Nvidia Advances AI Stack: Expands From Cloud To Colocation

Every organization wants to build a modern platform for its business. When it comes to AI, some organizations want to merge their platform, either from self-developed software or to use open-source software and commodity hardware as possible. Every AI platform buyer has an AI platform vendor. It is known to everyone that while Nvidia provides components to the hyper scaler platform, public clouds, the OEMs, and ODMs, to build an AI platform to promote, it also wants to be a leading AI platform provider.

It has the total capacity to lead the AI industry because it skillfully enhanced the AI frameworks that support its own business. Additionally, it had to build the distributed systems that its in-house researchers and developers utilize to meet the state quality of services. As a result, Nvidia became a system supplier in its own right, and now it is returning to control planes for training and inference frameworks.

Nvidia rolled out more of its software stack just before the Mobile World Congress 2021 and International Supercomputing 2021 events, which will start within a few days from now at the Equinix Analyst Day. At the same time, it announced the AI Launchpad services to expand its AI hardware from the public clouds. The hyper scaler platform controls the hardware designs and co-locates the datacenters that resemble a cloud but requires the customers to buy and install Nvidia DGX servers and host and manage by a third party with cloud-like pricing.

Equinix, the world’s biggest co-lo operator, is the first AI launchpad creator with high-performance pipes into all of the big public clouds and its Metro Edge locations. Undoubtedly, the public cloud, for others, is an expensive platform to perform production workloads over time, regardless of its scalable attribute.

You may think that having all of the primary public clouds structure and installing a GPU-accelerated framework would be adequate for Nvidia to control AI training workloads and get a robust platform to expand the use of GPUs for AI inference. However, to save the cost, enterprises worldwide adopt cloud technology and deploy data centers to protect critical data that offers a lower price, data security, and workload isolation. Enterprises want the cloud experience of reasonable pricing and easily expandable capacity without abandoning all control to a third-party cloud vendor. We all know that 50% of real-time utilization for cloud capacity is more sensible than renting or buying your infrastructure, which involves high investment. 

The process of investing in infrastructure is cumbersome and time-consuming. All the OEMs are trying to implement cloud infrastructure to turn all of their hardware cloudy regarding its consumption. They also wish to turn it into a physical asset that allows customers to control either on-premises or in a co-lo facility. In addition, the co-Los from Equinix connects immensely to the Internet support of the world, the backbones of the hyper-scalers, and the cloud builders (who transfer 70 percent of the world’s Internet traffic). A co-lo is a much better remedy than setting up your infrastructure.

Justin Boitano, the General Manager of Enterprise and Edge Computing at Nvidia, says, “Instead of in our customers saying,‘I have to buy servers. Hence, I will return post two or three months to get started,’ they can get started promptly. They can set up the infrastructure instantly and get the workload going rather than trying to build the infrastructure themselves. That is going to be helpful for the customers to get started on this journey quickly and add value to internal stakeholders before making higher capital investments.”

Companies should practice using the AI platform as soon as possible, and not everyone wants to deploy AI in the public cloud. Nvidia does not want to have the massive public clouds have so much leverage over it, either. Development and production are two completely different aspects; colocations have a much cleaner pricing model than clouds, which absorb customers with their high networking fees even if their compute and storage seems relatively inexpensive.

The pricing for AI Launchpad services intended to be in the “dollars per hour.” The Nvidia software stack is running controls to all these things. Additionally, with a VMware layer inclusion here, most enterprises are already paying for the VMware virtual infrastructure and using the same for computing, storage, and networking facility in the AI Launchpad service they are familiar with on-premises data centers. In the future, we predict that the VMware layer can be removed and replaced with a bare-metal container environment. Nevertheless, it may be an expensive option, and enterprises will pay it because they don’t have the opportunity to build their own AI container platform.

As a part of the AI Launchpad program, both Nvidia and Equinix are making investments in it. Equinix will own the hardware that customers deploy, including Nvidia, Dell, Hewlett Packard Enterprise, and Lenovo, which are OEM partners with Equinix and Nvidia. Thus, the OEMs will follow suit, and some of the ODMs may also take part in the AI launchpad program.

As per Boitano, “Equinix will roll out its AI Launchpad services in the late summer. It is starting in the United States and focusing on regions dominated by large enterprises, who want to start implementing AI in production and roll out globally.

As part of the announcement, Nvidia is rolling out its Base Command software, which is part of its enterprise AI stack. Nvidia could do its data preparation, and machine learning programs run on its supercomputers with this invention. The software costs $90,000 per month to run on Nvidia’s DGX systems. Boitano says that the company is working on getting it certified and available on OEM machines. For edge use cases, Nvidia is declaring the availability of Fleet Command in general, which is revealed at the GTC 2021 event earlier this year. The AI Enterprise is the Nvidia commercial runtime for evolving models and transforming them into inference engines. It orchestrates and manages GPU-accelerated systems at the edge that are running AI workloads. There was no disclosure on the pricing for Fleet Management.