Should Use This
Developers who already maintain a jailbroken PS5 and want to test GGUF models without additional hardware will find the port useful. Users seeking maximum speed or broad compatibility should run Llama.cpp on a conventional GPU instead.
Bottom line: PS5LM proves Llama.cpp can run natively on console hardware, but practical use stays limited to smaller models and users comfortable with jailbreaking.
The project shows that consumer game consoles can serve as unexpected inference platforms when the right toolchain is ported.
Top comments (0)