Cheap hardware is nice, but the Expert node bar in the doc is a spec line, not a budget line: "AMD EPYC 7203 or Intel with TDX + 128GB + 2TB enterprise SSD + 500 Mbps starting bandwidth" (https://niumeta.com/en/docs/36-niumeta.html). So the thing that decides whether your 7413 + H12SSL-i + 8x16G + PM9A3 build works is whether it meets that baseline, and the doc names 7203 specifically, so double-check your CPU against it rather than assuming any EPYC fits. The doc also puts the reference figure at about $1,500 for an Expert node and calls it "meet the spec, not an arms race", so $1,200 is in the same order of magnitude, not a loophole. Two things people usually skip: seats and tasks are assigned by the protocol, L2 is hard-capped at 12,999 and one identity runs one Expert node, so a parts list reserves nothing. And the entry is not open yet; the opportunity page still says land rent, App and the distribution window are not open. Good buy for the hardware, build now, run when it opens.
Cheap hardware is nice, but the Expert node bar in the doc is a spec line, not a budget line: "AMD EPYC 7203 or Intel with TDX + 128GB + 2TB enterprise SSD + 500 Mbps starting bandwidth" (https://niumeta.com/en/docs/36-niumeta.html). So the thing that decides whether your 7413 + H12SSL-i + 8x16G + PM9A3 build works is whether it meets that baseline, and the doc names 7203 specifically, so double-check your CPU against it rather than assuming any EPYC fits. The doc also puts the reference figure at about $1,500 for an Expert node and calls it "meet the spec, not an arms race", so $1,200 is in the same order of magnitude, not a loophole. Two things people usually skip: seats and tasks are assigned by the protocol, L2 is hard-capped at 12,999 and one identity runs one Expert node, so a parts list reserves nothing. And the entry is not open yet; the opportunity page still says land rent, App and the distribution window are not open. Good buy for the hardware, build now, run when it opens.