Ohio Supercomputer Center Releases “Upward” HPC GPU Cluster – High Performance Computing News Analysis

Ascending GPU array (Credit score: Ohio Supercomputer Heart)

Columbus, Ohio — The Ohio Supercomputer Heart (OSC) has formally launched its new high-performance computing (HPC) cluster for synthetic intelligence (AI), information analytics, and machine studying.

Ascend consists of Dell PowerEdge servers with 48 AMD EPYC CPUs and 96 NVIDIA A100 80GB Tensor Core GPUs with NVIDIA NVLink and linked by the NVIDIA Quantum 200Gb/s InfiniBand platform. The middle stated it triples the OSC’s capability for AI, modeling, and simulation and provides 2 petaflops of peak theoretical efficiency. The Pitzer and Owens clusters have present capacities of 5.5 petaflops, greater than 20 petabytes of disk storage and greater than 23 petabytes of expandable backup storage.

The middle stated tutorial and business shoppers might request group time, which enhances OSC’s present Pitzer and Owens teams. Ascend is OSC’s first GPU platform.

stated David Hudak, OSC’s govt director. “We’re thrilled to have the ability to supply this key new useful resource to the HPC neighborhood and assist prospects’ development in tutorial analysis and business applied sciences.”

Though Ascend is smaller than the Pitzer and Owens clusters, with solely 24 compute nodes, “its peak efficiency is on par with all of the others, primarily because of the efficiency that the GPUs add to those nodes,” stated Doug Johnson. OSC Affiliate Director.

Johnson stated the Ascend suite permits prospects to shortly switch information externally by 25Gbps Ethernet community interfaces and utilizing NVIDIA’s Quantum Platform with in-network computing capabilities.

David Hudak, OSC

“OSC’s Buyer Providers and Scientific Purposes groups will probably be out there to assist our prospects decide if their functions could make good use of progressive GPUs,” Johnson stated. “For some functions, there’s a enormous efficiency profit to utilizing GPUs and Ascend will enable our prospects to deal with among the points that our present programs cannot remedy.”

OSC Ascend was examined from late October to mid-December with a choose group of consumers to evaluate system efficiency and solicit suggestions from customers.

“A part of the purpose from the early rollout interval was to get a greater understanding of how consumer functions use the GPUs we assist within the system,” Johnson stated. “We are going to proceed to enhance software program and system administration as we be taught extra of what we encounter that helps early adopters and the system up and working for an extended time period.”

Yu Su, an assistant professor within the Division of Laptop Science and Engineering at Ohio State College, used Asc throughout the early testing interval for a number of analysis tasks within the discipline of pure language processing. Ascend’s NVIDIA GPUs allowed Su to make use of BLOOM-176B, one of many largest neural community fashions, for the primary time on OSC programs.

“Many different tasks—from answering questions by way of giant information bases to knowledge-based reasoning and language interfaces to relational databases—have benefited in my mixture of NVIDIA A100 GPUs on the rise,” Su stated. “College students report two to 3 instances quicker processing velocity than the A6000 GPU servers in my group.”

Bargeen Turzo, a graduate scholar in Steffen Lindert’s computational chemistry group at Ohio State, discovered that Ascend might course of capabilities that predicted protein complicated buildings quicker than Pitzer’s group might.

“For some giant proteins, I could not even get a single Bitzer prediction after doing the mathematics for a number of weeks,” Turzo stated. “Through the ascent, the identical calculation completed in 12 hours.”

Ascend follows the identical pricing construction because the Owens and Bitzer teams. Extra details about Ascend {hardware}, software program surroundings, file programs, batch particulars, and connection particulars will be discovered on the Ascend Assortment web page of the OSC web site.

Along with launching Ascend, OSC is within the planning levels of creating a high-performance computing cluster that may exchange its Owens cluster, scheduled for launch in 2023. The brand new cluster will initially work at the side of Owens and complement Pitzer and Ascend.

Rajesh Bohani, Dell Applied sciences

says Rajesh Bohani, Vice President, PowerEdge, Core Compute and Excessive Efficiency Compute, Dell Applied sciences. “Ascend’s AI capabilities, powered by highly effective PowerEdge XE8545 servers, will complement and exponentially develop the superior computing sources important to engineering innovation and scientific discovery, finally serving to advance human progress.”

“Trendy scientific computing is pushing the boundaries of HPC and AI at scale,” stated Timothy Costa, director of HPC and Quantum Product at NVIDIA. “As firms proceed to innovate and develop their portfolios, NVIDIA will empower OSC and different key researchers around the globe to advance engineering and scientific discoveries.”

“AMD EPYC processors assist researchers around the globe with the efficiency and productiveness wanted to reply a few of science’s largest questions,” stated Brock Taylor, director of high-performance computing, AMD. “We’re excited that the Ascend supercomputer will assist the Ohio Supercomputer Heart additional its mission of advancing ranges of synthetic intelligence, machine studying, large information, and information analytics.”

Supply: Ohio Supercomputer Heart

Leave a Comment