Briefly: Frontier, the world’s strongest supercomputer, is on-line however nonetheless removed from operational. Its director has confirmed studies that it’s experiencing a system failure each few hours, however insists that is par for the course.
Frontier is in a category of its personal. It has 9,408 HPE Cray EX235a nodes, every powered by an AMD Trento 7A53 Epyc 64-core CPU outfitted with 512 GB of DDR4, and 4 AMD Intuition MI250X GPUs / accelerators every outfitted with 128 GB of HBM2e. Summed, the system has 602,112 CPU cores and eight,138,240 GPU cores in whole, and 4.6 PB of each DDR4 and HBM2e.
In Could, Frontier joined the TOP500 as the primary supercomputer to interrupt the exascale barrier after it accomplished the HPL benchmark with a rating of 1.102 ExaFlops/s. Since then, the Oak Ridge Nationwide Laboratory in Tennessee, which manages the supercomputer, has been readying it for scientific analysis scheduled to begin in January.
Nonetheless, there have been studies that the launch of Frontier may very well be waylaid by extreme {hardware} failures. Searching for solutions, Inside HPC organized an interview with the Program Director at Oak Ridge, Justin Whitt. Within the interview, he confirmed Frontier was experiencing day by day system failures however asserted that was inevitable in such a big system.
“Imply time between failure on a system this measurement is hours, it isn’t days,” he stated. “So you have to be sure you perceive what these failures are and that there is not any patterns to these failures that you have to be involved with.” Whitt added that going a day with out a failure “could be excellent.”
“Our purpose continues to be hours.”
There have been rumors that the {hardware} issues have been being attributable to the brand new AMD Intuition MI250X, however Whitt refuted them. The MI250X is AMD’s strongest GPU/accelerator, and it solely sells it to pick out companions. It has 220 CUs containing 14,080 cores clocked at 1700 MHz in a 500 W bundle.
“The problems span plenty of completely different classes, the GPUs are only one,” Whitt remarked. “It has been a fairly good unfold amongst widespread culprits of components failures which have been a giant a part of it. I do not suppose that at this level that we’ve got plenty of concern over the AMD merchandise,” he added.
“We’re coping with plenty of the early-life sort of issues we have seen with different machines that we have deployed, so it is nothing too out of the unusual.”
Whitt conceded that the unprecedented scale of Frontier had made nice tuning it “just a little bit more durable” however stated they have been nonetheless following the schedule set again in 2018-19 regardless of delays attributable to the pandemic.
Head over to Inside HPC to learn the complete interview.
