I watched a video about the AMD BC-250, which is based on partially disabled PS5-style APUs. The creator reportedly played games on a single node, and it seems like adding more nodes—with proper power delivery and cooling—could make the system much faster. At the prices these boards originally sold for, it looked like an unusually powerful setup.
Why aren't systems like this more common for gaming? Is the main problem power consumption, cooling, Linux compatibility, or the fact that the hardware is weaker and more restricted than it appears? Can multiple BC-250 nodes be combined to improve performance in one game, or would each node only be useful for running a separate game or virtual machine?
4 Answers
The boards were originally made from partially defective PS5 APUs and aimed at cryptocurrency mining. Mining benefits from having lots of mostly independent devices working in parallel, so a box full of nodes made sense for that workload. Gaming is different because one game usually needs one graphics device with fast, coordinated access to its CPU, GPU, and memory.
They can be decent as inexpensive Linux machines or experiments, but they have a lot of practical disadvantages for a gaming PC: limited software support, awkward hardware configuration, uncertain reliability, and no useful way to pool multiple boards for normal games. Once you add the cost of power, cooling, storage, a suitable case, and troubleshooting, a conventional graphics card is usually much simpler and better supported.
The low prices made the BC-250 interesting, but the hardware was never a normal, fully enabled consumer APU. AMD disabled or locked parts of it, and it was designed around a very specific mining product. Recent community modifications can restore some functionality, but results vary because these chips started as defective or marginal parts. Even a good example isn’t automatically equivalent to a complete PS5, and it still needs unusual drivers, Linux setup, power delivery, and cooling.
You generally can’t combine several BC-250 boards into one faster gaming GPU. Games need to be specifically designed for multi-GPU rendering, and even older technologies such as SLI, CrossFire, and certain DirectX 12 modes had poor support and significant drawbacks. Adding more nodes usually means running separate game instances, not increasing the frame rate of one instance.
So a multi-node gaming setup would mainly be useful as a game server or for letting several people run separate games at once, rather than pooling the hardware for one player?