Graphics Virtualization at NVIDIA GTC 2016
Last week, I’ve attended NVIDIA’s GPU Technology Conference (GTC) in San Jose in Silicon Valley. This year’s focus of GTC was virtual reality, artificial intelligence and autonomous cars. NVIDIA’s CEO Jen-Hsun Huang made a number of spectacular announcements around these topics and showed impressive demos, including Steve Wozniak in a real-time VR simulation of a mission to Mars. Despite the fact that graphics virtualization was not mentioned in the keynote, it was a topic considered as special emphasis area and was covered in more than 40 sessions and talks within a separate track. There was a great line-up of speakers from various vendors, partners and customers. But there were also independent experts and industry analysts who shared their thoughts about GPU-accelerated remoting in general and GRID v2 in particular.
GTC is a great place to be when you want to listen to all relevant vendors like Microsoft, Citrix, VMware, IBM/Softlayer, Amazon and Frame presenting all their products that are playing nicely with NVIDIA GRID cards. Ultimately, attendees are able to see all these vendors’ product managers in one room, something that doesn’t happen too often. One of the sessions I personally enjoyed a lot was “Graphics Virtualization: Leveraging Microsoft’s New GPU Virtualization Technologies” presented by Chris Huybregts from the Microsoft Hyper-V team and Jeroen van Eesteren from the RDS team. Among the things they highlighted were the Windows Server 2016 RDP protocol improvements and the upcoming GPU-accelerated N-series Azure VMs. Derek Thorslund and Stephen Vilke from Citrix presented a session on next-gen HDX and Framehawk concepts optimized for WAN connections. In another session Pat Lee from VMware introduced performance details of their new Blast Extreme protocol. All these remoting protocols have in common that they have improved capabilities allowing them to take advantage of GPUs.
But there was more. We learned how NVIDIA customers like Bell Helicopter, National Aeronautics and Space Administration or Lockheed Martin are using GRID-accelerated remoting in their enterprise environments. Very impressive! We were able to see how software vendors like Esri, Dassault Systemes, Adobe, SOLIDWORKS and ANSYS can benefit from the significant performance improvements provided by the new NVIDIA GRID cards. Industry expert Thomas Poppelgaard introduced his favorite tools for monitoring NVIDIA GRID, allowing him to measure the impact of the GRID v2 performance improvements. And last but not least, Ruben Spruijt and I presented “Benchmarking and Scalability in the Virtual Workspace” with brand new test results. Ruben and I had a lot of fun, feel free to take a look at the video recording of our session. But what I enjoyed most was hanging out with other speakers, customers and the folks from Microsoft, Citrix, VMware and Nvidia. I had some great “side conversations” you can only have when meeting people face-to-face.
A highlight that most people may have missed is that NVIDIA made a public announcement about a change in the GRID v2 licensing model at the first day of GTC. In essence, NVIDIA made the licensing model simpler and they also adapted prices. They introduced an annual subscription license in addition to the existing perpetual license schema. Now there is an entry-level license model for shared multiuser environments (called Virtual Appliances), a mid-range license model for virtual desktops, and a high-end license model for virtual workstations supporting 4k screen resolution. An annual subscription costs $10, $50 and $250 while a perpetual license is $20, $100 or $450 per device. There’s a great article by Jits Langedijk that explains the licensing model in more detail.