"Oracle on Tuesday revealed it would field more than 18 zettaFLOPS worth of AI infrastructure from Nvidia and AMD by the second half of next year. This includes a cluster of 800,000 Nvidia GPUs capable of delivering up to 16 zettaFLOPS of peak AI performance - that's sparse FP4 in case you're wondering. The cluster, part of Oracle Cloud Infrastructure's Zettascale10 offering, is a big win for Nvidia, which isn't only furnishing the GPUs and rack systems, but also the networking."
"While Nvidia counts its billions, AMD expects to see 50,000 of its MI450X-series accelerators deployed at Oracle data centers in the second half of next year, with additional deployments expected the following year. First teased at AMD's advancing AI event in June, the MI450X will be offered in a rack scale architecture similar to Nvidia's NVL72 that it calls Helios. Each rack is equipped with 72 MI450X GPUs stitched together using an open alternative to Nvidia's high-speed NVLink interconnect called Ultra Accelerator Link (UALink)."
Oracle will field more than 18 zettaFLOPS of AI infrastructure from Nvidia and AMD by the second half of next year. The deployment includes a cluster of 800,000 Nvidia GPUs capable of up to 16 zettaFLOPS peak AI performance using sparse FP4. Nvidia will supply GPUs, rack systems, and Spectrum-X Ethernet switches to interconnect the GPUs. Oracle will offer multiple Nvidia AI services through its cloud. AMD plans to deploy about 50,000 MI450X accelerators in the same timeframe and more later. AMD's Helios rack design uses 72 MI450X GPUs linked via Ultra Accelerator Link and targets multi-exaFLOPS performance with large HBM4 capacity.
Read at Theregister
Unable to calculate read time
Collection
[
|
...
]