Blast extreme does indeed work with multiple monitors even when an NVIDIA GPU is present, however VMware currently only leverage NVENC on the GPU with one monitor, when >1 monitor the applications will still access the GPU but rather than NVENC CPU encode will be used.
If you’re not tied into VMware, then … Citrix have just released their version of the H.264 Hardware Encode, and it’s VERY good and does support multiple monitors. Not only that, but it comes with new Adaptive Display v2. This monitors the screen content and only uses the H.264 Encode for video type content, and uses still image compression and bitmap caching for other areas. This retains the quality lost by a full screen H.264 encode for text etc
Here are some links to it if you’re interested:
Scroll down to the bottom of the page to where it says "Use video codec for compression"
Thanks for the info! We were chosen by this potential customer to be one of a few providers to show them a demo. This 30 day demo won’t have GRID, but it will allow them to experience a VMware Horizon VDI experience. If after this demo we are chosen to be their hosted VDI provider we will need to really get down to brass tacks with GRID. They are heavy users of 3D accelerated applications and will require hundreds of virtual desktops powered by GRID.
Do you have any white papers that break down the performance impact in this scenario? How does this affect density of guests on a single GRID node (server) We’ll be using a hyperconverged platform where each node will already be hosting a VM ofr the hyperconverged infrastructure, and we need a VERY solid idea of the actual VDI footprint we can actually have on each node. Every user in this environment uses multiple monitors. Either 2 or 4
You mention above that you are one of a few providers to show the client a demo. However, with the information you have detailed within this thread, I have a few concerns, and I’d like to help you address these so you stand a better chance of surpassing the clients expectations and offering more performance than the other evaluations.
Have you considered your options when one of the other providers gives the client a GRID enhanced evaluation to assess rather than a standard (non GRID) one? As your client is a heavy user of 3D accelerated applications, I’m sure they would want to evaluate a platform that can deliver this performance (I know I would). If you do not have a platform that they can evaluate, I’d suggest that you direct them towards the Try GRID offering from NVIDIA. This runs on Horizon and I believe this is now running on Maxwell generation GPUs. You can access it here: http://www.nvidia.com/object/vmware-trygrid.html
One of the important things to note, is that you are not designing to deliver VDI. You are designing for Workstation replacement, and they are not the same. You design for them in the opposite way. You design for performance and user experience first and foremost, everything else comes a distant second. Once the client is happy with the performance (typically, you’ll out perform their local workstation, or at the very least, match its performance) you then scale the platform without impacting the performance or experience. I mention this because you plan to use a Hyper-Converged platform. Hyper-Converged platforms are not conducive to delivering the level of performance required for true Workstation replacement, they are designed for density and consolidation, which is the last thing you want with Workstation replacement, as the resources are contended. Also, with a Hyper-Converged platform, you are (excuse the term) "robbing Peter to pay Paul" in terms of resources, as each Node (as you already mention) requires management and that resource should be being used by the application, not the infrastructure. You should (in my opinion (for what it’s worth)) be looking at a Converged platform for the appropriate level of performance (for Workstation replacement).
Regarding the multi-monitors and NVEnc, I’m not aware of any white papers on this. Typically, you’ll evaluate the resources required within a demo lab to gain an understanding of the requirements, as there are again many variables such as resolution, refresh rate, how the screens are being used, application type, how the applications are being used etc etc. Certainly, for this, as the GPU won’t (currently) be used for all monitors, the faster the CPU the better (unfortunately, further echoing my point in an earlier post, which we both agreed on).
As it stands now I believe the customer has given a price point for us to target. I’m not making the decisions on whether to use converged or hyperconverged, I’m just a slave to the process. Our current platform is a traditional converged platform featuring a VCE vBlock (Cisco UCS blades, EMC VNX storage, VMware). This is where the demo will live. Because this intial demo has to be provided to the customer for 30 days free of charge, and we don’t currently use GRID, we have decided to only deliver standard desktops.
