Seeking to solidify its place because the dominant international provider of chips that assist generative AI workoads, Nvidia introduced new GPUs and servers in addition to a spread of latest software program choices on the SIGGRAPH convention in Los Angeles this week.
On the {hardware} aspect, Nvidia introduced a brand new line of servers, the OVX collection. The server line is designed to make use of as much as eight of the corporate’s L40S GPUs. The GPUs are primarily based on the corporate’s Ada Lovelace structure, which succeeded Ampere because the microarchitecture in use in its important line graphics playing cards. Every L40S packs 48GB of reminiscence and is designed with advanced AI workloads in thoughts, boasting 1.45 petaflops of tensor processing energy.
It is much like the method Nvidia has taken previously with its client graphics card designs, in that the corporate plans to promote some OVX servers instantly and as reference designs, however different producers (on this case, Dell, ASUS, Gigabyte, HPE, Lenovo, QCT and Supermicro) will function international system builders. The L40S will develop into out there within the fall, and the corporate mentioned that OVX programs will go on sale quickly after.
As a part of an improve to its AI Enterprise software program line, Nvidia additionally launched a brand new product referred to as AI Workbench, which is designed to be a type of self-assembly package for AI builders. The system comes with pretrained fashions and an array of instruments that can be utilized to customise them, with the thought of saving appreciable growth time. Nvidia additionally introduced quite a few options designed so as to add generative AI capabilities to its different product traces,together with an AI developer “co-pilot” for its Omniverse 3D imaging software program.
How Nvidia targets completely different units of customers
Most of the firm’s latest AI-related releases are focused at completely different customers — together with cloud service suppliers, builders, and server makers. That’s a key a part of Nvidia’s technique, in response to Shane Rau, analysis vice chairman at IDC.
“If the tip buyer’s a cloud service supplier, they could simply need, say, a server GPU board,” he mentioned. “Some clients want to purchase the Nvidia silicon but additionally purchase the entire system round it — LVX, OVX, and so forth. Then possibly the following stage is you purchase the {hardware} however possibly you additionally want some coaching.”
One other necessary strategic level, in response to Rau, is Nvidia’s flexibility. That flexibility began as way back as 2012, when the corporate launched its first server GPUs, with the CUDA developer atmosphere that allowed them to be reprogrammed and optimized for various duties, and has continued with the varied AI-related items of software program that Nvidia has launched. The one place, in truth, the place the corporate tends to cease providing options is when it will encroach instantly on an finish consumer’s personal area.
“AI will be very end-user particular,” Rau mentioned. “Normally the tip consumer brings in their very own experience — agriculture, monetary evaluation, and so forth. So Nvidia wanst to convey the extent of resolution that you simply’re wiling to spend money on all the way in which as much as your particular area, however you present the particular experience.”
It’s been a extremely profitable technique for the corporate within the AI market, Rau added, provided that Nvidia is the most important supplier of silicon for AI use by a long way.
“I’d say this was at all times within the playing cards for them,” he mentioned.
Copyright © 2023 IDG Communications, Inc.
