This is a summary of the questions arising from the interactive chat Q&A from a webinar presented by Jared Cowart from NVIDIA for the Citrix User Group in May 2016. We have checked and enhanced the answers but readers should be aware that as products evolve the answers may change and as such all information should be queried against the current product documentation. We aim to release these FAQs for the benefit of those who attended the webinars and also the wider community. Users with follow-on questions are encouraged to post on the NVIDIA GRID Forums where GRID support, product and engineering staff answer questions.
Q: Can we have a recording of this presentation/Demo?
A: A recording of the webinar is available here: https://www.mycugc.org/p/do/sd/sid=146
Q: Any suggestions on improving zooming and panning in AutoCAD Civil 3D 2016? We are seeing delays in remote locations from the VDI servers.
A: It's very hard to comment on a specific operation in a specific application. In general we need more information, for example what is the windows OS - many applications behave very differently on Win 7 (GDI/GDI+) vs Win 8.x and up (Microsoft changed OS to DirectX based stack). Panning and Zooming in general work best when the vGPU framebuffer is well matched to the parts being used so it is worth assessing the framebuffer usage as you may not be using a vGPU profile with sufficient support for your usage (see information on monitoring framebuffer: https://nvidia.custhelp.com/app/answers/detail/a_id/4108/). Users application specific queries are encouraged to discuss such issues on our forums: https://gridforums.nvidia.com/
Q: Are there performance differences between VMware and XenServer?
A: At NVIDIA we have not noticed any significant performance differentiator between the hypervisors with respect to the GPU features including vGPU. Customer hypervisor choice is usually dictated by management tool integration, other hypervisor feature and specification differences, as well as existing infrastructure. vGPU is available with VMware Horizon View on vSphere/ESXi and Citrix XenDesktop and XenApp with XenServer and vSphere/ESXi.
Q: Are these GPU cards available for Cisco UCS chassis blades?
A: Yes the Tesla M6 is designed for blades - Customers need to check their hypervisor vendors HCL (Hardware compatibility list) for information on how to do this is here: . For bare metal use of GRID GPUs, customers should check NVIDIA's certified server list, here: http://nvidia.custhelp.com/app/answers/detail/a_id/4116. Cisco have recently (April 2016) announced a reference architecture for M6 cards on UCS will be released soon: https://twitter.com/FrankCAnderson/status/717820448478613504
Q: Do you have the same CLI console for VMware instead of XenServer?
A: ESXi and XenServer both provide a CLI (command Line Interface) but these are different and the commands different. The documentation in the User Guide for the product is specific to each hypervisor and outlines the commands you can use on each.
Q: For company with low budget, is there any model of NVDIA card for servers other than GRID for sharing GPU?
Q: Quadro cards can be via GPU-passthrough under XenAppp on bare-metal, vSPhere/ESXi and XenServer; GRID Enterprise support is not available on Quadro products.
GRID 2.0 and up offers Enterprise support direct with NVIDIA. vGPU for XenApp offers some benefits for ease of management/monitoring above GPU-sharing via pass-through/vDGA. Also note that the number of Quadro cards in server chassis is typically limited to one.
Shared GPU for VDI (XenDesktop/View) is only available via vGPU and only on GRID cards. Very soon after the webinar NVIDIA announced a new GRID card, the M10, aimed at high-scalability to reduce TCO, you may wish to look at this new offering.
Q: Have you seen any benefits of using Windows 10 over Win 7 or 8 as far as HDX 3D Pro is concerned?
At the time of writing, Citrix does not support GPU acceleration or HDX 3D Pro with Windows 10 and so we cannot comment. Windows 10 in general benefits greatly from the availability of a GPU, the extent to which Citrix or VMware protocols leverage GPUs is a question for those vendors.
Q: How important is it to match driver version of host and guest?
A: It is essential to match drivers on the host and guest. This is the configuration tested and certified to work. GRID 2.0 and up customers have 24/7 Enterprise support when using a certified version. Mismatched drivers may well lead to problems and are also unsupported.
