At the Global Supercomputer (ISC 2022) trade exhibit, HPE shown blade systems that will electricity two exascale supercomputers established to appear on line this year — Frontier and Aurora. Unfortunately, HPE experienced to use advanced and electricity-hungry components to get unprecedented computing performance. For this reason, each equipment use liquid cooling, but even large h2o blocks simply cannot conceal some intriguing style peculiarities the blades element.
Equally Frontier and Aurora supercomputers are developed by HPE employing its Cray EX architecture. Even though the devices leveraged AMD and Intel hardware, respectively, they use higher-performance x86 CPUs to run normal tasks, and GPU-based mostly compute accelerators to operate very parallel supercomputing and AI workloads.
The Frontier supercomputer builds on HPE’s Cray EX235a nodes (opens in new tab) powered by two AMD’s 64-core EPYC ‘Trento’ processors showcasing the firm’s Zen 3 microarchitecture improved with 3D V-Cache and optimized for large clocks. The Frontier Blades also appear with eight of AMD’s Instinct MI250X accelerators (opens in new tab) featuring 14,080 stream processors and 128GB of HBM2E memory. Every node gives peak FP64/FP32 vector efficiency of around 383 TFLOPS and peak 765 FP64/FP32 matrix efficiency of approximately 765 TFLOPS. Both of those CPUs and compute GPUs employed by HPE’s Frontier blade use a unified liquid cooling method with two nozzles on the entrance of the node.
The Aurora blade (opens in new tab) is at this time termed just like that, carries an Intel badge, and does not have HPE’s Cray Ex model number nevertheless, possibly because it still demands some sharpening. HPE’s Aurora Blades benefit from two Intel Xeon Scalable ‘Sapphire Rapids’ processors with in excess of 40 cores and 64GB of HBM2E memory for every socket (in addition to DDR5 memory). The nodes also aspect six of Intel’s Ponte Vecchio (opens in new tab) accelerators, but Intel is peaceful about the exact specs of these beasts that pack over 100 billion transistors each and every (opens in new tab).
One point that catches the eye with the Aurora blade established to be utilized with the 2 ExaFLOPS Aurora supercomputers (opens in new tab) is mysterious black containers with a triangular ‘hot surface’ sign found upcoming to Sapphire Rapids CPUs and Ponte Vecchio compute GPUs. We do not know what they are, but they could be modular complex power supply circuitry for more versatility. Following all, back in the day, VRMs have been removable (opens in new tab), so employing them for extremely ability-hungry parts might make some perception even today (assuming that the accurate voltage tolerances are fulfilled), especially with pre-output hardware.
Once again, the Aurora blade utilizes liquid cooling for its CPUs and GPUs, nevertheless this cooling process is solely distinct from the one particular utilized by Frontier blades. Intriguingly, it appears like Ponte Vecchio compute GPUs in the Aurora blade use unique h2o blocks than Intel demonstrated (opens in new tab) a number of months in the past while we can only marvel about feasible good reasons for that.
Curiously, the DDR5 memory modules Intel-based blade takes advantage of arrive with somewhat formidable warmth spreaders that glance even larger than all those utilised on fanatic-grade memory modules. Trying to keep in head that DDR5 RDIMMs also carry a ability management IC and voltage regulating module, they in a natural way want superior cooling than DDR4 sticks, particularly in area-constrained environments like blade servers.