GigaIO Wins TACC Contract for Composable HPC – AI Infrastructure – insideHPC – insideHPC

San Diego, March 10, 2022 – GigaIO, maker of data center rack-scale architecture for artificial intelligence and HPC, today announced that production has begun on their composable disaggregated infrastructure testbed in the Lonestar6 system at the Texas Advanced Computing Center (TACC) at The University of Texas at Austin.

Lonestar6 is a 600-node system utilizing Dell Technologies servers powered by AMD EPYC “Milan” CPUs and A100 GPUs from NVIDIA. It is the first CDI platform at TACC. CDI is designed to deliver “pools” of compute over a software-defined PCIe-based memory fabric, providing access to more processing power and storage when needed, and allowing for sharing of resources, thereby increasing utilization.

GigaIO is providing Lonestar6’s fabric infrastructure, including switches, cards, cables, JBOGs (Just a Bunch Of GPUs), and composition software. “GigaIO’s composable infrastructure solution democratizes access to expensive specialized resources such as accelerators, which it shares across users and workloads in a way that is simple to implement for IT managers,” said Alan Benjamin, CEO of GigaIO. “FabreX will allow TACC to use any accelerator for any job without limitation, in a future-proof solution that can continue to grow as their processing needs grow.”

GigaIO said its composable infrastructure platform pairs flexibility with the agility of  cloud computing, “allowing researchers to build completely customized and otherwise impossible servers for their AI and HPC workflows. GigaIO is the only CDI vendor that can transform each server to an entire rack using only PCIe for the absolute lowest latency and highest bandwidth throughout.”

GigaIO said it does this through its universal composable fabric, FabreX, “that can transform a rack of servers into a true rack-scale system without proprietary architecture lock-in.

“Making this happen requires both the ability to disaggregate and compose resources to servers, and the ability to run internode communications on the same network,” the company said. “FabreX orchestrates workloads by configuring any resource on the fly and integrating networking, storage, memory, and specialized accelerators into a single-system cluster fabric. Composing resources with FabreX dramatically lowers OpEx and CapEx costs due to increased resource efficiency, while improving both serviceability and upgradeability.

GigaIO said TACC chose its CDI offering for its ability to let customers “seamlessly choose the best hardware accelerator (or number of accelerators) for each workload; the green savings on cooling, power, and footprint available through increased utilization and minimized server requirements; and most importantly, because of GigaIO’s truly open platform for heterogeneous architectures. GigaIO offers the only composable solution that does not require any proprietary orchestration software. Instead, the company is committed to being an open standards platform, and to working with leading northbound integration software vendors to integrate natively for ease of deployment.”

TACC Executive Director Dan Stazione with GigaIO rack

“GigaIO allows us to mix the type and number of accelerators attached to each node, varying them with the particular mix of jobs at any given moment in time,” said Dan Stanzione, executive director of TACC. “GigaIO’s ability to efficiently scale resources across all open standards software and hardware is just one of the reasons we selected them and will be continuing to work with them on future cluster enhancements.”

TACC designs and operates some of the world’s most powerful computing resources. The center’s mission is to enable discoveries that advance science and society through the application of advanced computing technologies. Lonestar6 is the latest cluster environment that will be utilized for this work, including computational fluid dynamics, material science, and climate science.

Spread the love

Leave a Reply

Your email address will not be published.