NVIDIA GRID Days 2016 Recap, Key Takeaways, Technical Deep Dive and Lessons Learned!
Last week in Santa Clara, I was honored by NVIDIA to be one of thirteen professionals to join NVIDIA enterprise graphics teams for the first ever GRID Days event! GRID Days was a fun packed two day event networking with industry professionals, and getting an inside look into the NVIDIA enterprise graphics teams, namely GRID, Quadro, and Tesla. During this event we had exclusive access to the testing and development teams, product management and marketing, executive sponsors, along with management and support teams. Wow! Much of the roadmap and content required that all professionals sign a Non-Disclosure Agreement (NDA), so I will be careful only to share agenda topics and details that are publicly available at this point. Our ever so patient NVIDIA hosts for the week included: Amanda Saunders, Rachel Berry, Luke Wignall, and the rest of the amazing NVIDIA GRID team!
In order from left to right: Amanda Saunders, Thomas Poppelgaard, Sean Massey, Barry Coombs, Benny Tritsch, Fred Devoir, Emily Apsey, Dane Young, Ruben Spruijt, Jits Langedijk, Magnar Johnsen, Tobias Kreidl, Kurt Miller, Alex Cooper, and Rachel Berry. As you can see from the image above, we have had an excellent representation from the global end user computing and virtualization community, not just in North America, but Europe, Middle East, and Asia (EMEA) as well!
Day 1 – Introductions and Overview!
The first day of the event started with a bang as we unexpectedly ran into a dear friend of the virtualization community, Mark Templeton, former CEO and truly the heart of Citrix for many years.
Mark then proceeded to allow me to drive his Tesla from the hotel to NVIDIA HQ through Santa Clara (Thomas riding shotgun!)
I’m pretty sure Jits, Magnar and Sean in the back were terrified for their lives as I kicked on ‘Insane mode’!
Once we arrived at the NVIDIA office, MarkT shared with us some pretty cool stuff he’s been working on, then we got started.
Introductions were led by Rachel as we enjoyed lunch, along with a little vision casting about the future of NVIDIA as it relates to community and influencer programs. More to come on that later!
Scott Fitzpatrick provided a brief history and overview of NVIDIA graphics in the enterprise and what Quadro could do to address modern business challenges.
We were then guided through an exclusive NVIDIA Demo Room tour, which was by far the best and most complete showcase of NVIDIA consumer and professional products! Thanks Robert Perry!
NVIDIA Visual Computing Appliance up close and personal!
Overview of NVIDIA products in the consumer gaming industry:
A beautiful showcase of NVIDIA in the automotive industry:
Simulated demo of the self-driving car:
After coming down from a graphics goodness high in the demo room, freshly minted GRID VP and GM Jim McHugh shared a little bit of his perspective and vision for NVIDIA GRID in the Enterprise. This session was particularly useful as it was the first time I had an opportunity to speak openly about the past, current, and future challenges in NVIDIA GRID adoption! I suspect many of my industry peers were also very honored by Jim’s openness in the road table discussion. His personality and approach to this discussion paved the way for more transparent discussions with the NVIDIA GRID team. Leaving that discussion, I am very hopeful for NVIDIA’s future direction and strategy in the enterprise graphics market.
This discussion was followed by a private NDA demo in NVIDIA’s consumer graphics group (GeForce) to see what they’re cooking up next!
Hint taken from www.geforce.com website banner:
After a very full day, the NVIDIA team hosted us at a very nice dinner at Campo Di Bocce in Los Gatos. Bocce ball fun was had by many!
Day 2 – Technical Deep Dives!
Day two was most definitely the more technical day, and after drinking through a firehose all day, we were all spent! The first session was led by NVIDIA’s fearless CTO for GPU Virtualization and Brit Andy Currid, going under the hood of GRID vGPU! Andy’s session was super helpful as he walked us all through NVIDIA’s silicon and software architecture and why certain decisions were made.
If you’ve never seen or heard Andy present before, I’d highly recommend you watch the recording of his vGPU presentation from GTC 2014. Andy is the man!
