Grid 2.0

K1和K2在2016年底显然已经结束，加上不会为K1 / K2 GPU提供Linux VM支持，这种情况更加恶化。
这也意味着在某些时候也会切断驾驶员的发展。



以下为原文

The situation is exacerbated by the K1 and K2 apparently being end-of-lifed at the end of 2016, plus there will be no Linux VM support provided for the K1/K2 GPUs. It also means driver development will also be cut off at some point.

哎哟!
这听起来真是个坏消息： -  /。
我无法找到nVidia的生命终结政策。
到生命结束时，你的意思是：
a）这些卡的销售终止（SW支持/硬件更换将继续提供几年）
b）或这些卡的生命/支持结束（不再需要更换硬件和软件开发）？
在假期之前向我的服务器供应商（思科）的本地产品经理发出消息 - 新卡应该在2016年1月底之前开始发货，并且他们应该有更多的经销商信息。
2016年1月中旬左右的定价。
Spenatek



以下为原文

Ouch! This is really bad news to hear :-/.

I was not able to find an End of Life policy of nVidia. By end-of-life do you exactly mean:

a) end of sale of those cards (SW support/HW replacement would continue to be available for a few years)
b) or end of life/support of those cards (no more HW replacement and SW developement)?

Spoke to local product manager of my servers supplier (Cisco) before the holidays - the new cards should start shipping by end of January 2016 and they should have more information for resellers incl. list prices near mid-January 2016.

Spenatek

Spenatek，
它可能意味着没有更多的驱动程序被释放。
当然，就硬件而言，现有卡将被维护，但没有新的驱动程序，这意味着在许多情况下不再需要操作系统升级以保持兼容。
我认为EOL还没有正式文件。
关于GPU直通许可，如果有人感兴趣，你肯定需要一个“工作站扩展”许可证开始，因为这显然是唯一一个支持GPU直通（再次，请参阅http://www.nvidia.com/object/
grid-technology.html以获取详细信息）但是仍然不清楚通过直通连接的每个VM是否需要在其上安装单独的许可证，或者是否可以由任何进行直通连接的VM使用“池”许可证。
令人难以置信的是，在这段时间之后，这仍然是如此令人困惑和无证。
除此之外，我甚至无法让像戴尔这样的大型零售商获得许可信息！
据说CDW-G现在有许可价格，所以我可能会问他们。
这是一次极其令人沮丧的全面体验。



以下为原文

Spenatek,
It probably means for one that there will be no further drivers released. Sure, existing cards will be maintained as far as the hardware goes, but without new drivers that means in many cases no more OS upgrades to remain compatible. I don't think there is an official document out on the EOL yet.

Regarding GPU passthrough licensing, if anyone is interested, you definitely need a "Workstation Extended" license to begin with, as that is evidently the only one that supports GPU passthrough (again, see http://www.nvidia.com/object/grid-technology.html for details) but it is still not clear if every VM connecting via passthrough needs a separate license installed on it or if a "pool" of licenses could be drawn upon by any VM making a passthrough connection.

It is unbelievable that after all this time that this is still so confusing and undocumented. That aside, I cannot even get a big retailer like Dell to get me licensing information! Supposedly CDW-G has licensing pricing now, so I may ask them. This has been an extremely frustrating experience all-around.

谢谢Tobias。
对于passthrough以及OpenCL / CUDA，如果我没有记错，你需要一个8GB的vGPU配置文件，这实际上意味着每个GPU 1个VM =每个M60卡2个VM ...
关于许可 - 根据http://images.nvidia.com/content/pdf/grid/guides/quickstartguide-v01.pdf，看起来你有一个许可证池，由许可服务器管理，许可证是由
VM中的nVidia驱动程序。
因此，如果在驱动程序正确指向许可证服务器的情况下启动VM，它将获得适当的许可证，如果关闭VM，许可证将返回池中。
我在官方的nVidia文件中读到了某些内容，我现在无法找到，如果没有适当的许可，FPS将降低到无法使用的费率。
这意味着您可以安全地使用nVidia驱动程序启动VM而无需nVidia许可证（GPU可以正常工作，但无法用于常规工作）。
Spenatek



以下为原文

Thanks Tobias.

