Nvidia Tesla与Nvidia GRID vGPU购买SUMS服务是否正确理解了所有内容

你好
1）是的，您已经正确理解，但是对于GRID 5（可从9月1日获得），vWS许可证已被vDWS取代。
2）是的，M60对于Bentley MS来说足够强大，但你现在应该看看NVIDIA P40。
P40拥有24GB的FrameBuffer，但它只有一个GPU。
也就是说，它运行在更新，更高效的Pascal架构上。
展望未来，由于GRID的软件开发，P40将拥有更多可用功能。
3＆amp;
4）GRID由“并发用户”许可。
因此，您只需要与用户同时访问GPU资源一样多的许可证。
因此，对于2x vDWS用户，您需要2x vDWS许可证/ 20x vDWS用户，您需要20x vDWS许可证。
对于像Bentley MS这样的应用程序，您应该使用XenDesktop，而不是XenApp。
使用XenApp，您将遇到性能/争用问题。
我使用过Bentley，所以我很清楚它是如何使用资源的。
它可能非常耗费资源。
我只能根据自己的经验提出建议，但根据Bentley型号的大小，您可能需要每个用户8GB，这也取决于您使用的显示器数量和分辨率。
在提交部署策略之前测试解决方案的性能和负载，因为结果将确定您采用的路径和VM的规格，然后定义每个物理服务器可以容纳的用户数。
M60是双GPU卡，它有2x 8GB GPU可用。
由于单个服务器中有2个M60，因此您可以使用4个GPU。
如果您可以使用4GB vGPU配置文件，那么只要您没有达到CPU或RAM限制，您就可以在物理服务器上获得8x用户。
如果您遇到资源限制，那么您显然必须减少用户数，直到性能恢复到合适的水平。
但是，当您向外扩展时，您显然需要更多物理服务器来支持您的20个用户。
如果您选择P40，您将有更多的容量选项，因为有不同类型的vGPU配置文件可用。
我在这篇文章中附上了GRID 5所有可用vGPU配置文件的.jpg：https：//gridforums.nvidia.com/default/topic/1286/general-discussion/nvidia-announce-grid-5-update-/
因此，如果您要购买1x P40，并且您能够使用4Q vDWS配置文件，那么您可能会对每个物理服务器的6个用户感到满意（请记住，这些是高性能工作站用户，而不是标准用户，因此密度必须降低
）。
或者，如果您想使用8Q vDWS配置文件，那么购买2x P40，每台服务器将获得6个用户（CPU和RAM允许）
如上所述，您真正知道哪种规格最适合您的工作负载的唯一方法是测试并查看一切如何。
但是，我仍然建议您使用XenDesktop，而不是XenApp。
而且我相信你会知道，Bentley的CPU非常重，所以请确保你拥有最新的CPU，具有高时钟速度和核心数，因为Bentley同时使用它们！
问候



以下为原文

Hi

1) Yes you've understood correctly, however with GRID 5 (which is available from Sept 1st), the vWS license has been replaced by vDWS.

2) Yes, the M60 will be powerful enough for Bentley MS, however you should now be looking at the NVIDIA P40. The P40 has 24GB of FrameBuffer, however it only has a single GPU. That said, it runs on the newer, more efficient Pascal architecture. Looking ahead, the P40 will have more features available to it because of the software developments of GRID.

3 & 4) GRID is licensed by "Concurrent User". So you only need as many licenses as you have users concurrently accessing the GPU resource. So for 2x vDWS users, you need 2x vDWS licenses / 20x vDWS users you need 20x vDWS licenses.

With applications like Bentley MS, you should be using XenDesktop, not XenApp. With XenApp, you will run in to performance  / contention issues. I've used Bentley, so I'm well aware of how it uses resources. It can be extremely resource hungry.

I can only advise on my own experiences but depending on the size of your Bentley models, you may very well need the full 8GB per user, this also depends on how many monitors you use and the resolution of them.

