South Africa recycling supercomputer hardware

[Luis]

South Africa recycling supercomputer hardware

South African researchers are reusing decommissioned supercomputer hardware to bolster the country's high-performance computing infrastructure.

An initiative led by the Council for Scientific and Industrial Research (CSIR) sees this retired research computing hardware used at the University of the Witwatersrand and at institutions across Africa.

 

[KernelPanic42]

Great work

 

[Justanotherguybythesea]

Yes, you need a good supercomputer, not the latest greatest, use the last generation, or the one before, which, while it is not as cutting edge, does have a major advantage in that it will do 95% of the tasks, at a few percent lower performance, and at a fraction of the cost to buy. Yes power use is slightly more per unit of work, but it will take a decade before that extra will make up for the difference in cost of buying new. Very common for older hardware to be used for less demanding jobs, simply because it still has a good lifetime ahead, and the new job is within the capacity as well.

 

[brattex]

Yes, you need a good supercomputer, not the latest greatest, use the last generation, or the one before, which, while it is not as cutting edge, does have a major advantage in that it will do 95% of the tasks, at a few percent lower performance, and at a fraction of the cost to buy. Yes power use is slightly more per unit of work, but it will take a decade before that extra will make up for the difference in cost of buying new. Very common for older hardware to be used for less demanding jobs, simply because it still has a good lifetime ahead, and the new job is within the capacity as well.

In this case, some of these HPC Ecosystems Project systems aren't even "run of the mill" repurposed systems. These are decommissioned systems that - when switched OFF - were ranked at #20 in the world (Top500).

It's a major testament to Dell engineering that these systems have been running (reliably, nogal) for 15ish years.

Even more so given that they were threshed to pieces from 2012 to 2017 as the flagship HPC system for Texas Advanced Computing Center.

Rightly stated that in this case the major cost is no longer procurement but operational. We've been working on novel approaches for green computing now to maximise the lifespan of these systems and to minimise the cost. DVFS and Core Scaling are showing some interesting opportunities to map out a strong Joule/FLOP relationship.