内部网格GPU概念证明

它们需求量很大并且仍然很昂贵，因为在某种程度上，不涉及许可费用。
为什么不尝试使用基本的Quadro卡作为PoC，只需在以Windows Server作为主机的直通模式下运行它？



以下为原文

They are in high demand and remain expensive because, in part, there are no licensing fees involved. Why not try a basic Quadro card as a PoC, and just run it in passthrough mode with a Windows Server as your host?

托比亚斯，你好！
再次感谢您的输入。
我会尝试Quadro，我有它，但它不支持vGPU。



以下为原文

Hello yet again, Tobias! Thanks again for your input.

I would try a Quadro, which I have available, but it doesn't support vGPU.

为什么您的客户对基于云的GPU实例不信任？
Azure，AWS，Fra.me，Cloudalize，3XS以及许多其他人对基于云的GPU实例没有任何问题？
你的客户不喜欢什么？
至于今天测试Kepler GPU。
我会救你的麻烦......不要打扰。
现在使用现有技术绝对浪费时间。
您的客户将不会了解性能，因为新技术真的要好得多。
开普勒已经被两次超越，当沃尔特击中GRID时，这将是三代过时的。
你要好好让客户等到你能以最低限度使用麦克斯韦并对其进行测试。
在Kepler上测试的另一个问题是，即使是GRID软件也会非常老旧，因为至少Maxwell需要运行5.x.
如果您的客户想要在此期间进行测试，那么现在就在Passthrough中运行Quadro。
这将让您了解在新的GPU架构上可以期待什么，如果他们想要同时测试多个会话，只需将其作为XenApp / RDS VM运行。
问候



以下为原文

Why does your client have no faith in Cloud based GPU instances? Azure, AWS, Fra.me, Cloudalize, 3XS and many many many others have no issues with Cloud based GPU instances? What is it your client doesn't like?

As for testing Kepler GPUs today. I'll save you the trouble ... Don't bother. Absolute waste of time with existing technologies out there now. Your client will learn nothing about performance as the newer technologies really are that much better. Kepler has already been superceeded twice and when Volta hits GRID, that'll be three generations out of date. You're far better asking your client to wait until you can use Maxwell at an absolute minimum and test on that. Another problem with testing on Kepler is that even the GRID software will be really old, as you need at least Maxwell to be running 5.x.

If your client wants to test in the meantime, then just run Quadro in Passthrough for now. That'll give you an idea of what you can expect on the newer GPU architectures, and if they want to test multiple sessions simultaneously on it, just run it as a XenApp / RDS VM.

Regards