Test the solution for performance and load before committing to a deployment strategy, as the results will determine which path you take and the specification of your VMs which will then define how many users you can fit per physical server. The M60 is a dual GPU card and it has 2x 8GB GPUs available. As you have 2x M60s in a single server, you have 4x GPUs to use. If you can get away with a 4GB vGPU profile, then you'll be able to get 8x users on a physical server as long as you don't hit CPU or RAM limitations. If you run in to resource limitations, then you'll obviously have to reduce the user count until the performance returns to a suitable level. However you'll obviously need more physical servers to support your 20 users when you scale out.

If you choose the P40, you'll have more options in terms of capacity, as there are differing types of vGPU profile available. I've attached a .jpg of all available vGPU profiles for GRID 5 in this post: https://gridforums.nvidia.com/default/topic/1286/general-discussion/nvidia-announce-grid-5-update-/

So if you were to purchase 1x P40, and you were able to use the 4Q vDWS profile, then you may be happy with 6 users per physical server (remember, these are high performance workstation users, not standard users, so the density must decrease). Or, if you wanted to use the 8Q vDWS profile, then purchase 2x P40 and you'll get 6 users per server (CPU and RAM permitting)

As said, the only way you'll really know which specifications are best for your workload, is to test and see how everything performs. However, I would still recommend that you use XenDesktop, not XenApp. And As I'm sure you'll know, Bentley is very CPU heavy, so make sure you have the latest CPU generation with high clock speed and core count, as Bentley uses both!

Regards

BJones，
感谢您提供有关P40的更多信息。
我们希望在运行VMware vSphere 6.5和Horizo​​n 7.x的Dell PowerEdge R740服务器中使用P40卡。
但是，我发现P40上没有与VMware Horizo​​n特别相关的文档，另外P40没有在VMware的HCL上列出。
您何时希望为P40和VMware发布更多信息和支持？
谢谢！



以下为原文

BJones,  

Thank you for the additional information above about the P40.  We have a desire to utilize the P40 cards in Dell PowerEdge R740 servers running VMware vSphere 6.5 and Horizon 7.x.  However, I am finding NO documentation on the P40 specifically related to VMware Horizon and additionally the P40 is not listed on VMware's HCL.  When do you expect further information and support to be released for the P40 and VMware?

Thank you!

嗨TOuimet
为了保持这个主题相关性，我在你原来的帖子上做了回复 -  https://gridforums.nvidia.com/default/topic/1286/general-discussion/nvidia-announce-grid-5-update-/
问候
本



以下为原文

Hi TOuimet

To keep this Topic relevent, I've responded here on your original post - https://gridforums.nvidia.com/default/topic/1286/general-discussion/nvidia-announce-grid-5-update-/

Regards

Ben

亲爱的，B。琼斯。
感谢您的支持！
我们将使用Citrix XenDesktop。
对于服务器，我们计划使用512Gb的ECC。
2个至强处理器E5-2690v4。
2 x Nvidia Tesla M60
此配置基于4个用户启动。
这个配置会不够用？



以下为原文

Dear, B. Jones.

Thank You for your support!
We will use Citrix XenDesktop.

For server we are planning to use 512Gb of ECC. 2 Xeon Processors E5-2690v4.
2 x Nvidia Tesla M60

This configuration based for 4 users for start.

Will be this configuration enough?

嗨谢尔盖
是的，那些CPU应该没问题，如果在测试期间发现可以使用4GB vGPU配置文件，那么它可以为服务器添加更多用户。
还有一个8 Core 3.2Ghz v4可能是另一种选择。
512GB在4个用户之间分配很多。
你可以使用远远少于这个，但仍然有足够的。
您正在考虑分配每个VM多少CPU / RAM？



以下为原文

Hi Sergey

Yes those CPUs should be fine, and if during testing you find that you can use the 4GB vGPU profile, then it gives you the potential to add more users to the server. There is also an 8 Core 3.2Ghz v4 which could be another option.

512GB is a lot to split between 4 users. You could use far less than that and still have more than enough.

How much CPU / RAM are you looking at allocating each VM?