At Mobile World Congress 2019 in Los Angeles, California, NVIDIA made a number of announcements focused on enabling the 5G mobile edge. Early adopter customers of the EGX platforms include Walmart, BMW, Procter & Gamble, Samsung, NTT East, as well as the cities of San Francisco and Las Vegas.
NVIDIA EGX Platform
You may recall the announcement from our piece NVIDIA EGX Platform for Edge AI Inference Launched. Essentially, this is NVIDIA’s vision for lighter weight GPU enabled servers to IoT Jetson Nano devices that can sit in the edge. Think of them as a counterpart to the DGX and AGX lines for the data center and the robotics industries.
A big part of that includes EGX boxes that are designed to sit in telco racks at the edge as our cover picture shows. One may notice that this is perhaps the strangest airflow design with the DIMMs blocking airflow to the GPUs. We think this is a render as many of the partner edge servers we have seen for these applications, like the Wiwynn EP100 Short Depth 3U5N, use more standard airflow.
With today’s announcement there are a few big bullets on the ecosystem side:
- Microsoft plans to work to more tightly integrate NVIDIA’s accelerators into its edge designs to expand the reach of Azure services
- Red Hat is enabling the edge services for 5G operators via OpenShift. Red Hat is an important enterprise Linux distribution but also has a number of orchestration tools in its stack that can help manage far-flung nodes.
- Ericsson announced that it is working with NVIDIA and Mellanox SmartNICs to enable virtualized 5G radio access networks (RANs). These virtualized RANs allow 5G operators to use more generic compute resources to enable differentiated service delivery in the future 5G networks.
Final Words
Overall, these are good things, but we did not get a new major hardware announcement. We are still talking about the NVIDIA T4. You can read STH’s NVIDIA Tesla T4 AI Inferencing GPU Benchmarks and Review as well as our Analysis of Our NVIDIA Tesla T4 Review for more on that platform.
