Discovery should do higher than that. As well as, in accordance with the RFP, it should help every part from small workloads utilizing 20% of the nodes to those who require the complete system. It have to be expandable with “new, novel architectures,” and interoperate with and help related DOE experimental person amenities and different ORNL Management Computing Facility (LCF) infrastructure.
It additionally have to be operational earlier than the tip of Frontier’s service life, function inside OLCF’s operations and utilities funds, present a productive programming atmosphere for customers, and “proceed to make progress and lead in dramatically enhancing power effectivity throughout the ecosystem.”
Oh, sure — and it have to be AI-friendly, after all.
Discovery ought to “be on the forefront in supporting area scientists and software builders as they discover and combine transformational AI applied sciences to speed up discoveries in science, power, and safety issues of nationwide significance,” the RFP outlined.
As effectively, OLCF added in its RFP, “We envision a large spectrum of use circumstances starting from inverse design and management of advanced methods akin to energy grids and nuclear reactors, to generative AI and foundational fashions that combine textual content and pictures which might be typically unstructured, high-resolution, and from multi-modal knowledge sources.
“Executing AI-empowered computing campaigns and workflows will place new calls for on the system structure, presumably requiring extra interconnect bandwidth and an optimized storage layer that may deal with very excessive charges of I/O operations (IOPS) centered on random reads.”
