Nvidia wants to steer customers and partners to buy select GPU servers from OEMs, ODMs and other system builders that have been certified to provide optimized AI performance as part of a new program.
The company announced the new Nvidia-Certified Systems program, which it called an industry first because of the emphasis on testing the servers for “real life” workloads for AI and ensuring that such workloads can be supported in large server deployments.
Nvidia kicked off the program with 14 certified GPU servers from Dell Technologies, Hewlett Packard Enterprise, Gigabyte, Supermicro and Inspur. However, the chipmaker said it anticipates the server roster to grow nearly 70 from at least 11 system makers in the future. There is no cost for OEMs and other partners to participate in the certification program.
All systems have been certified to provide optimized performance for AI applications, frameworks and domain-specific software development kits from Nvidia’s NGC repository of GPU-accelerated software. Customers who buy certified systems will also receive enterprise support from Nvidia and OEM partners
Adel El-Hallak, director of product management for NGC at Nvidia, said while the chipmaker will continue to support those who want to build their own systems, the company will recommend customers to buy servers that are certified under the new program.
“Ideally, we’d like them to buy the Nvidia-certified systems because we know that these will run optimally [and] meet the performance threshold that we expect,” he said.
El-Hallak said Nvidia partners should focus on adding new value around the creation of AI applications rather than trying to tinker with systems when Nvidia is already doing the groundwork.
“We’d rather that they spend the time doing what they do best, which is creating AI and deriving new business values rather than spending time figuring out how to [put] these systems together,” he said.
The Nvidia-Certified Systems include servers with as many as eight A100 GPUs as well as high-speed InfiniBand or Ethernet adapters from the company’s Mellanox networking business. The systems can also include Mellanox ConnectX-6 SmartNICs and BlueField-2 DPUs.
In order to pass certification, the servers must pass tests on deep learning training and inference, machine learning algorithms, intelligent video analytics and network and storage offload, all of which are based on software from Nvidia’s NGC catalog.
Servers that have been certified so far include the Dell EMC PowerEdge R7525, Gigabyte R281-G30, HPE Apollo 6500 Gen10 System, Inspur NF5488A5 and Supermicro A+ Server AS -4124GS-TNR.
Matthew DuBell, a consulting solutions architect for the Business and Analytics Advisors group at St. Louis, Mo.-based World Wide Technology, a top Nvidia AI partner, said the new certification program shows that GPU computing and accelerated computing are “here to stay in the enterprise.”
“Nvidia has definitely taken a huge lead in being able to offer these types of programs, and it’ll be interesting to see how other players in the market react,” he said.
DuBell said the program will help take out a lot of the guesswork of how to optimize systems for AI while also giving customers more confidence in the infrastructure investments they’re making.
“Organizations don’t want a heterogeneous mix of server vendors or types. They tend to prefer a baseline from an OEM, so being able to offer a validated design and performance from a preferred OEM is going to be a benefit to a lot of IT organizations,” he said.
For World Wide Technology, the certification program will allow the value-added reseller to focus on providing its own differentiated services for AI and data science, which include managing the data, the building out of initial AI pipelines and deploying large production environments, according to DuBell.
“A lot of people will talk about how data is the next oil. But when you think about oil, it’s not useful until it’s refined,” he said. “So being able to help a company refine its data so that it’s truly insightful and allows them to grow their business is really what we’re focusing these data services on.”
Dominic Daninger, vice president of engineering at Burnsville, Minn.-based Nor-Tech, an Nvidia high-performance computing system integrator, said he is not concerned that the certification program puts more emphasis on pre-built systems over those that Nor-Tech may custom build.
“If there aren’t other issues in the way, we would probably go for a certified system just because it’s one less surprise you might get,” he said. “But there’s usually many factors weighed [when selling servers], and certification, we look at it as a plus.”