I won’t share the entire deck that Andy presented at GRID Days, but here are a couple of the key topics from his presentation:
- GRID vGPU Architecture
- Memory Management
GRID vGPU Architecture 101:
Scheduling and Timesharing (Sequential vs. Parallel):
Overview of SR-IOV:
Comparison of SR-IOV to GRID vGPU (Why NVIDIA chose the vGPU architecture they did):
Obviously there are a ton more slides than I shared here, but I hope you get a sense for the overall content. Looking forward to finding a similar presentation from Andy in the near future that I can repost and share as this presentation was tremendously valuable.
Huge thanks to Andy and the NVIDIA GRID team for this session. I’ve been interested in GRID since the pre/early days of VGX, and this was the first time I heard many of these advanced concepts laid out this way! Thanks Andy!
In the next highly technical session, we got to hear from Luke Wignall (Performance Engineering Director) and Erik Bohnhorst (Performance Engineering Architect) on testing and benchmarking methodology for graphics in a virtualized environment! These guys rock!
Luke and Erik started the discussion by resetting expectations for traditional testing methodologies for VDI, and why they do things the way they do:
Overview on Timeslicing:
Why Benchmarks Need Not Apply:
And finally, my favorite part of their session! Huge thanks to Jason Lee and the NVIDIA team for seriously blowing our minds with…
…Reusing latency testing methodologies from the GeForce team within NVIDIA. Click to Photon is the scientific name for a series of tests and methodologies that assess the impact of latency, either by the device driver, virtualization environment, or other environmental conditions! Click to Photon is scientifically measured using an oscilloscope.
Using an automated method to click and measure response time from click to color change, the NVIDIA team is going all the way when measuring the impact of latency at all levels. By far the coolest thing I saw as it relates to scientific remote protocol latency testing methodologies! Check it out…
How to scientifically measure protocol, encode, decode and responsiveness latency (on a LAN or WAN) using the Click to Photon method:
Obviously we were all incredibly intrigued how this worked and could be used to scientifically measure all kinds of latency tests!
The important thing to note is that this isn’t necessarily used to measure LAN or WAN latency, there’s plenty of other tools that can be used to show RTT, much like a simple Ping test would do. Click to Photon is intended to measure the complete round trip time from the second a mouse is clicked, or keyboard button is pressed, until a change is detected on the screen. Obviously this means it’s capturing a lot more in the process.
According to the NVIDIA team, distinguishable and acceptable human interactivity is usually around 150ms for the Click to Photon test. This is something that NVIDIA has proven with their Cloud Gaming service popularized with the Shield products. To learn more, read the section about Game Consoles Will Be Obsolete in 2 to 3 Years in the following blog post. A 100% local PC-based experience is typically between 50-100ms for the Click to Photon test. This is fairly typical for gamers using GeForce GPUs and doing high end PC Gaming. Tolerances for cloud gaming services, or remote graphics will obviously be slightly higher, even for a LAN based experience. The question is, what’s tolerable and acceptable to the eye and abilities of the human brain to process in terms of responsiveness.
You can see how this would be different than refresh rate (function of display capabilities), or frames per second (function of GPU horsepower), as many GPUs and displays can push far more than that. A 4K 120Hz display for example has the ability to refresh frames every 8ms (1000ms / 120). Ultra-responsiveness in a gaming world doesn’t mean your responses keep up with refresh rates, simply that it appears that way to the brain. Virtual Reality headsets for example must push these higher refresh rates, otherwise it can create nausea (thanks for that tip Rachel!).
Now that we’ve seen a little bit of the scientific method, let’s evaluate how NVIDIA used it in recent testing of the newly announced Horizon Blast Extreme protocol that will leverage their hardware encoder (NVENC):
I really like this testing methodology, especially when it’s used to evaluate multiple options for remoting from a single vendor. Keeps the discussion unbiased as each major vendor implements the render, encode, encapsulate, send, decode, render process slightly differently. In the visual above, you can see their testing results showed a 50% improvement factor in responsiveness in nearly every test when comparing Horizon 7 with PCoIP to Horizon 7 with Blast Extreme (GPU using NVENC). Under heavy load, letting the GPU do the work improved the responsiveness by 58%! We can also see that the entire process using Blast Extreme with GPU (NVENC) is only 100% slower than the local laptop. Not bad considering everything that’s involved in that round trip! Seriously impressive results, and even better knowing the testing methodology behind these tests is highly scientific.
