zlacker

>>mindra+(OP)
As someone with a similar background to the writer of this post (I did avionics work for NASA before moving into more “traditional” software engineering), this post does a great job at summing up my thoughts on why space-based data centers won’t work. The SEU issues were my first though followed by the thermal concerns, and both are addressed here fantastically.

On the SEU issue I’ll add in that even in LEO you can still get SEUs - the ISS is in LEO and gets SEUs on occasion. There’s also the South Atlantic Anomaly where spacecraft in LEO see a higher number of SEUs.

>>kevdev+N6
Single event upsets are already commonplace at sea level well below data center scale.

The section of the article that talks about them isn’t great. At least for FPGAs, the state of the art is to run 2-3 copies of the logic, and detect output discrepancies before they can create side effects.

I guess you could build a GPU that way, but it’d have 1/3 the parallelism as a normal one for the same die size and power budget. The article says it’d be a 2-3 order of magnitude loss.

It’s still a terrible idea, pf course.

>>hedora+yw1
It strikes me that neutral network inference loads are probably pretty resilient to these kinds of problems (as we see the bits per activation steadily decreasing), and where they aren't, you can add them as augmentations at training time and they will essentially act as regularization.