For the passthrough as well as OpenCL/CUDA, if I'm not mistaken, you need a 8GB vGPU profile, which effectively means 1 VM per GPU = 2 VM per M60 card...

Regarding licensing - according to http://images.nvidia.com/content/pdf/grid/guides/quickstartguide-v01.pdf it looks like the you have a pool of licenses, managed by a licensing server and the license is requested by the nVidia driver in the VM. So if you power-on a VM with the driver correctly pointing to license server, it'll acquire an appropriate license, if you power-off the VM, the license is returned to the pool.

I read somewhere in an official nVidia document, which I'm now unable to find, that without proper licensing FPS will be reduced to unusable rate. That mean you can safely boot a VM with nVidia driver without a nVidia license (GPU will work, but would be unusable for regular work).

Spenatek

谢谢你，斯彭泰克。
这是一个可怕的许可模式IMO。
启动VM并分配许可证并不意味着VM必须使用GPU直通实例。
考虑一下VM通过XenDesktop在本地运行其大多数应用程序的情况，但是一些需要GPU的应用程序托管在XenApp服务器上并通过GPU直通访问。
您所说的是，即使无法访问XenApp服务器，仍然会仅仅因为VM已启动而分配许可证，无论它是否真正使用了GPU。
这意味着你必须为你打开的每个虚拟机都拥有一个许可证（在大多数情况下，这意味着所有这些都是，所有的时间），即使只有少数人在任何特定时间都使用GPU直通。
因此，基本上，这是双重浸入，使您支付8GB GPU直通虚拟机的费用，然后不仅适用于任何使用直通的虚拟机，而且适用于所有正在运行的可能使用GPU直通的虚拟机。
换句话说，您被迫购买可能在某些时候访问GPU直通的任何和所有虚拟机的许可证，即使它在任何给定时间只有一小部分具有活动连接。
从本质上讲，许可模式使客户为在任何给定时间点可能比实际需要的使用量更少的情况付费。
太棒了... ：-（
而且，是的，你对降低的比率是正确的;
据我所记，它低至3 FPS。



以下为原文

Thank you, Spenatek.
This is a terrible licensing model IMO. Powering on a VM and allocating a license doesn't mean the VM will necessarily make use of a GPU passthrough instance. Consider the case that a VM runs most of its apps locally, say via XenDesktop, but some apps requiring a GPU are hosted on a XenApp server and accessed via GPU passthrough. What you are saying is that even if there is no access to the XenApp server, the license will still be allocated just because the VM is powered up, whether it's really making use of the GPU or not. That means you have to have a license for every VM you have turned on (which in most cases, that means all of them, all the time) even if only a few are really making use of GPU passthrough at any given time. So, basically, this is double-dipping, making you pay for an 8GB GPU passthrough VM and then not only for any VM making use of passthrough, but for all VMs that are running that might make use of GPU passthrough. In other words, you are forced to buy licenses for any and all VMs that may at some point access the GPU passthrough, even if it's only a fraction of them at any given time that have active connections. In essence, the licensing model makes customers pay for what in many cases may be less usage than actually needed at any given point in time. Just wonderful... :-(

And, yes, you're correct about the rate being lowered; it's down to 3 FPS, from what I recall.

托比亚斯，
许可证是永久性的，许可证没有维护成本。
每个永久许可证都需要SUMS的第一年（支持，更新和维护）。
第一年后，您无需继续使用，也可以使用订阅期限内可用的软件功能。
有关SUMS优点的完整详细信息，请访问：http：//www.nvidia.com/object/grid-support.html



以下为原文



Tobias,
The license is perpetual, there are no maintenance costs for the licenses.

The first year of SUMS (Support, Update and Maintenance) is required for each perpetual license.
After the first year, you are not required to continue and may use the software features available during your subscription term.

