Microsoft Server 2012R2/RemoteFX是否与M6/M60兼容？

嗨Trollkeeper，
REmoteFX的工作方式有点不同（有时称为vGPU），所有需要使用GPU的是与Windows 2012 R2兼容的GPU，而RemoteFX协议对GPU的使用将由Microsoft GPU提供。
M60 / M6 / M10都支持Windows 2012 R2，所以你没问题，因为没有直接的互操作（可能为什么信息在互联网上如此轻松）。
我会尝试写一些正式的文档。
RemoteFX“vGPU”不是硬件-vGPU（根据Citrix / VMware），并且在两者之间使用OS，您将看不到与任何GPU供应商的hardware-vGPU相同的好处。
最好的祝愿，
雷切尔



以下为原文

Hi Trollkeeper,

The way REmoteFX works is a bit different (it's sometimes called vGPU), all it needs to use the GPU is a GPU compatible with windows 2012 R2 and the RemoteFX protocol's use of the GPU will be provided by the Microsoft GPU. The M60/M6/M10 all support Windows 2012 R2 so you will be fine as there is no direct interop (probably why info is so light on it on hte internet). I'll try and get some formal documentation written up.

The RemoteFX "vGPU" isn't hardware-vGPU (as per Citrix/VMware) and with the OS in between you won't see the same benefits as hardware-vGPU from any GPU vendor.

Best wishes,
Rachel

嗨RachelBerry
您的信息在RemoteFX上不准确
“在Windows Server 2012中，RemoteFX vGPU允许虚拟机中的渲染和计算操作使用虚拟主机上的物理GPU进行硬件加速。在Windows Server中提供硬件加速体验涉及多个组件”
链接到完整文章：
https://blogs.msdn.microsoft.com/rds/2014/06/06/understanding-and-evaluating-remotefx-vgpu-on-windows-server-2012-r2/
您所说的是Server 2012中的RDP情况，但是当您打开RemoteFX时它确实使用了硬件。
我可以确认我的K1网格卡的区别。
如果我通过将客户端上的网络管道丢弃到100 MBit来关闭RemoteFX，我会安装一个不同的虚拟图形适配器，它可以作为基于软件的图形控制器运行。
当我将其切换到千兆位时，我会在具有硬件加速功能的虚拟机上安装不同的驱动程序。
我测试了与RemoteFX兼容的SolidWorks。
软件和硬件支持之间的区别非常明显。
遗憾的是，Autodesk套件没有支持K1虚拟化的驱动程序，只能运行软件模式。
由于缺乏对OpenGL的支持，许多其他软件套件也仅运行基于软件的模式，这应该是Server 2016中支持的.DirectX是目前唯一支持RemoteFX的协议。
但问题是是否支持M6 / M60卡。
仅仅因为有卡的驱动程序并不意味着可以使用在Hyper-V上启用RemoteFX的所有重要复选框。
如果那个复选框不可用，那么在卡上花费数千美元绝对没有用，因为你将无法使用它。



以下为原文

Hi RachelBerry

Your information is not accurate on RemoteFX

"In Windows Server 2012, the RemoteFX vGPU allows render and compute operations in a virtual machine to be hardware accelerated using a physical GPU on the virtualization host. There are multiple components involved in delivering a hardware accelerated experience in Windows Server"

link to full article:

https://blogs.msdn.microsoft.com/rds/2014/06/06/understanding-and-evaluating-remotefx-vgpu-on-windows-server-2012-r2/


What you are saying is the case with RDP in Server 2012, but when you switch on RemoteFX it does use the hardware.  I can confirm the difference on my K1 Grid cards.  If I switch down RemoteFX by dropping the network pipe on the client to 100 MBit I get a different virtual graphics adapter installed which does operate as a software based graphics controller.  When I switch it up to gigabit I get a different driver installed on the virtual which has hardware acceleration.

I tested SolidWorks out which is compatible with RemoteFX.  The difference between software and hardware support is very noticeable.  The Autodesk suite unfortunately doesn't have a driver to support the K1 virtualization and runs software mode only.  Many other software suites only run software based mode as well due to the lack of support for OpenGL, something that is supposed to be supported in Server 2016.  DirectX is currently the only supported protocol for RemoteFX.

The issue is however if the M6/M60 card is supported.  Just because there are drivers for the card does not mean that the all important check box to enable RemoteFX on Hyper-V will be available.  And if that checkbox is not available then there is absolutely no use in spending the thousands of dollars on the card as you will get no use of it.

