Developer Turns Jailbroken Sony PS5 into Local AI Server: GPU-Accelerated Inference of 27B-Parameter Qwen3.8-27B Model at 20.9 tokens/s
A developer has converted a jailbroken PS5 into a local AI inference server, leveraging GPU acceleration to run the Qwen3.8-27B model at 20.9 tokens/s, a significant improvement over earlier CPU-based approaches.
Key Takeaways
- Key Highlight：A developer has converted a jailbroken PS5 into a local AI inference server, leveraging GPU acceleration to run the Qwen3.8-27B model at 20.9 tokens/s, a significant improvement over earlier CPU-based approaches.
- Innovation & Tech：Highlights advancements in Qwen, Developer, Turns, demonstrating rapid progress in model capabilities.
- Industry Impact：Reported via IT之家 (CN), offering actionable signals for developers and technology leaders.
Developer abhi245y, building on Mert Cobanov's PS5LM project, developed PS5LM Plus, a native GPU application that enables a jailbroken PS5 to run the Qwen3.8-27B large language model locally, achieving an inference speed of 20.9 tokens/s.
The early PS5LM version relied on the CPU to run models, with the 27B-parameter model generating only about 1 token per minute on average, making it largely impractical. The new version achieves an order-of-magnitude performance improvement by porting the AMD RADV driver from the Mesa open-source project and leveraging the Vulkan API to tap into GPU computing power.
This experiment demonstrates the potential of consumer gaming consoles in non-traditional AI scenarios. The AMD RDNA-architecture GPU in the PS5 offers considerable floating-point computing capability, which can play a role in model inference tasks.
However, using a gaming console as an AI inference device still faces numerous limitations, including memory capacity, thermal design, and software ecosystem maturity. This approach is better suited for tech enthusiasts and experimental scenarios, and remains far from practical deployment.
Such explorations reflect the community's ongoing interest in localized AI inference. As large model parameter counts grow, computing demands continue to rise, and finding low-cost, accessible inference hardware has become one of the directions developers are paying attention to.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.
Industry Insights & Analysis
As artificial intelligence rapidly evolves, breakthroughs surrounding Qwen, Developer, Turns, Jailbroken are shifting toward scalable, robust real-world implementations.
Driven by both open-source ecosystems and proprietary model architectures, the integration between compute optimization, data engineering, and agentic workflows is accelerating. This development provides a strategic benchmark for upcoming AI tooling and developer workflows.