Complete details about the benefits of SUMS can be found at: http://www.nvidia.com/object/grid-support.html

托比亚斯，
每个活动的传递实例都需要通过许可证服务器实现的许可证。
如果您有六个VM配置为使用具有两个GPU的单个M60，并且每个VM配置为在两个GPU之一上使用GPU-Passthrough，则此服务器上可同时使用的最大VM数将为两个。
每个VM在启动时都会签出虚拟工作站扩展许可证，并在关闭时释放它。
许可证不是按配置的VM，而是每个活动/引导的VM。



以下为原文



Tobias,
Each active pass-through instance would require a license that would be fulfilled via a license server.
If you have six VMs configured to use a single M60 with two GPUs, and each VM is configured to use GPU-Passthrough on one of the two GPUs, the maximum number of VMs that can be used simultaneously on this server would be two.
Each VM would checkout a Virtual Workstation Extended license when it starts up and releases it when it is shutdown.
The license is not per-configured VM, it is per active/booted VM.

杰里米
谢谢你的回复。
回复：
1）在第一年之后，期望一个人放弃支持和更新是荒谬的！
你要做的最后一件事就是大量投资某些技术，并将其冻结超过一年的时间，以达到目前的驱动程序和支持状态。
这是一种非常不现实的方式，使其看起来更具成本效益。
2）谁在实验室不使用时关闭虚拟机？
它们一直在运行，人们经常在许多环境中登录/关闭（是的，我知道，教育环境是大海中的小鱼）。
显然，这种使用模式的目标是拥有稳定用户的环境，他们整天在CAD / CAM项目上工作，在工作站上不做任何其他事情。
再次，教育环境是非常不同的。
这种许可模式不适合它，并最终增加了大量的额外成本。



以下为原文

Jeremy,
Thank you for your responses. In reply:

1) That is ludicrous to expect one to drop support and updates after the first year! The last thing you'd want to do is to invest heavily in some technology and freeze it beyond a year's time to its current status of drivers and support. That is a very unrealistic way of making it appear to be more cost effective.

2) Who shuts down VMs when not in use in a lab? They run all the time and people log on/off frequently in many environments (yes, I know, educational environments are small fish in a big ocean). Apparently this model of usage is target toward environments with steady users who work on CAD/CAM projects all day long and do nothing else on their workstations. Again, the educational environment is very different. This licensing model is a poor fit to it and ends up adding significant additional cost.

以下为原文

我也看到了这一点。
我希望看到精美的印刷品，加上第一年后，您将看到仍然维护费用。
我现在认为纯粹用于GPU直通的Quadro M6000可能是一种选择，因为它与K2组合的GPU相当或稍好一些。
它刚刚于2015年3月发布，因此应该存在一段时间，并且在Citrix平台上受支持。
我的一个问题是关于扩展，如https://gridforums.nvidia.com/default/topic/561/xenapp-with-nvidia-grid/autodesk-revit-and-user-density/中所述。
我们也可能决定等待它，看看GPU市场会发生什么。



以下为原文




I did see this, as well. I'd like to see the fine print, plus after the first year, you are going to see still maintenance fees.

I am now thinking a Quadro M6000 purely for GPU passthrough might be one option as it's about the equivalent or slightly better than the combined K2 GPUs. It was just released in March 2015, so should be around for a while and is supported on Citrix platforms. My one concern is about scaling, as noted in https://gridforums.nvidia.com/default/topic/561/xenapp-with-nvidia-grid/autodesk-revit-and-user-density/. We may also decide to just wait it out and see what happens with the GPU market.

这个优惠在欧洲也有效吗？



以下为原文



Is this offer valid also in Europe?

这个优惠在欧洲也有效吗？
是



以下为原文



Is this offer valid also in Europe?

Yes

如果有人想知道这个优惠有什么限制，据我所知，这里是他们的：
每位客户仅需2个M60或4个M6单位。
1美元许可证的数量受许可证类型的限制。
例如，客户可以购买最多：
•64个Virtual PC Edition许可证，或
•16个Workstation Edition许可证，或
•4个Workstation Extended Edition许可证，或
•它们的组合，最多4个特斯拉M6或2个特斯拉M60 GPU支持的最大CCU数量。



以下为原文

In case anyone wants to know what the limitations are on this offer, here they are, as far as I am informed:

Only 2 M60 or 4 M6 units per customer.
The number of $1 licenses is limited by the license type. For example, a customers may purchase up to:

• 64 Virtual PC Edition licenses, or
• 16 Workstation Edition license, or
• 4 Workstation Extended Edition licenses, or
• a combination thereof, up to the maximum number of CCUs supported by 4 Tesla M6 or 2 Tesla M60 GPUs.