雷切尔的信息绝对准确。
RemoteFX虚拟图形适配器使用API​​拦截机制，该机制从来宾VM中的Microsoft驱动程序转换API调用，并将这些调用传递给Hyper-V主机中的硬件驱动程序。
您在K2上看到的是Microsoft驱动程序拦截并转换这些API调用并将它们传递给Hyper-V主机中的Nvidia驱动程序。
值得注意的还有几点
Remote FX虚拟图形适配器仅在功能齐全的RemoteFX会话中启用，任何其他连接方法都禁用它。
它也仅限于DirectX，2012R2版本不支持OpenGL（虽然Server 2016可以）。
所以我很惊讶你看到Solidworks使用GPU（它可能在非常低的功能级别上运行在CPU上）。
无论如何，你问过M60对Hyper-V的支持。
对于M60，目前还没有正式的裸机部署支持，这种类型的安装是因为它安装在Server 2012R2中（不支持核心服务，你必须拥有GUI）。
有几个原因可能会在未来发生变化。
但是，如果您很高兴测试不支持的功能，Tesla M60在技术上可以与Hyper-V 2012R2一起使用，以提供虚拟图形功能。
您将需要主机中的其他图形适配器来安装您的操作系统，您仍然需要在vPC版级别上为每个正在运行的VM获取许可证。
获得这些许可证并因此访问许可证门户后，您就可以下载适用于Windows驱动程序的GRID。
评估许可证是可用的btw。
随着即将发布的Server 2016，Nvidia计划支持Tesla M60，M6和新推出的DDA M10主板。
自Server 2016 TP4以来，此功能已可用于试用，因此使用与上述相同的驱动程序，您可以测试该功能。
如果您有M60卡并需要评估许可证，可以在此处申请：
http://www.nvidia.com/object/grid-evaluation.html



以下为原文



Rachel's information is absolutely accurate.

RemoteFX virtual graphics adapter uses an API interception mechanism that translates API calls from the Microsoft driver in the guest VM and passes those calls to the hardware driver in the Hyper-V host.

What you see with the K2 is the Microsoft driver intercepting and translating those API calls and passing them to the Nvidia driver in the Hyper-V host.

It's also worth noting a couple of things

The Remote FX virtual graphics adapter is only enabled in fully featured RemoteFX sessions, any other connection method disables it.

It's also limited to DirectX, OpenGL is not supported in the 2012R2 release ( Server 2016 does though). So I'm surprised you're seeing Solidworks use the GPU (it may be running on CPU at a very low feature level).


Anyhow, you asked about M60 support for Hyper-V.

In the case of the M60 there is currently no official support for bare metal deployments, which this type of installation is becasue it's installed into Server 2012R2 (no support for core services, you must have the GUI).

There are several reasons for this which may change in the future.

However, if you're happy testing out an unsupported feature, the Tesla M60 can technically be used with Hyper-V 2012R2 to deliver the virtual graphic functionality. You will need an additional graphics adapter in the host to install your OS and you will still require a license for each running VM at the level of vPC Edition. Once you have these licenses and hence access to the license portal you will be able to download the GRID for Windows drivers. Evaluation licenses are available btw.

With the upcoming release of Server 2016 Nvidia plans to support  Tesla M60, M6 and the newly announced M10 boards for DDA. This feature has been available for trial since Server 2016 TP4, so with the same drivers as above you can test the functionality.

If you have an M60 card and require evaluation licenses, you can request them here:

http://www.nvidia.com/object/grid-evaluation.html

RemoteFX“vGPU”将使用GPU，因此在某种程度上会发生硬件加速，但因为它是通过操作系统有一个adidtional cerittion / support层，你获得的加速水平将低于vSphere / XenServer vGPU。
因此，虽然GPU将使您受益，但您将无法获得与硬件vGPU堆栈相同的优势。
这有助于使其更清晰吗？
对不起，如果不清楚...请提出更多问题......



以下为原文



RemoteFX "vGPU" will use the GPU so hardware acceleration to some extent will occur but because it's via the OS there is an adidtional ceritfication/support layer and the level of acceleration you get will be below that of vSphere/XenServer vGPU. So whilst a GPU will benefit you, you won't get the same level of benefit as on the hardware vGPU stacks.

Does that help make it clearer? Sorry if it wasn't clear... please do ask more questions...

