for starters, at the loads they’re running at, they have literally hundreds of gpu failures a day. How do you propose doing that in space?
Comment on An in-space construction firm says it can help build massive data centers in orbit
FaceDeer@fedia.io 3 days agoThey'd better not try to sell them to anyone who has access to an engineer, then. Just a single engineer will bring the whole scheme crashing down.
vrighter@discuss.tchncs.de 3 days ago
FaceDeer@fedia.io 3 days ago
Include spares.
I hope they're reading this thread and taking notes, they probably didn't think of that.
vrighter@discuss.tchncs.de 3 days ago
and the infrastructure and robotics to replace them, of course.
Assuming 200 nvidia H100 failures a day (conservativo, reality is worse) that’s an extra ~340kg of weight you’d need to launch per day. Which is an extra 120 tons yearly.
FaceDeer@fedia.io 3 days ago
So, one Starship launch per year. Doesn't sound like a problem.
vrighter@discuss.tchncs.de 3 days ago
and forget about running 4nm chips in space. shit has to be radiation hardened, which means bigger process nodes and higher energy cost, and lower speed
FaceDeer@fedia.io 3 days ago
Another thing they probably didn't think of. Nobody's run chips in space before.
vrighter@discuss.tchncs.de 3 days ago
they did think of it. lots of people have. I just mentioned what was required. Rad hardened processors are usually 10 to 20 times slower than what we have on the surface
FaceDeer@fedia.io 2 days ago
So don't use rad-hardened processors, put them in a radiation vault instead. Those become more mass-efficient the more hardware you're putting inside them.
Really, I assure you the people proposing these things have put more thought and expertise into it than a bunch of random Fediverse commentators.