Keeping this in mind, would it make sense to frame it not as anomalous but as a reaction to a competing fascist/imperialist power?

Thank you to TrueAnon for validating me

There's supposed to be a refresher course coming up so if I get in I can take some notes and DM you the details

I don't think I can ballpark an estimate because it's been a while since I did the intro course. It might be a max of 96 GPUs per compute node in a pod but I'm not sure how many per pod + how many pods
And of course the hardware is replaced Just-In-Time so there's no stockpiles on site

As in total GPUs?

I can't speak with a ton of confidence on the exact process either because my team isn't directly involved in the hardware support

The RMA process is handled by a third party vendor. I'd estimate on average 1-2 per week spread across 4-6 DCs although I don't know the exact count of how many SuperPod deployments. It doesn't sound like a lot but comparing to normal server hardware it's waaay more frequent that HDD or RAM failures

Pretty sure H100s and A100s only, I'm not sure what the rollout strategy is for Blackwell/if it's already been rolled out at newer builds

Plus the heat generated is nuts. We've had shitcoin miners able to operate in normal air cooled data centers but AI loads require a complete redesign for power/cooling

It's the combination of load + heat. The cooling requirements for NVIDIA's SuperPods are on another level. If you ran your PC full tilt on and off multiple times per day every day in high heat you'd have some component fail much sooner

I work for a data center and just want to confirm the failure rate for the GPUs is insane. Our teams replace multiple per week. And it's not even defect related, the heat and load just wear them out so quickly

Literally can't think of even a single idpol topic the NDP were focused on in the last election