嘿。
如何将sdk grid 2.0芯片送到Maxwell？
目前，我们的团队正在开发基于GRID SDK版本2.3的平台：2014年4月2日，用于芯片Kepler。



以下为原文

Hey. how to get the sdk grid 2.0 chips to Maxwell? At the moment our team is developing a platform based on GRID SDK Version 2.3: April 2, 2014 for chips Kepler.

现在我们正在开发一个使用GRID SDK版本2.3捕获和编码基于Kepler（K1 / K2和K340 / K520）芯片的系统。
目前，有一个新的芯片组Maxwell（M60 / 6），它早先使用GRID SDK。
GRID SDK版本4.x 2015年9月或更早版本。



以下为原文

Now we are developing a system to capture and coding based chips Kepler (K1 / K2 & K340 / K520) using GRID SDK Version 2.3. At the moment, there was a new chipset Maxwell (M60 / 6) which uses GRID SDK earlier. GRID SDK Version 4.x September 2015 or earlier.

这个优惠在欧洲也有效吗？
是
杰森，谢谢。
如果我有兴趣，应该与谁联系？
nVidia或OEM供应商（在我们的情况下等于思科）？
你能提供任何联系吗？
提前致谢。
托比亚斯，谢谢你对这些局限性进行了抨击。
问候，
spenatek



以下为原文



Is this offer valid also in Europe?

Yes

Jason, thanks. Who should I contact about this offer if I'm interested? nVidia or OEM vendor (equals Cisco in our case)? Can you provide any contacts? Thanks in advance.

Tobias, thanks for claryfing the limitations.

Regards,

spenatek

否 -  GRID 2.0仅限Maxwell（M60 / 6）和未来的架构。
它不会被添加到开普勒。
如果您在vGPU中需要Linux或4K显示器，则它仅为GRID 2.0。
为什么NVidia不支持Linux guest虚拟机用于K1 / K2 vGPU共享（K1 / K2不应该有任何技术问题支持它）？
我们可以！
（非官方）
SteamOS也是linux，它与vGPU K1 / K2一起工作（几乎）。



以下为原文


No - GRID 2.0 is Maxwell (M60/6) only and future architectures. It will not be added to Kepler. If you require Linux or 4K displays in vGPU, it's GRID 2.0 only.

Why NVidia does not support linux guest for K1/K2 vGPU sharing (there should not be any technical problem to support it for K1/K2) ?

Yes, we can! (unofficially)

SteamOS is also linux and it works (nearly) fine with vGPU K1/K2.

这个优惠在欧洲也有效吗？
是
杰森，谢谢。
如果我有兴趣，应该与谁联系？
nVidia或OEM供应商（在我们的情况下等于思科）？
你能提供任何联系吗？
提前致谢。
托比亚斯，谢谢你对这些局限性进行了抨击。
问候，
spenatek
OEM /思科应该能够提供帮助。
它可通过区域分销商（英国EMEA的Ingram）向经销商提供。
如果您在获取详细信息时遇到问题，请发送给我



以下为原文



Is this offer valid also in Europe?

Yes

Jason, thanks. Who should I contact about this offer if I'm interested? nVidia or OEM vendor (equals Cisco in our case)? Can you provide any contacts? Thanks in advance.

Tobias, thanks for claryfing the limitations.

Regards,

spenatek

OEM / Cisco should be able to help. It's available to resellers through the regional distributor (Ingram in EMEA).

If you have trouble getting details send me a PM.

嗨，大家好，
我不想深入了解定价的细节或许可的复杂性。
但是将GRID 2.0拆分为在特斯拉硬件GPU之上购买的软件产品的概念背后的逻辑。
这两个博客应该解释它在质量保证方面应该有什么好处，并允许在HPC和图形之间重新利用...想法？
https://virtuallyvisual.wordpress.com/2016/02/04/mode-switching-on-nvidia-m60-grid-2-0-cards-modeswitch-tool/
https://virtuallyvisual.wordpress.com/2016/02/05/nvidia-m60-gpus-how-they-support-cuda-opencl-and-compute-in-graphics-mode/
该软件产品的实际细节和定价是业务部门的详细信息。
但是从技术方面和工程过程;
以及虚拟化硬件重新利用的好处。
虽然采用顶级软件购买GPU的新模式。
我觉得有道理。
但是想听听别人的想法。
雷切尔



以下为原文

Hi guys,

I don't want to get into the details of pricing or the complexity of licensing. But the logic behind the concept of spliting GRID 2.0 into a software product bought on top of a Tesla Hardware GPU.

