Over the last couple of days, I attended Citrix Synergy in Anaheim and E2EVC in Los Angeles. Six days packed with great information about cloud and remoting technology – and hanging out with the virtualization community folks. Here’s what I think were the most interesting things I learned.
The Synergy keynote was delivered by Mark Templeton and he opened his session with the announcement that Intel will be adding GPU cores to their CPUs, allowing for rich application services. Wow, what a start! Only four weeks after Ruben, Shawn and I published our whitepaper on GPU-accelerated graphics remoting, this technology gets so much love. Additional information about the Intel Crystalwell and GVT announcement was disclosed at the Synergy day-2 general session. Crystalwell is the codename for Intel’s Haswell CPUs with Iris Pro GT3e GPU cores and eDRAM on one single die. An example is the Intel Xeon E3, specifically designed for servers. Such a Crystalwell CPU can be combined with XenServer, creating the foundation for what Intel calls Graphics Virtualization Technology or Intel GVT for short. Pretty much like NVIDIA vGPU, Intel GVT is a GPU virtualization solution with mediated or brokered graphics pass-through. If you want to know more about this, download the 3D Graphics for Virtual Desktops Smackdown whitepaper here and read my related article from April 9.
The other Synergy keynote announcement that I found most exciting was Citrix Workspace Services (CWS) for Microsoft Azure. Generally speaking, CWS was designed for controlling creation, management and delivery of distributed virtual user workspaces across different cloud platforms. Combined with Microsoft Azure as the first cloud platform supported, this could be the future of XenApp and XenDesktop. Simon Bramfitt published a great analysis of Citrix Workspace Services. What I like most about CWS is that it uses a management interface that is built on top of Azure Pack for Windows Server. This seems to bring Citrix into a pretty similar position with respect to Microsoft like they were in the WinFrame times: Citrix embraces and extends a core Microsoft product using unique programming interfaces. If combined with Framehawk for HDX improvements in the future, this all may well create a cloudified workspace environment with superior user experience. Let’s hope that Citrix doesn’t drop the ball on this one…
Now here is my favorite announcement from E2EVC, but it’s somewhat related to the Intel GVT announcement from Synergy. Nick Pandher from AMD delivered a nice E2EVC session where he disclosed that they are working on a new technology that provides hypervisors and virtual machines access to shared and virtualized GPU cores. That alone is not too exciting, but AMD does this with an interesting twist. The big difference to NVIDIA’s vGPU is that AMD does not require GPU vendor-specific software components in the hypervisor or the parent partition (= Dom0). AMD’s intention is to do this all in hardware, making an individual GPU core look as if it were multiple cores. In other words, if I understood it right, they want to move the GPU resource broker to the hardware level. As a result, any hypervisor would see a pre-defined number of virtual GPUs that behave like physical AMD GPUs. This means that even Hyper-V would be able to take advantage of this technology without any architectural changes. As soon as such graphics cards are available in conjunction with an updated set of hypervisor-specific synthetic graphics drivers for VMs, this technology may bring the graphics remoting concept called API Intercept to a whole new level.
Three announcements from three vendors, but one general direction: Improving user experience when delivering remote Windows desktops and applications from the private and the public cloud. It will be interesting to see which of these products will resonate in the market.
This article first appeared on Benny's blog on May 11th, and is used with permission.