Here’s an example of what the oscilloscope looks like during the first test represented (I prefer looking at pretty bar graphs personally):
Now that you’ve seen this testing methodology and results, if you’re interested in learning more, feel free to reach out to Jason Lee! Magnar has already started to experiment with simulation of Click to Photon using software, check it out!
Fred was able to share a couple details about his amazing NVIDIA GRID powered GPU environment:
Once again, Fred has proven to all of us that his GPUs are bigger than ours!
The other sessions on day two were largely roadmap and feedback sessions, with lots of Q&A between the NVIDIA team and attendees. We talked a bit about what’s missing in the current offerings, marketing and support resources (website, forums, etc.), barriers to adoption in the VDI mass market, and support and documentation. Rachel led a session focused around the NVIDIA community and ways to gain more collaboration and interaction with the practitioners in the industry! Stay tuned for future announcements about what the NVIDIA team has in store!
As things wound down for the day, we had one last surprise in order…Gandalf the Geek, GPU Wizard!
Gandalf the Geek meets Gandalf the Grey (Tobias Kreidl) for the first time at GRID Days:
Tobias and I met for the first time this week, but were instant friends as he became a freshly minted Citrix Technology Professional (CTP) earlier this year and quickly became known as Elder Gandalf. Our mutual respect is largely due to being on polar age spectrums in the CTP program (oldest and youngest), and I was always taught to respect Wizards and Elders.
After a bit of a laugh and guest appearance by Gandalf, Jim McHugh came back to wrap things up and collect feedback from the two days we shared together. At one point he asked me to take off the Gandalf hat, I guess it was a bit distracting!
The night concluded and we all left learning a little more about our industry, each other, and NVIDIA GRID.
I am really appreciative for the opportunity to have been part of this journey with the NVIDIA GRID team, and look forward to future opportunities to deep dive on GRID, Quadro, and Tesla! Really exciting stuff happening in the industry as a whole and I can’t wait to see what it means for End User Computing and Virtualization. I’m particularly excited about the work NVIDIA is doing with the Support, Update and Maintenance Subscription (SUMS) program to drive further value from dedicated support resources. This has been much needed to the NVIDIA enterprise ecosystem for quite some time! It takes a lot of work to bring together support on a lot of different levels, including ESRI, Autodesk, Dassault, PTC, and many others. Overall, this should be a very good year for adoption.
If this blog post was useful to you, and you are intrigued, I’d highly recommend that you register and attend the GPU Technology Conference (GTC) North America in just one short month!
This year’s GTC will be at San Jose McEnery Convention Center on April 4-7, 2016, and is one of my favorite industry conferences!
Jen-Hsun Huang (CEO of NVIDIA) is one of the most inspiring people that you will ever see present, he’s truly unique in our industry…
If you’re not able to attend GTC live, be sure to put a reminder in your calendar to watch the Keynotes live stream.
All the best GPU related announcements come out during the keynotes, and it’s an excellent opportunity to broaden knowledge about consumer and professional graphics, and even the autonomous car!
If you are in EMEA and would like to attend GTC, I’d highly recommend checking out the GPU Technology Conference coming to Amsterdam September 28-29, 2016! Check it out!
Once again, a huge thank you to the NVIDIA team for hosting the group this week. I know it’s a lot of work to put on an event like this, and we’re very grateful for everything you do! Credit goes to the NVIDIA team and my fellow peers for the awesome collection of pictures captured in this post. If you want to read more about this week at GRID Days, check out the following:
Barry Coombs – NVIDIA GRID Days, Day 1
Barry Coombs – NVIDIA GRID Days, Day 2
Sean Massey – It’s Time To Reconsider My Thoughts on GPUs in VDI…
Jits Langedijk – NVIDIA GRID Days- My thoughts and what I have Learned
Looking forward to seeing you soon!