Tomahawk Ultra is here, and its ground-breaking 250ns switch latency at full 51.2 Tbps throughput is redefining Ethernet for AI and High-Performance Computing (HPC).
When deployed with Scale-Up Ethernet (SUE), Tomahawk Ultra enables sub-400ns XPU-to-XPU communication latency, including the switch transit time — setting a new benchmark for tightly synchronized AI compute at scale.
But we’re not stopping there. We’re also introducing SUE-Lite, which has the same low-latency and lossless characteristics as full SUE, but with further reductions in the silicon footprint and power consumption of Ethernet interfaces on AI XPUs and CPUs.
Welcome to a new era of performance for Ethernet in HPC and AI Scale-up.
(link to media kit)
Tomahawk Ultra is here!
Welcome to a new era of performance for Ethernet in HPC and AI Scale-up. Tomahawk Ultra’s ground-breaking 250ns switch latency at full 51.2 Tbps throughput is redefining Ethernet for AI and High-Performance Computing (HPC).
virtual graveyard 
Hinterlasse einen Kommentar