Nvidia (Nasdaq:NVDA) has kicked off the Computex systems hardware tradeshow with the news it has opened the NVLink interconnect technology to the competition with the introduction of NVLink Fusion.
NVLink is a high-speed interconnect born out of its Mellanox networking group which lets multiple GPUs in a system or rack share compute and memory resources, thus making many GPUs appear to the system as a single processor.
Now in its fifth generation, NVLink supports up to 1.8 TB/s of bi-directional bandwidth per GPU, supporting up to 72 GPUs per rack. That’s 14 times faster than the alternative, PCI Express 5, which tops out at 128GB/s of bandwidth.
Until now, NVLink has been limited to Nvidia GPUs and CPUs, but with NVLink Fusion, non-Nvidia semi-custom accelerators will be able to use it. Nvidia says there will be two configurations for NVLink Fusion: for connecting custom CPUs to Nvidia GPUs and for connecting Nvidia’s Grace and future CPUs to non-Nvidia accelerators.
“This gives customers more choice and flexibility, while expanding the ecosystem and creating new opportunities for innovation with Nvidia at the center,” said Dion Harris, senior director of HPC, Cloud, and AI at Nvidia on a pre briefing Computext conference call with journalists.
“Nvidia ecosystem brings together leading partners to help hyperscalers deploy custom scale up computing solutions quickly and efficiently on our platform, from custom ASIC designs and NVLink fusion integrations … By combining this broad partner network with Nvidia technology, NVLink fusion, adopters can accelerate their time to market for AI factories,” he added.
Nvidia’s generosity only extends so far, however. AMD can’t use it to connect its Epyc CPU with its Instinct GPU, for example. There has to be some Nvidia product involved in the connection.
But that isn’t stopping licensees. Already the first licensees of NVLink Fusion include MediaTek, Marvell, Alchip, Astera Labs, Synopsys and Cadence.
There is an alternative to Nvidia NVLink, that is the multivendor Ultra Accelerator Link (UALink) consortium and its recently released Ultra Accelerator Link Specification. The UALink 200G 1.0 Specification defines a low-latency, high-bandwidth interconnect for communication between accelerators and switches in AI computing pods.
More Nvidia news:
>