Driver建议

嗨巴里
目前可用的驱动程序是367.64 / 369.71，因此您只需进行一些修改。
GRID驱动程序本身不太可能对您的PCoIP丢失负责。
至于保持它们是最新的，较新的驱动程序通常包括功能和性能更新，错误修复，兼容性更新，并提供比旧驱动程序更好的稳定性。
它们通常也会更安全。
我个人总是尝试运行最新的驱动程序包，但是像任何更新一样，对于驱动程序更是如此，我在将其引入生产平台之前对其进行评估。
使用GRID，在更新生产平台之前，您显然有2个要评估的驱动程序（主机和VM）。
如果您找不到特定应用程序供应商的任何驱动程序建议，那么他们可能没有任何或只是不想提出这些建议。
我对此的建议是针对最新的，但显然要确保在将它们引入您的平台之前进行彻底的测试。
GRID驱动程序更新计划的一个关键考虑因素是，与典型的Windows更新不同，您必须将主机驱动程序与VM驱动程序配对才能完成升级，如果您有许多支持GPU的主机，这实际上可以成为一个
必须定期进行大量工作并且可能会变得有点痛苦，特别是如果您从更新的驱动程序中获得的收益很少，或者因为它们不包含任何与您相关的错误修复或您可以看到的改进。
我想这仅仅取决于您如何管理您的基础架构以及您拥有多少备用容量，以及升级过程的简单程度以及您的团队对BAU任务的忙碌程度。
为了缓解在预生产测试期间未获取的任何问题，您通常不会同时更新整个GPU产品，因为如果要解决任何问题，回滚不是一个非常快速的过程。
然后，您必须考虑哪些VM驻留在哪些主机上，因为主机/ VM驱动程序将不同。
但是，我尝试将更新过程考虑在内，因为我在设计平台时更容易维护，但是必须保持驱动程序同步可能会有点痛苦，现在VMware已经宣布了它的vGPU HA能力
保持司机同步将变得更加重要！
...
问候
本



以下为原文

Hi Barry

The current drivers available for you are 367.64 / 369.71, so you are a few revisions behind.

The GRID drivers themselves are unlikely to be responsible for your PCoIP dropouts.

As for keeping them up to date, newer drivers will usually include functionality and performance updates, bug fixes, compatibility updates and offer better stability than older drivers. They'll typically be more secure as well. I personally always try to run the latest driver packages, but like any update and even more so with drivers, I evaluate it before introducing it into a production platform. With GRID, you obviously have 2 drivers to evaluate (Host and VM) before you update your production platform.

If you can't find any driver recommendations from the specific application vendors, then the chances are they either don't have any or just don't want to make those recommendations. My advice on that, aim for the most recent ones, but obviously make sure you throughly test before introducing them into your platform.

A key consideration to your GRID driver update schedule, is that unlike typical Windows updates, you have to pair the Host driver with the VM driver in order to complete the upgrade, and if you have lots of GPU enabled Hosts, this can actually become a large piece of work to have to undertake regularly and can become a bit of a pain, especially if you stand to gain very little from the updated drivers either becasue they do not contain any bug fixes relevent to you or improvements that you can see. I guess it just depends how you manage your infrastructure and how much spare capacity you have as to how simple the upgrade process is and also how busy your team are with BAU tasks.

To mitigate any issues that have not been picked up during pre-production testing, you typically wouldn't update your entire GPU estate at the same time, as the roll back isn't a very quick process if you were to encouter any issues. You then have to consider which VMs reside on which Hosts as the Host / VM drivers will differ. However, I try and factor the update process in when I'm designing platforms so they're easier to maintain, but having to keep the drivers in sync can be a bit of a pain, and now that VMware has announced its vGPU HA ability, keeping drivers in sync will become even more important! ...

Regards

Ben