Happy to celebrate the amazing work of the #hpc teams at the Department of Energy and the NNSA at #sc24!

Last talk for #sc24 #pmbs24 is on using adaptive scheduling for helping to improve the performance of parallel file systems in #hpc clusters.

Heading into our last #pmbs24 session for the day with a discussion on power profiling for #hpc GPU machines.

Before the #sc24 #pmbs24 break we are discussing AI assisted design space analysis of #hpc Arm processors.

Back from lunch at #sc24 with a #pmbs24 paper on Arm’s ( #hpc ?) matrix extensions and whether we can assess likely performance by studying the Apple M4 chip.

Last #sc24 #pmbs24 talk before lunch! A comparison of Intel Ponte Vecchio #hpc GPU results at Argonne National Laboratory and the University of Cambridge. conferences.computer.org/sc-wpub/pdfs...

Carlo joins #pmbs24 to present performance results from AMD’s #hpc MI300 APU - the processor that will power the NNSA El Capitan exascale #supercomputer.

Next up - exploring what’s going on with Intel #hpc processors at #pmbs24 #sc24.

CPU benchmark bake off at #sc24 #pmbs24! Some of the first #hpc benchmarking of NVIDIA Grace.

Turns out that #hpc workloads on Perlmutter at NERSC are running more than 1/3 of their cycles using single precision. #sc24 #pmbs24

Back from the break with coffee and a best paper talk by Brian Austin from NERSC! Brian will talk about system wide roofline profiling to understand complex multi user #hpc workloads. #sc24

Next up! Performance modeling of AI foundation models, a paper by NERSC at the #pmbs24 #hpc conference. Room 303 at #sc24!

We are under way with the #pmbs24 #hpc workshop at #sc24. The first talk describes work from Argonne National Laboratory on benchmarking inference throughput when using open source frameworks and AI models.