Q: How is user density affected for VDI? If you deploy GRID the card profiles seem to significantly restrict hardware density - therefore do you feel a vGPU is something all modern virtual desktops should have or is it only really for sessions that require it and most "Standard" users would continue to utilise standard HDX graphics acceleration features
A: Users with even moderate GPU usage usually find other resources such as CPU, bandwidth or IOPS are exhausted on the server before other resources. Users who do find the scalability limited by the number of users supported by a GPU card can add additional non-GPU enabled VMs to hosts, this can be a very good strategy for maximizing resource usage as users who do not need a GPU use applications and resources very differently. Users with high scalability needs can also use XenApp vGPU or GPU-sharing (via passthrough) where higher consolidation ratios can be achieved on GRID cards for lightweight applications.
NVIDIA have recognized the need for a high-scalability card and shortly after this webinar announced the new M10 product which doubles the number of VDI users that can be supported. It's a high density card supporting up to 64 vGPU users, allowing up to 128 vGPU users per dual card chassis. See: http://nvidianews.nvidia.com/news/nvidia-grid-delivers-100-graphics-accelerated-virtual-desktops-per-server
Q: Is GRID supported on XenApp 7.7 and/or 7.8?
A: Yes. GRID GPUs are supported for XenApp on GPU-passthough and vGPU, although you should always check the hypervisor version support for your particular server (http://nvidia.custhelp.com/app/answers/detail/a_id/4116). GRID cards can also be used for XenApp on bare-metal provided the server/GPU combination is on the NVIDIA certified server list: http://www.nvidia.com/object/grid-certified-servers.html. XenApp added support for vGPU back in XenApp 7.5 (announced here: https://www.citrix.com/blogs/2014/11/25/citrix-supports-nvidia-grid-vgpu-for-xenapp/), on the GRID 1.0 at the time it was not possible to assign a whole physical GPU as a vGPU so the advice was to use GPU-sharing. With vGPU profiles now available representing a whole GPU, the benefits of management and monitoring make vGPU a good choice for XenApp.
Q: I need to virtualize about 50 desktops for use by remote users connecting with a satellite connection. It has a 700ms latency, which solution will support a good user experience with that kind of latency?
A GPU solution will provide hardware acceleration to the application and the server to raise the image quality and framerate. How well the network can support those frames will always be an issue. For interactive work 700ms will not provide a great interactive experience (you cannot beat the speed of light).
Raw latency alone does not affect interactive or user experience. The jitter (variation in the latency) will be a big factor too.
Depending on your bandwidth and packet loss you might either find Citrix's Framehawk or Thinwire compatibility (Thinwire+ released in XD7.6 FP3) provide a better experience to default protocols at high latency. For some Citrix protocols tuning the TCP window at high latency can provide effective mitigation, see: CTX125027 - How to Optimize HDX Bandwidth Over High Latency Connections.
Users of VMware may find benefits from the GPU-accelerated H.264 compression provided by the hardware encode offered by Blast Extreme for both Windows and Linux. Hardware encode is currently only available for the Linux VDA for Citrix.
A major factor will be what applications your users are using and the level of interactivity they need. Some graphical uses such as video streaming are very tolerant to high latencies with buffering available and lower user interaction (see this Citrix blog for an example: https://www.citrix.com/blogs/2014/05/23/is-running-xendesktop-remotely-on-a-server-8000km-away-with-700ms-latency-on-an-airplane-with-an-oculus-rift-virtual-reality-application-insane/)
If users demand a highly interactive experience 700ms is unlikely to provide it and you might want to explore using a data center closer to the users, there are many service providers and cloud offerings available from both Citrix and VMware and their partners for VDI.
Q: Is 3D image capture supported over XenApp? We're running VMware ESXi.
If the 3D capture hardware is supported over USB into the XenApp session then yes. This is a question for the 3D Scanning solution provider and Citrix to address.
Q: Is it choppy on your side or is it spinning clean?
A: During the webinar the model was spinning cleanly on Jared machine - unfortunately GTM is limited and provided a bottleneck so it looked jerky to remote viewers. This is always a problem with webinars and even youTube videos which can compress recorded videos.
Even recording demo video in software can be misleading, so Citrix has released a blog on how to make hardware captured videos, here: https://www.citrix.com/blogs/2014/07/30/introduction-to-video-capture-hardware-for-hdx-product-demonstrations/
Q: Please describe what's required on endpoints - both Windows and zero thin-clients such as Dell Xenith.