As far as your comment regarding using the GRID test drive, I don’t think that is an option because it is only for 24 hours and would probably not be an acceptable ‘demo’ to the customer. It would be nice for nVidia to have a whitebox demo that could be branded by a company working on a proposal, and be usable for an extended period of time. Even if there were a small fee it would be better than investing the time and money into setting up our own demo environment. We’re not strictly a VDI provider, so we may not ever use that equipment again.
I’m familiar with vBlock. We use bespoke FlexPods ourselves, no Blades currently though. We have the same configuration options in terms of network and compute :-)
Are your Blades B200 M3 or M4? If M4, and assuming your VMware licenses are correct, could you not evaluate a pair of M6s from one of your suppliers? These are great little cards considering the form factor and install very easily. Then using http://www.nvidia.com/object/grid-evaluation.html you could add some GPU into your evaluation.
It won’t give you everything you need, but at least by doing this you would be able to run some of their 3D applications.
My manager has had talks with reps at nVidia and I have been in contact with someone. Nobody has ever offered up a card to test with. However, the AMD rep I talked to offered one right away. At this point we may not have enough time to integrate the GPUs and do the required testing before our demo is due to be delivered. This is a fairly rushed timeline.
(I know AMD’s FirePRO card is a bit of a different solution, but I just wanted to point out they offered us a card to test without asking.)
edit we run all B200 M3 blades, and 1 B440 M2. We are a VMware VSPP and our licensing is all squared away.
I’ve sent you a PM. Let me know what you think …
Turns out the customer did not end up requiring the 10-desktop demo, and went ahead and moved us on to the next phase. Now we actually do need to provide a demo, and here is the sticking point. We’re going to run this on Nutanix, and my manager said that the Nutanix ‘nVidia’ person says we cannot mix profiles per card. Everything we’ve been talking about here indicates we can have 2 different profiles on the same card - (1 per GPU). My manager and I were just discussing this and he initially mentioned having to buy 2 cards for the demo to be able to deliver 2 profiles. He said the Nutanix person mentioned something about licensing being the limiting factor. So now my question is does VMware have some sort of licensing restriction that you know of that would prevent 2 different profiles being used on 1 M60 card?
As long as the Hypervisor is licensed for vGPU, there are no "Hypervisor" vGPU profile limitations. Whether Nutanix applies some sort of limitation is up to them.
You CAN run mixed profiles on a single M60. The M60 has 2x GPUs, you need to run matching profiles on the same GPU, not same Card.
Depending on the vGPU profiles you choose, you may need to change the M60s default assignment policy from "Breadth-First" to "Depth-First". You can get additional detail from the "GRID vGPU User Guide" (Section 2.2.9 - Page 26) that comes with the GRID drivers.
Alternatively, start 1x VM with each Profile. This will automatically place them on separate GPUs. You can then start the rest of the VMs and they will automatically get assigned to the appropriate GPU.
Regards
This sounds like there’s confusion between the term card and GPU.
There are multiple GPU’s on each CARD
a GPU cannot mix profiles, but each GPU on a CARD can run a different profile.
Not sure what is going on, but I had a call today with Eric and Randall from nVidia and I think we’re getting some things ironed out. We talked a bit more about Blast Extreme and PCoIP as it relates to performance and the number of monitors the end-user can have. Also discussed the M10 as that may end up being a very good fit for us if Win10 VDI really does perform measurably better with GPU acceleration. As far as confusion about the cards, I’m pretty spot on with that. Maybe I should just use the term ‘die’ and ‘card’. It’s easy for me to keep straight, and maybe not so easy for others.
There was a conversation around Blast Extreme and VMware’s multi-monitor restriction when it comes to NVENC - just wondering what is the latest status?
Also, when Blast Extreme is in dual monitor mode, will the primary monitor be utilizing NVENC and only the secondary require CPU encoding?
Hi KJSteven,
since Horizon 7.0.3 there is multi-monitor support with Blast Extreme and NVENC. So this hopefully answers your question.
Regards
Simon
Appreciate the response Simon. I think my excitement may have been shortlived - the release notes of 7.0.3 https://pubs.vmware.com/Release_Notes/en/horizon-7-view/horizon-703-view-release-notes.html does tick H.264 support for multi-screens BUT will NVENC be utilized (for multi-screens) if GRID is present rather than using the CPU for the encoding?