Nvidia OVX (SuperPod): New Multi-GPU System Counts Digital Twins

NVIDIA omniverse It is a collaboration platform on which the digital world can be built and connected. nvidia speaks”digital twinReality. Their calculations place great demands on the hardware used. Along with OVX, Nvidia presents its solution based on GPU.

What is Nvidia Omniverse?

With Omniverse, users can work with other users in selected 3D software apps in real time, develop interactive AI avatars or create digital twins of reality – but for what? An application example for digital twins is the simulation of changes or processes that cannot be so easily mapped to reality.

Simulation in a digital environment

For example, car manufacturers can set up and run entire factories in Omniverse and test the effects of small changes in a reproducible manner under similar conditions. For example, the behavior of autonomous robots in the omniverse can be evaluated risk-free with the Isaac SIM replicator. Ericsson, on the other hand, is already building entire cities digitally into the Omniverse in order to optimize 5G expansion in relation to the spread of radio waves. And in Germany DB Netze is rebuilding the group’s railway network as part of the “Digital Rail Germany” project to optimize the operation of the actual network and be able to better respond to disruptions. And Amazon uses Omniverse to train software for robots that pack products into packages.

The platform uses Universal Scene Description (USD), originally developed by Pixar as “HTML for 3D Rendering”, to communicate between professional applications from 3D environments including Adobe, Autodesk, Blender, Graphisoft, Mathworks or other Unreal to maintain compatibility. device.

Ubiquitous needs computing power

The more complex the digital twin, the more intensive the computation and computation of the simulations going on. With Nvidia OVX Servers, the group now wants to provide Omniverse customers with the right hardware solutions. The approach to pure AI workloads is similar to that of the Nvidia DGX, only in this case it’s not only about the AI ​​but also the workloads it provides.

Nvidia OVX Server for Omniverse

Nvidia OVX Server for Omniverse is based on eight Nvidia A40 graphics cards, the server version of the Nvidia RTX A6000 with GA102 full expansion and 48 GB GDDR6 (ECC), connected via three ConnectX 6 switches with 200 GBit/s Huh. , two Intel Xeon Platinum 8362 OVX servers with 32 cores each, 1 TB RAM and 16 TB storage space are the key data points. The Intel Xeon is in for quite a surprise after the AMD Threadripper has been repeatedly used in the Nvidia DGX.

OVX Superpod = Up to 32 OVX Servers

Customers for whom eight A40s aren’t enough for their Onmniverse project can have 8, 16, 24, or even 32 servers combined. Nvidia calls the biggest expansion phase “SuperPods” and stresses that SuperPods can still be added. Switch Spectrum-3 with 400 Gbit/s (Data Sheet, PDF,

The Nvidia OVX Server and OVX SuperPod are expected to be available later this year through partners Inspur, Lenovo and Supermicro. Nvidia did not name the prices. Nvidia now also rents DGX systems.

Access to the Omniverse via GeForce Now

Nvidia OVX brings the computing power needed for the Omniverse to a new level for the user, but at the same time Omniverse is opening up to users without certified hardware in the computer at GTC 2022. In the future, users will be able to access selected Omniverse functions from almost any end device via Omniverse Cloud, including 3D applications for collaborative editing of projects or AI avatar fields.

The applications used are run on cloud computers, that is, on computers from the gaming streaming platform GeForce Now. nvidia takes over From now on Applications are accepted for the early access phase.

ComputerBase has received advance information about this material from Nvidia under Nvidia. The only requirement was the earliest possible publication date. The manufacturer exercised no influence over the report, and was under no obligation to publish it.