However, the author notes several caveats, including limited software support for uniting multiple NPUs, unknown RAM capacity on Hailo chips, and bandwidth constraints due to the Pi's PCIe Gen 2 bus. Despite these challenges, the author believes it's not unreasonable to expect people to hack together systems with multiple NPUs, given the more attractive price of the Hailo-8L.
Key takeaways:
- The author is attempting to build a stable AI PC with all the AI accelerator chips they own, including Hailo-8, Hailo-8L, and Coral Dual Edge TPU, among others.
- They have 63 potential TOPS of neural compute available, but only 55 are available due to the limitations of the Alftel 12x PCIe M.2 adapter card they're using.
- They were able to compile the Coral's apex driver on Pi OS 12 and use it with CodeProject.AI, and also use the Hailo for inference, though examples of how to use multiple Hailo's are not easy to find.
- There are several caveats to this setup, including limited software support for uniting multiple NPUs, potential limitations on the amount of RAM available on-chip, and bandwidth constraints due to the Pi's PCIe Gen 2 bus.