Nvidia program gets GPUs slotted in edge services from Digital, EdgeConnex and more

DGX hardware (which incorporates Nvidia Volta GPUs) and software were created to carry artificial intelligence workloads and are optimized for AI training.

Earlier this month, Digital Realty said it had launched its Data Hub component and services line with Nvidia DGX systems. The combination, according to Digital Realty executives, makes it possible to quickly deploy AI and machine learning workloads on their PlatformDigital offerings which help companies deal with changing data, control and networking demands.

Executives say that Data Hubs on PlatformDigital create data-exchange centers on which buyers can securely build out organizational artificial intelligence and machine learning projects. In short, Digital Realty can quickly deploy a rack of equipment and network connectivity at any one of 25 data centers around the globe, including markets like Fortaleza (Brazil) Houston, Texas (US), and Seoul (South Korea), among others. Availability is planned for more markets later this year. Specifications from Digital Realty show DGX pods take up a minimum of 2 cabinets at 35kW per cabinet.

Also this month, Core Scientific launched its pilot AI-platform as a service, the so-called Cloud for Data Scientists. The pilot is located at an Equinix International Business Exchange data center. It incorporates Nvidia’s DGX compute-acceleration systems and is designed for AI and deep learning roles.

EdgeConneX has also incorporated DGX systems into its colocation edge facilities, which are found in 30 markets in North and South America and Europe. In doing so, it is taking artificial intelligence and machine learning capabilities from distant cloud services and placing them close enough to buyers that latency becomes a much smaller factor for information technology performance.

Global spending on infrastructure services (including colocation) for public edge cloud is expected to reach $6.2B by 2025, according to a forecast from Analysys Mason. That sum represents 19% of the $34B overall public edge cloud market opportunity.

Analysis

Where Digital Realty and Equinix have offered data center services in more traditional datacenter markets, EdgeConneX has facilities in places like Denver, CO; Tallahassee, FLA; Warsaw, Poland; and Santiago, Chile. For some customers, Digital Realty’s locations will serve as an extended geographic “edge” datacenter. EdgeConnex’s data centers are more akin to aggregation edge facilities that are in secondary or tertiary data center markets (i.e., outside of San Jose, CA, Ashburn VA, etc.) yet are still capable of supporting the power requirements of GPU workloads.

Like most issues with deployment of edge analytics services, the customer’s particular use case will dictate which provider can support the required latency for their applications.

The larger takeaway is that datacenter providers of all sizes and locations are offering tailored solutions that go beyond standard X86 rack configurations for general-purpose workloads. We see GPU-focused offerings being one of a growing variety of solutions that will offer specific processor and network configurations tailored to use cases such as data processing. We expect to see next-generation AI chips from companies like Brainchip, Tachyum, and more as well as new networking chips to power future edge data center offerings.

Jim Davis, Principal Analyst, Edge Research Group

Article Topics

AI/ML  |  datacenter  |  Digital Realty  |  edge datacenter  |  EdgeConneX  |  Equinix  |  GPU  |  Nvidia