Citrix Synergy 2013 is taking place in Los Angeles this week. One of the most talked-about demos from the opening keynote was from NVIDIA where they showed GPU-powered Windows desktop applications delivered via HDX, including live photo editing, video editing, and 3D mapping. We've written quite a bit about NVIDIA in the past year, including reviewing the announcement of their GRID K1 and K2 GPU server cards. (I also mentioned that the release of these cards signifies a major breakthrough in the general usability of VDI.)
Gabe was able to spend some time with NVIDIA's Will Wade at Synergy and recorded some interesting conversations. Key takeaways include:
- Gabe loved the demo. "Holy S***"
- NVIDIA's been talking about GRID and their VDI and RDSH capabilities for more than a year, so they're glad to finally bring it to market now in XenDesktop.
- There are two options: dedicated GPU pass-through and virtual GPU.
- Dedicated GPU pass-through means each VM gets one GPU. (That VM can be a desktop VM or a server VM running RDSH which can use the physical GPU for multiple session.)
- GPU virtualization means that up to 8 users can share a single physical GPU.
- NVIDIA has GRID server cards in two models: The K1 and K2. They are conceptually the same, though the K1 has four lower power GPUs and the K2 has two high power GPUs.
- These NVIDIA GPU virtualization products have the regular every day NVIDIA drivers in the guest VM. So there's nothing funky going on. If your app works with NVIDIA, it works with GRID and HDX.
- There is no more software licensing component. (Yay!) Just buy the cards from NVIDIA (well, from your server vendor) and you never have to pay them again.
- Sample pricing (from the add-in options for a D720 from Dell.com in the US) is $2289 for a K1 and $4182 for a K2.
Here's a quick video of Gabe getting an overview. (Video link opens a new page because I don't know how to embed more than one video per post.)
Gabe and Will also discussed graphics and GPU acceleration for XenApp. Gabe opened the conversation with, "We've been talking about this for years. We've seen this Sea-Doo demo a hundred times. So why are you showing that here? What's actually new?"
Will answered by showing a side-by-side of the old Sea-Doo demo with what they can do today. Basically the old version looked laughably cartoonish. The GPU-powered demo had full 3D shading, lighting, shadows, etc. All of this was powered from XenApp. If you want to use a GPU or GRID with XenApp, that's no problem. The GPU sharing works the same, with most people choosing to dedicate a GPU to the XenApp VM which in turn shares it as needed with each user session.
Here's the conversation between Gabe and Will as they discuss GPUs in XenApp:
Finally, Gabe and Will move to the white board as will shows how he explains just what the heck is going on with XenApp, XenDesktop, XenDesktop App Edition, virtual GPUs, GPU pass-through, hypervisors, etc.
In terms of support, they can do dedicated GPU pass-through (one GPU per VM) on both VMware vSphere and Citrix XenServer. At this time the virtual GPU "slicing and sharing" is only available on XenServer, though Will reminded us that "this is Synergy," and that "as soon as those guys announce, we'll talk about it." So we can assume that GPU virtualization is coming to vSphere, and possibly to Hyper-V.
So what do you think? Are these as cool as we think they are? Do you agree that adding NVIDIA GRID cards to Citrix-based VDI servers should be a requirement for everyone?