The end-point requirements will be dictated by the virtualization stack and protocols (e.g. HDX/ICA) and the capability of the end-client software (e.g. Citrix receiver). Using an NVIDIA GRID GPU on the server adds no constraints on the client but for a good user experience the client should be able to keep up with the server's ability to supply high-quality visuals and frames.
For Citrix when using NVIDIA GRID we'd recommend you look for an HDX 3D Pro certified end-client with hardware decode. If you have additional needs such as smartcard, multi-monitor or USB you would probably want to look at the clients HDX Premium certified status as the HDX 3D Pro level covers graphical capabilities only. There is plenty of information on Citrix own site, here: https://www.citrix.com/blogs/2016/01/04/finding-vdi-thin-clients-just-got-easier/
VMware have an equivalent certification program, here: https://www.vmware.com/resources/compatibility/search.php?deviceCategory=vdm
Q: We are getting ready to replace our VMware ESXi Hypervisors running XenDesktop 7.6 and are thinking about GRID and GPU....am I correct that customers who only use Published Apps would benefit from more GPU and GRID?
Any user delivering applications which utilize hardware acceleration will benefit. Applications as ubiquitous as the browser, Adobe Acrobat Reader and Microsoft Office benefit in addition to other more obvious applications. Acceleration is available regardless of whether the user accesses a published desktop or a published application.
Q: We are looking to utilize GRID with XenDesktop 7.6 (Win8 and Win10) for our power users to enhance the overall experience. They do not utilize high-performance 3D apps, only Office and browser sites. Would HDX 3d Pro still be the recommended VDA to use? Or would the Standard VDA be recommended?
The applications can leverage hardware GPU sharing via vGPU on both VDAs, the difference is that the 3D Pro VDA accesses the framebuffer of the GPU directly. The standard VDA is tuned differently for text and visual quality and bandwidth for office applications. Since office applications use the framebuffer less heavily the standard-VDA is likely to be a good choice. We'd advise you use one of the standard templates released in XD7.6 FP3 for configuring your HDX environment and VDA in the first instance (see https://www.citrix.com/blogs/2015/10/28/simplify-hdx-policy-administration-to-amplify-the-user-experience/).
Q: What applications would benefit from Virtual PC (vPC)?
vPC is intended to provide GPU support for business and office users, including those with multi-monitor demands, see M60-1B profiles etc). Applications such as browsers, unified communication software (e.g. Cisco Jabber), video/YouTube and office (Powerpoint etc) should benefit.
Q: What is best vGPU solution to just increase graphics experience for VDI with windows 7 and windows 10?
It will depend on your applications, if they are regular office applications rather than 3D/CAD, Virtual PC (vPC) is likely to be the best solution for you. It is likely that the newly announced M10 card used with VDI would be a good solution for you.
Q: What is typically user per server ratio in VDI environment? How many VDI instances can you run on single Physical machines under Autodesk Revit intensive workloads with single K1 GRID card.
The K1 cards was primarily aimed at office users (for improving windows experience) and as such for intensive 3D workloads and large models the M60 (successor to K2) would be a much better choice.
The application usage and workflows the users actually use for a product like Revit varies greatly and will dictate consolidation ratios. It's advisable to test with real users as real-life CAD workers often have bursty GPU usage and benefit from NVIDIA's time-sliced methodology.
The GRID performance team do produce a number of application guides with general guidelines for some applications that can be found on the main GRID website under the tabs Resources->Application Guides (http://www.nvidia.com/object/grid-enterprise-resources.html).
Revit is one application that has been assessed (alibi on VMware - the advice should be relevant to a Citrix stack too) - see here: https://www.nvidia.co.uk/content/grid/resources/vGPU_App%20Guide_Revit_1.1_FINAL.pdf
Q: What Linux distro are you showing?
A: During the webinar Jared showed XenDesktop's Linux VDA with RHEL 7. Both VMware and Citrix offer Linux VDAs that can use GRID GPUs on a range of distros. Which distributions are supported and whether both vGPU and GPU-passthrough are supported differs between VMware and Citrix and customers should check on the current level of support which is evolving rapidly for both vendors.
Q: Where do you acquire the vGPU manager ISO?
