Nvidia launched a monster field yesterday referred to as the HGX-2, and it’s the stuff that geek goals are manufactured from. It’s a cloud server that’s presupposed to be so highly effective it combines high-performance computing with synthetic intelligence necessities in a single exceptionally compelling package deal.
You already know you need to know the specs, so let’s get to it: It begins with 16x NVIDIA Tesla V100 GPUs. That’s good for 2 petaFLOPS for AI with low precision, 250 teraFLOPS for medium precision and 125 teraFLOPS for these instances whenever you want the very best precision. It comes commonplace with a 1/2 a terabyte of reminiscence and 12 Nvidia NVSwitches, which allow GPU to GPU communications at 300 GB per second. They’ve doubled the capability from the HGX-1 launched final yr.
Paresh Kharya, group product advertising supervisor for Nvidia’s Tesla information heart merchandise, says this communication pace permits them to deal with the GPUs primarily as a one big, single GPU. “And what that permits [developers] to do isn’t just entry that huge compute energy, but additionally entry that half a terabyte of GPU reminiscence as a single reminiscence block of their packages,” he defined.
Sadly you gained’t be capable of purchase one in every of these packing containers. Actually, Nvidia is distributing them strictly to resellers, who will doubtless package deal these infants up and promote them to hyperscale information facilities and cloud suppliers. The fantastic thing about this method for cloud resellers is that once they purchase it, they’ve the complete vary of precision in a single field, Kharya stated.
“The good thing about the unified platform is as firms and cloud suppliers are constructing out their infrastructure, they will standardize on a single unified structure that helps the complete vary of high-performance workloads. So whether or not it’s AI, or whether or not it’s high-performance simulations, the complete vary of workloads is now attainable in only a single platform,”Kharya defined.
He factors out that is notably necessary in large-scale information facilities. “In hyperscale firms or cloud suppliers, the primary profit that they’re offering is the economies of scale. If they will standardize on the fewest attainable architectures, they will actually maximize the operational effectivity. And what HGX permits them to do is to standardize on that single unified platform,” he added.
As for builders, they will write packages that benefit from the underlying applied sciences and program within the actual stage of precision they require from a single field.
The HGX-2 powered servers shall be accessible later this yr from accomplice resellers, together with Lenovo, QCT, Supermicro and Wiwynn.