AMD and Intel are improving GPU performance but struggle with scalability due to limited high-speed interconnects. Nvidia's NVLink and NVSwitch enable scaling up to 72 GPUs, while competitors max out at eight. Many industry players are supporting the Ultra Accelerator Link (UALink) as an alternative. Contrarily, Broadcom believes Ethernet technology suffices, citing its capability for monitoring and debugging. Broadcom is focusing on its scale-up Ethernet (SUE) for systems with over 1,024 accelerators, suggesting that its approach may rival or exceed Nvidia's existing capabilities.
"There is a huge benefit to having the same technology for all parts of the network. There's a lot of goodness that comes with using Ethernet as far as monitoring, telemetry, and debugging tools."
"Our position is you don't need to have some spec that's under development that maybe you'll have a chip a couple of years from now."
"Broadcom claims that scale-up Ethernet will support scale-up systems with at least 1,024 accelerators using any Ethernet platform."
"While Nvidia can support 576 accelerators, there haven't been any known deployments that scale beyond 72 GPU sockets."
Collection
[
|
...
]