• 9 Posts
  • 81 Comments
Joined 5 months ago
cake
Cake day: June 10th, 2025

help-circle

  • Ah my mistake. As I’m sure you’ve found you can certainly get USB-SATA adapters. They’ll be $15-20 each, so you’d realistically be better off getting 2 2-drive enclosures since that would be about the same price but much cleaner. I’ve used Sabrent for this for a while and they’re fine. There are occasional usb disconnects so it’s not good for anything mission critical. And never do anything port-powered when connecting drives over usb, always use parts that get their own wall power.



  • Buy a cheap LSI card on ebay, they can usually be had for around $15-20. Make sure it’s either in HBA or IT mode, or that there’s a way to put it into that mode. If it’s in HBA/IT mode, you can then just use it like more SATA ports. Buy a pack of LSI-SATA cables (there are two kinds, get the kind that includes the SATA power connector). Then you can put the card inside your computer and the drives anywhere that the cables will reach.













  • I haven’t seen any consumer NPUs that will aid with training. They’re mainly used for accelerating image effects in photoshop or blurring your background in zoom. Most aren’t even any good for inference offload. Inference and especially training take a good GPU with a large amount of VRAM (expensive) or something like a ryzen strix halo with a ton of system RAM (also expensive). With model quantization you might run modestly sized models, but you would be training tiny, tiny models at best. Think thousands of parameters, not the billions or trillions used in the LLMs you know and love.