AMD Supercomputer Frontier, still not fully operational.

Lakados

[H]F Junkie
2FA
Joined
Feb 3, 2014
Messages
13,956
https://www.techspot.com/news/96254...computer-has-hardware-failure-every-day.html?
TLDR;
The system is online but performance is hampered because it suffers hardware failures every few hours.
The project has been riddled with rumors of huge problems from the MI250x being very temperamental and that they are causing lots of setbacks, the 250 series is AMD's first chiplet-based GPU.
Their project lead refutes that the GPUs are the cause by saying the system has lots of categories where it's failing and GPUs are just one, but they aren't out of the ordinary for a system of that scale.
 
I would hope they can get the gpus down to a reliable level. Especially considering nearly all gpus are extremely stable even in cluster currently. Hard to know what exactly is causing issues without any decent details
 
Back
Top