The AMD BC-250, a PS5 APU repurposed from cryptocurrency mining boards, can now unlock all 40 graphics compute units (CUs)—up from the default 24 CUs shipped from the factory. Tom's Hardware benchmarked a fully configured build that costs approximately $350 ($200 for the board, ~$150 for case, fans, PSU), or as low as $400-500 total from scratch, versus Valve's newly released Steam Machine at over $1,050. Community-developed BIOS patches and GPU drivers enable the full 40-CU configuration, which can also unlock two additional Zen 2 CPU cores (bringing it from 6 to 8 cores) with comparable performance.
Unlocking all 40 CUs yields 17-28% performance gains in demanding titles: Cyberpunk 2077 achieves 42 FPS (vs. 36 FPS at 24 CUs), Hitman 3 hits 69 FPS (vs. 58 FPS), and smaller jumps across other AAA games. The board features 16GB shared GDDR6 memory, a Zen 2 CPU and RDNA 2 GPU—the same IP lineage as PS5 silicon, though significantly cut down from retail console specs. However, it is not a turnkey device: it requires custom BIOS flashing, Linux OS, active cooling, and extensive tinkering to reach stability.
The BC-250 has experienced a bifold lifecycle: first as a high-margin crypto-mining card (pre-2023), then as a discovery on secondary markets (eBay, AliExpress) where units now sell for $120-200 used. The community-driven unlock (as of mid-2026) has reinvigorated interest in the platform. Each chip is a silicon lottery—some boards stable at 40 CUs, others maxing out at 32 or 36—and the APU's thermals remain constrained even with active cooling.
For architects: the BC-250 is neither a consumer product nor enterprise hardware. However, it represents a case study in silicon scarcity economics and secondary-market repurposing. If you are designing low-cost inference systems or exploring heterogeneous compute for budget deployments, the availability of $200 PS5-class APUs with 16GB unified memory and working open-source ROCm support (as of August 2026) warrants evaluation. Larger implications: expect continued mining-rig repurposing as AI workloads migrate to edge devices.