Major tech computing firms have formed a consortium to define a brand new open standard for connecting AI accelerators. NVIDIA was not invited to change into a part of the group, although the corporate is by far the biggest supplier of AI GPUs.
AI data centers must move massive amounts of knowledge with very low latency. High-bandwidth data processing on GPUs is amazingly fast, however the challenge is moving data inside and between clusters of those AI accelerators in data centers.
NVIDIA has developed NVLink, its proprietary high-speed interconnect specifically designed for communication between its GPUs. The problem is that NVLink is proprietary and subsequently only works with NVIDIA GPUs.
AMD, Broadcom, Cisco, Google, Hewlett Packard Enterprise (HPE), Intel, Meta and Microsoft have announced the formation of the Ultra Accelerator Link Promoter Group. The goal of the group is to define and promote an open standard called Ultra Accelerator Link or UALink.
The idea is that UALink might be adopted by the industry as the usual solution for high-bandwidth, low-latency data transmission between AI accelerators in data centers.
Similar protocol standardization efforts have been critical to the technology industry previously. Because now we have open standards just like the PCI bus, Ethernet, or TCP/IP, hardware and software from different manufacturers can interconnect.
This could possibly be one among the explanations NVIDIA wasn't invited to the party. If the consortium of technology firms can agree on an open networking standard for the industry that isn't influenced by NVIDIA's technology, it could help break the quasi-monopoly NVIDIA seems to have.
AMD and Intel are direct competitors of NVIDIA within the GPU market, and Microsoft and Google are each developing their very own AI hardware.
“An industry specification might be critical to standardize the interface for AI and machine learning, HPC (high-performance computing) and cloud applications for the following generation of AI data centers and implementations,” the consortium said in an announcement.
Version 1.0 of UALink is anticipated to be ready within the third quarter of 2024 and might be made available to firms joining the Ultra Accelerator Link (UALink) consortium.
NVIDIA's absence doesn’t necessarily mean they’re permanently excluded. The consortium may decide to welcome them in the longer term, and NVIDIA may decide to adopt UALink if it gains widespread industry acceptance.