These two blogs should explain how it should have benefits in terms of QA and allowing repurposing between HPC and graphics... thoughts?


https://virtuallyvisual.wordpress.com/2016/02/04/mode-switching-on-nvidia-m60-grid-2-0-cards-modeswitch-tool/


https://virtuallyvisual.wordpress.com/2016/02/05/nvidia-m60-gpus-how-they-support-cuda-opencl-and-compute-in-graphics-mode/


The actual details and pricing of how that software product looks, is one for the business units. But from the technology side and engineering process; as well as the virtualisation hardware repurpose benefits. Although a new model of buying GPUs with software on top. I think makes sense. But would like to hear others thoughts.

Rachel

嗨雷切尔，
我个人的想法是，基于K2的解决方案仍然是主流方式，这也是一种非常好的方式。
在我们看到更强大的CPU之前，K2提供了一个非常好且经济实惠的解决方案，具有广泛的部署方法选项，并在典型的2 CPU / 2 K2单元中实现CPU / GPU性能之间的良好平衡。
如果您想在谈话中包含真实客户，我认为很难将技术方面与业务角度分开，因为需要使用该技术的客户也需要以某种方式获取和维护产品。
不幸的是，今天，M6 / M60或“GRID 2.0”是我认为的“业务部门”，是K2的新主流替代产品，尽管在考虑到CPU性能通常无论如何首先出现时都没有意义
，由于新的许可模式（根据整个系统的成本计算），使得X个普通CAD用户所需的服务器单元的成本在5年内几乎翻倍。
根据我的经验，除少数情况外，这是不可行的。
M6 / M60专门推出的新功能进一步增强了这一信息，并且当K2宣布退役时，谣言正在流传。
我知道您早些时候在Citrix做了一项精湛的工作，将社区的想法传达给业务人员，从而有效地实现了积极的开发周期。
我一直希望NVIDIA现在会发生类似的情况:-)或许有一个声明表明K1 / K2将得到支持和购买，直到2020年左右，一个不需要许可证的替代品将在
未来的某些方面，或者目前的价格点会降低。
我认为许可模式本身很有意义，如果定价大约是目前的20％，并且对XenApp这样的共享桌面有完整的支持。



以下为原文

Hi Rachel,

My personal thoughts are that K2 based solutions are still the mainstream way to go and that it is also a very good way. Until we see more powerful CPUs the K2 offer a very good and affordable solution with a broad option of deployment methods and a good balance between CPU/GPU performance in a typical 2 CPU/2 K2 unit.

If you want to include real-life customers in the conversation it is in my opinion difficult to separate the technical aspects from the business perspective as customers who want to use the technology are required to somehow also acquire and maintain the products.

Unfortunately today, M6/M60, or "GRID 2.0", are, by "the business units" I assume, launched as the new mainstream replacement for K2, despite there being little sense in this when considering that CPU performance often cap out first anyhow, making the cost for the server units required for X number of normal CAD users almost double up over a 5 year period due to the new licensing model (that is calculated on the cost for the whole system). It is, based on my experience, not feasible except in a select few cases.

This message is further enhanced by new features being exclusively launched for M6/M60 and rumors are floating around on when the K2 will be announced as retired.

I know you have done a superb work earlier at Citrix in channeling the community thoughts back to the business people and in-so-doing effectively enabling a positive development cycle. I keep my hopes up that something similar will happen now at NVIDIA :-) Perhaps a statement that K1/K2 will be supported and purchasable until say 2020 or so, a replacement that does not require licenses at that premium cost level will be announced at some point in the future, or that the current price points will be lowered. I think the licensing model in itself makes a lot of sense, if the pricing was about 20% of currently and there was intact support for shared desktops such as with XenApp.