The second days for GRID Days focused on two main aspects, the first getting deep with the nVidia GRID technology with everything from architecture to sizing methodologies and secondly our thoughts around building a community program, what is important to us in community programs and what should it contain.
The technical sessions on GRID were really valuable to me, the first session focused on the architecture of the GRID GPU and how it works in a vGPU context, the slides below show some of the layers of the architecture with the various engines on the card and the summary. I intend on writing my own blog post on this or at least seeing if I can post a link to the content for the readers of this blog, by understanding the architecture it will help you understand, design and support GRID in your VDI environment.
We were also told about how nVidia were doing some user experience testing internally with a program they call click to photon, effectively their goal is the speed of light and they aren't happy until they reach it. Click to photon testing allows them to measure how long the response is in ms between the user clicking a mouse on the client and the screen refresh from the the remote desktop happening on the local screen. As you can see in the results below with Horizon View 7, Blast Extreme and nVidia GRID offload they have been able to substantially reduce the latency to below 150ms, 150ms is an important figure to get below as it [UPDATE WHY IT IS IMPORTANT HERE]
I was amazed to see this glanuraity in the testing and also the ingenuity in the equipment used to completed the tests, they explained that the hardware was generally available equipment which cost no more than $600.
Another of the sessions that really stood out to me surrounded how to size and design your VDI desktops for graphical users, it started first by explaining how benchmarks were calculated and the use of the benchmarks but concluded that in a real life environment the benchmarks don't translate to a good user experience and that end user testing is the only solution to getting GPU sizing correct. The methodology focused around starting out by finding out what framebuffer was being used at max by the user on their current desktop and then using this figure to help select the profile before moving on to consider how many users could effectively share the card whilst still delivering the correct user experience.
I'm a really big fan of having sizing methodologies, it allows the industry to have a clear guideline as to how you should be designing and implementing technology, of course you are then able to use this and add your own methods and tools on top. I look forward to digging into the ways of doing this and blogging my results in the coming weeks.
There were a number of other great sessions during the day that I won't go into here, covering customer case studies from the customers that had joined us for the two days and collective thoughts on the industry, GPU adoption and community programs.
The evening was spent at a restaurant at the Levi Stadium where the conversation continued alongside lovely food and wine.
I have really enjoyed my few days in the US, I have been amazed by the power of the technology available from nVidia and inspired by the levels of knowledge of their team and my fellow attendees. I am now looking forward to going away and digging even deeper into their technology, it would also be cool to setup my own click to photon lab (I don't think I am the only person thinking that though, I suspect @drtritsch and the Smackdown team will be me though)
My doodle from day 2 can be seen below and I will upload a time lapse soon.