GoKawiilReddit developer ports llama.cpp to run natively on jailbroken PS5 GPU
A Reddit user known as AdIllustrious7246 built a fork of PS5LM, a llama.cpp port for jailbroken PS5 consoles, that uses the PS5_Vulkan graphics driver to run AI language models on the console's GPU instead of its CPU. The update lets the PS5 run a 27-billion-parameter model at 20 tokens per second, and smaller models like Gemma 4 E4B and Llama 3.2 3B at 61 and 124 tokens per second respectively, far faster than the earlier CPU-only version that was limited to about 6.25 GiB of memory.
GoKawiil's interpretation of the reporting above, not reported fact.
Shifting the workload to the PS5's AMD GPU removes the memory and speed bottlenecks that previously caused crashes, showing that console hardware not originally designed for AI can be repurposed for local, offline LLM inference. This progress builds on PS5_Vulkan, a broader effort to unlock GPU access on jailbroken PS5s, which could fuel further homebrew experimentation as GPU hardware for traditional AI setups remains expensive and scarce.
- A new fork of PS5LM runs llama.cpp natively on the PS5 GPU via the PS5_Vulkan driver.
- Performance jumped significantly: a 27B model hits 20 tokens/sec, and smaller models reach up to 124 tokens/sec.
- The GPU-based approach overcomes the 6.25 GiB memory limit and freezing issues of the earlier CPU-only payload version.
Sony PlayStation 5 Console — If you're intrigued by this jailbreak project turning the PS5 into a local AI inference machine, you'll need the hardware to experiment with yourself. The PS5's unified memory and GPU architecture are exactly what's being leveraged here to run large language models offline at surprising speeds.
See Sony PlayStation 5 Console on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: tomshardware.com — Jhet Borja, 2026-10-16
Published there as: “Jailbroken PS5s can now run local AI models at over 100+ tokens per second”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.