大家好，
刚刚跟进最初的问题：我们最终在Dell PowerEdge R730服务器上购买了Tesla M60用于测试目的。
戴尔不会在使用M60 GPU购买的服务器上安装任何操作系统，因此它没有操作系统。
问题是我们无法让Hyper-V RemoteFX识别M60，所有选项都显示为灰色，好像没有检测到这个GPU或不相容。
我们在2012 R2和2016 TP5上都有这个结果。
服务器有一个用于操作系统安装的板载Matrox视频卡。
GPU驱动器是从NVidia网站下载的，我们尝试了多个驱动程序版本无济于事。
我们做错了什么想法？
任何帮助表示赞赏。
谢谢，
谢谢，
IT_Vancouver



以下为原文

Hi All,

Just following up on the initial question here: We did end up purchasing a Tesla M60 on a Dell PowerEdge R730 server for testing purposes. Dell doesn't install any OS on the servers purchased with M60 GPU, so it arrived with no operating system.

Problem is we have not been able get Hyper-V RemoteFX to recognize the M60, all options are greyed out as if this GPU is not detected or is incompatible. We this on both 2012 R2 and 2016 TP5, with same results.

The server has an on-board Matrox video card for OS install. The GPU drives were downloaded from NVidia site, we tried multiple driver versions to no avail. Any ideas on what we're doing wrong? Any help is appreciated.

Thanks,


Thanks,
IT_Vancouver

嗨，
如果您查看我上面的帖子，您会发现我们不支持包含Hyper-V的裸机操作系统部署中的M60。
如果您仍希望以不受支持的方式测试Hyper-V，则在虚拟化中使用M60进行图形处理所需的驱动程序需要许可证，而NVIDIA.COM无法提供这些驱动程序。
您只能从许可证门户获取它们（见下文）
您还需要将M60配置为以图形模式运行（它们以计算模式运行）。
通过此途径访问评估许可证。
虽然不受支持，但仍需要许可才能使用。
在这种情况下，您可以与VMware的vSGA配合使用，因此您需要为配置为使用GPU的每个正在运行的VM（即RFX适配器）提供“虚拟PC”版许可证



以下为原文

Hi,

If you take a look at my post above you'll see that we don't support the M60 in bare metal OS deployments which includes Hyper-V.



If you still want to test Hyper-V in an unsupported manner, then the drivers you require to use the M60 for graphics in virtualisation require a license, and are not available from NVIDIA.COM. You can only get them from the license portal (see below)

You will also need to configure the M60 to run in graphics mode (they ship in compute mode).



Access to evaluation licenses is via this route.



Although unsupported you would still need to license it for use. In this scenario you would be comaprable to vSGA from VMware so you would require "virtual PC" edition licenses for each running VM configured to use the GPU (i.e. the RFX adaptor)

嗨大家好，我对使用直通GPU的Windows Server 2016上的DDA有疑问。
我的Dell PowerEdge T630在我的主机上安装了4个Tesla M60，并使用正确的驱动程序正确配置。
我已正常安装Hyper-V和First Thing我注意到，我无法在设置上的RemoteFX部分看到我的物理GPU，最后我发现开机即用的Tesla M60不支持RemoteFX，因为
在基于Compute而不是Graphics模式的架构上。
这对我来说没关系，因为我不打算使用RemoteFX，我想测试DDA，所以我没有打扰自己出现在我的Hyper-V上的GPU（这是我的第一个问题）。
即使我没有将我的GPU用于RemoteFX，我是否还需要在hyper-V上看到GPU以使DDA完全正常运行？
我已经通过powershell命令来禁用/卸载，并最终将GPU安装到我的Windows 10 VM，通常没有错误。
我已将我的主机上使用的相同GPU驱动程序安装到我的VM上，它安装正确，我也看到Perfmon上的GPU设置，一切似乎都很好。
问题是，每当我尝试运行简单的GPU强烈测试时，性能都很糟糕，它让我相信，出于某种原因，即使我的TESLA M60安装到VM，VM仍然使用默认的Hyper-V
集成GPU。
在我的设备管理器上，我看到了板载的Hyper-V视频卡和NVIDIA TESLA M60的正下方。
这让我觉得我应该以某种方式优先考虑TESLA M60而不是集成Hyper-V。
我想知道我错过了什么？
无论如何，我需要在我的Hyper-V控制台上显示GPU才能使用DDA 100％功能？
先谢谢你



以下为原文

Hi Guys, I have a question regarding DDA on Windows Server 2016 using pass-through GPU. My Dell PowerEdge T630 has 4 Tesla M60 installed on my host properly configured with the proper Drivers.
I have installed Hyper-V normally and First Thing I noticed is that I cant see my physical GPU on the RemoteFX section on the Settings, which at the end I figured out that Tesla M60, out of the box, is not supported for RemoteFX because on the Architecture based on Compute rather than Graphics mode.
That is ok for me, since I am not going to use RemoteFX, I want to test DDA instead, so I didnt bother myself with the GPU showing up on my Hyper-V (this is my first Question).