A: For GRID 1.0 (K1/K2) products you can download the GPU drivers and software from: http://www.nvidia.com. For GRID 2.0 and up, customers should obtain vGPU drivers and software from the NVIDIA licensing portal. Advice on how to do this for customers and also pre-evaluations is available, here: http://nvidia.custhelp.com/app/answers/detail/a_id/4120/
Q: Which management tool is being displayed here to show the GPU usage on the GRID card?
A: Jared was using nvidia-smi as detailed in http://nvidia.custhelp.com/app/answers/detail/a_id/3751/ and http://nvidia.custhelp.com/app/answers/detail/a_id/4137/kw/nvidia-smi. Jared was also using a unsupported community tool (written and supplied as is by one of our architects) GPUProfiler, an overview of this tool can be viewed here: http://on-demand.gputechconf.com/gtc/2016/video/S6504.html. The source code and tool using the NVAPI to call the functionality of nvidia-smi is provided here: https://github.com/JeremyMain/GPUProfiler/releases.
Q: If I use XenApp to publish full desktop is that licensed as applications (vApps) or as Virtual PC?
A: For all usage from an RDSH solutions such as XenApp the license required is vApp. As long as the profile associated to the VM is either A profile, or passthrough.
In the event that you choose to deploy Pro-Vis applications via RDSH / XenApp and you require Quadro functionality you would then require a vWS license and a Q profile or passthrough associated with the VM.
Q: Can we get the CLI from NVIDIA for VMware?
The command line interface (CLI) is a built in feature for each hypervisor, automatically installed with each hypervisor and provided by Citrix/VMware/etc. The NVIDIA vGPU product comes with full documentation on the CLI commands available to manage vGPU for each CLI and hypervisor.
Q: What are the differences in target use case between Kepler and Tesla?
Kepler is a GPU-generation (Family). The latest (at time of writing) GRID GPUs M60/M6 are Maxwell generation cards. Maxwell is the next-generation to Kepler. Tesla was the previous generation to Kepler.
Tesla is however also an NVIDIA product line for HPC. The GRID 2.0 M60 and M6 are based on _Tesla_ products that can be used for compute and but be configured to "graphics" mode for use with the GRID software and incorporate Quadro (graphics product line).
If you need a GPU for compute you should look at the HPC (High performance compute) products in the Tesla product family such as the K80 (note a Kepler generation card from the Tesla product line).
If your question is about the differences between Kepler generation GRID cards and Maxwell Generation cards, you should look at the GPU specifications. Maxwell GRID cards can support blades (M6) and are generally a high-specification to the GRID 1.0 Kepler cards (K1/K2) - more cores, more H.264 encode streams etc.
Q: For web browser acceleration would I require Virtual Workstation?
No, browsers can be very GPU intensive and will benefit from the GPU-acceleration supported by vPC; if you are likely to be using office apps then virtual PC (vPC) is a good solution and more cost effective. There are profiles available, such as the M60-1B ("B" for business) aimed at user like yourself who may be using lighter weight graphical applications but have the need for features such as multi-monitor, within vPC licensing.
Q: I am not using any virtualization at this point. We are going to completely revamp our storage system and want to properly plan for migrating to a VDI environment. We currently have NetApp 9 TB and 2 ESXi VWware boxes with about 8 Servers. No desktops are virtualized.
A: There are lots of places where you can find information or assistance for planning a VDI strategy including GPUs. I would suggest you explore:
· The case studies on others' usage: https://virtuallyvisual.wordpress.com/useful-links/remote-graphics-case-studies/
· The options to work with a local NVIDIA Preferred Partner experienced in such deployments: http://www.nvidia.com/object/partner-locator.html
· Attending webinars, user group events or conferences for your VDI supplier of choice (Citrix/VMware)
· Asking any questions you may have on our user forums: https://gridforums.nvidia.com
Q: Is there any benefit to assigning more than 1 vGPU profile to a VM?
A: This is not currently supported by hypervisor vendors. One vGPU profile per VM should be assigned. If you need more GPU resource for a VM we would advise allocating a larger vGPU profile. Only a few professional applications can utilize two GPUs well and are rarely used with VDI and as such very few applications would benefit. If this changes and there is customer demand NVIDIA and our hypervisor partners would of course revisit this.