Intel Omni-Path Interconnect: One and Done
Intel Corp.’s plans to make a big splash in the network fabric market for linking HPC, AI and other high performance workloads has apparently belly-flopped.
The chip maker confirmed to EnterpriseAI the outlines of an earlier report by the website CRN that it has jettisoned plans for a second-generation version of its Omni-Path interconnect. "Intel will no longer offer Intel OmniPath Architecture 200 (OPA200) products to customers," the company said in response to an email seeking confirmation of the CRN story.
An Intel spokesman also noted that the company would continue to "actively sell, maintain and support" its first-generation Omni-Path Architecture. "OPA100 continues to be a productive part of the DCG portfolio and we are continuing to sell, maintain and support OPA100. Intel continues to invest in connectivity solutions for our customers."
Intel had been waging an uphill battle to crack the HPC fabric market, attempting to compete with market leader Mellanox (NASDAQ: MLNX) and its market-leading Enhanced Data Rate (HDR) InfiniBand framework. The second-generation Intel fabric promised up to 200 Gbps of low-latency connections for server clusters running HPC and AI workloads.
Intel has for several years been targeting the booming HPC interconnect market, in which computing extends into the network. That capability prompted GPU leader Nvidia (NASDAQ: NVDA) to win a bidding war for Mellanox in March, paying $6.9 billion for the network fabric leader.
Mellanox began shipping a 200-Gbps version of its HDR InfiniBand interconnect earlier this year and plans a 400-Gbps version, according to CRN.
Intel highlighted its recent acquisition of Barefoot Networks is an example of its strategy of "supporting end-to-end cloud networking and infrastructure."
Editor's note: This story has been updated.