Even though I am not using my GPU for RemoteFX, do I still need to see the GPU on hyper-V in order to have DDA fully functional?

I have gone through the powershell commands to disable / Dismount and finally mount the GPU to my Windows 10 VM normally without errors. I have installed the Same GPU driver used on my host to my VM, it installed properly, I also see the GPU settings on Perfmon, all seems to be good. The problem is, whenever I try to run a simple GPU intense test, the performance is terrible, it makes me believe that for some reason, even though my TESLA M60 is mounted to the VM, the VM is still using the default Hyper-V Integrated GPU.
On my device manager I see both onboard Hyper-V video card and right below the NVIDIA TESLA M60. that makes me also think that I should somehow prioritize the TESLA M60 over the Integrated Hyper-V one.

I would like to know what I am missing?

No matter what, I need to have the GPU showing up on my Hyper-V console in order to use DDA 100% Functional?


Thank you in advance

None


以下为原文

Hi edenlincoln,

I've written up some notes on Hyper-V DDA, https://virtuallyvisual.wordpress.com/2017/01/18/nvidia-grid-and-microsoft-windows-server-oss-and-hyper-v/


It includes how to get the DDA driver - have you got the right driver and eval license?

Best wishes,
Rachel

嗨雷切尔，
感谢您的回复。
我还没找到你的帖子（我肯定在看错了地方）。
我会一看，我会在服务器上重试该程序后立即回复您。
只是我上次测试的快速更新。
环境：
戴尔动力T630
4张TESLA M60卡（8 GPUS）
1）成功为DDA选择了一个（从主机卸载并将其安装到VM）。
1.1）已成功将Tesla M60驱动程序从主机安装到VM，正常安装，但我注意到性能不佳（似乎它仍然在集成的Matrox板载卡下运行）
PS：此测试在没有任何GRID驱动程序或许可证的情况下运行。
这是来自NVIDIA网站驱动程序的裸装置。
我测试了M10驱动程序的Windows 10 x64和Server 2016版本，性能没有变化。
第二次尝试（一旦我从我的eval帐户运行Linux ISO媒体，一旦我将GPU从计算模式更改为图形模式，服务器就崩溃了）。
现在，在安装任何新的Windows Server 2016之后，一旦我安装了GPU驱动程序（M60），服务器永远不会重新启动。
题：
1）我应该在测试环境中使用哪种驱动程序？
适用于Windows 10 / Server 2016的常规TESLA M60 x64或来自eval配置文件的GRID？
2）谈到Hyper-V，如果我无法在hyper-v上找到M60 GPU以使DDA正常工作或GPU必须出现在Hyper-V上才能让DDA在我的VM上运行，这是否重要？
上次我将GPU传递给VM并且即使性能很糟糕也能正常工作，但Hyper-V管理器设置的Dropbox上没有可用的GPU。
感谢您的帮助。
伊甸园奥利维拉



以下为原文

Hi Rachel,

Thanks for your reply. I havent found your post (I was definitely looking at the wrong place). I will take a look and Ill get back to you as soon as I retry the procedure on my Server.

Just a quick update on my last test.

Environment:

DELL POWEREDGE T630
4 TESLA M60 cards (8 GPUS)

1) Successfully selected one for DDA (Dismounted it from host and mounted it to a VM).
1.1) Have successfully installed the Tesla M60 Driver from the host to the VM, installed normally, however I've noticed performance was not good (Seems it was still running under the integrated Matrox onboard card)

PS: This test was run without any GRID driver or license isntalled. it was a bare installation from the driver from NVIDIA Website.
I have tested both Windows 10 x64 and Server 2016 versions of the M60 driver, no changes on the performance.

Second attempt (Once I ran the Linux ISO media from my eval account, Server has crashed once I have changed the GPUs from Compute to Graphics mode). Now, after any new windows Server 2016 installation, once I install the GPU driver (M60) server never comes back up.

Question:

1) Which Driver should I use for my test environment? the regular TESLA M60 x64 for windows 10 / Server 2016 or the GRID from the eval profile?

2) Speaking about Hyper-V, Does it matter if I cant find the M60 GPU on my hyper-v in order to get DDA working or the GPU MUST show up on Hyper-V in order to get DDA working on my VM?
Last time I pass-through the GPU to the VM and it worked even though performance was terrible, there was no GPU available on the dropbox from Hyper-V manager Settings.

Thank you for your help.

Eden Oliveira