M60 for something other than virtual, and vendor specific versions

Virtual GPU Forums General Discussion
1

Hi There,

I’ve been working with one of my trusted vendors to get a demo server to be used with VMware VDI (horizon view). He has finally sourced me a reasonable spec server (Lenovo) but was unable to get the M60 to go with the demo. He has indicated that his price is north of $8,000 for a Lenovo sanctioned M60.

I’ve found on eBay and Amazon M60 cards, active, both HP branded and what looks like nVidia OEM M60’s. I’d include links but I’m not sure they are permitted on this forum. They are all running $4000.

As this is just a demo project, and I’m not likely to get buy in on VDI until this time next year, if I go ahead and buy the M60, it would be nice to have an alternative home for it. So there are two questions here:

  1. Is anyone using HP M60/Dell M60/OEM M60 in their server that is not HP/Lenovo/Dell for VDI? I can’t imagine this is like hard drives where Dell/HP check for specific firmware.

  2. Once I’ve completed my demo, and returned the server, and am sitting with this shiny M60 card, are there any desktop applications (non-virtual) that would utilise an M60?

In supportive additional detail for #2, I have an in house GIS guy that runs an older Quadro M4000 for ESRI ArcGIS Desktop and Pro (10.whateverismostrecent). Could I shove this M60 into his workstation, and in addition to his M4000, would ESRI take advantage of it? Essentially I’m looking for a use case to justify the $4000 spend.

Thanks in advance,

Tim

2

Hi Tim

I’m going to use Cisco as an example, as that’s who we use for our platforms.

  1. Certainly when you purchase the GPUs through Cisco, they do have a little modification to enable UCSM to be able to see / manage them. This is not a performance modification, it just allows the system to recognize that it is a Cisco supplied GPU. Whether this is the same for Lenovo / HP / Dell etc, I’m unsure. However, this doesn’t bind them to a specific vendors chassis and they can be used in other chassis.

If someone were to purchase an M60 outside of Cisco, but installed it in one of their servers, it will absolutely work, but you won’t be able to manage it through UCSM and therefore lose some of the benefits, and you would need to ensure it was supported separately, as although the configuration will be supported, it (the actual GPU) won’t be supported by Cisco.

When Cisco supplied GPUs are installed in Cisco servers, this configures the server in a slightly different way to non-Cisco supplied GPUs. I’m not sure how much detail I’m allowed to say on here, so in the interests of my NDA, I’ll leave it there. Suffice it to say there are definite differences between the GPU variants, and for the best overall performance, management and support, you should use the vendor supplied versions when using their servers.

So in answer to your question, yes, you’ll be able to use those M60 GPUs in any supported server, but also yes, there are potential differences (as mentioned, certainly with Cisco) between those and the vendor supplied ones. A slightly long winded answer, but I thought it worthwhile letting the community know a bit more detail.

  1. Even in a Bare-Metal install, don’t forget you need to license the M60 to be able to use it properly. However, that’s not even your biggest issue with using it like that, the M60 is designed to sit within a server due to the airflow required to cool it. If you add it into a PC / Workstation, you risk seriously damaging it due to the lack of adequate cooling. My advice, sell it after you’ve completed your POC, and either buy a Quadro to replace the M4000 or save the money for your next project.

As you currently have a server without a GPU installed, be sure that when you purchase the M60, you also check whether you need the GPU enablement kit for your Lenovo server. Typically, this consists of some low profile CPU heat-syncs and the GPU power cable. When you buy the GPUs from the vendor when purchasing a server, this is usually highlighted so you don’t forget it, and yes, you do need it.

Regarding the cost of these (and other) GPUs, Nvidia actually sells them at a very good price. What happens after that is obviously outside of their control. There are vendors such as SuperMicro that offer less expensive Nvidia GPU packages that you may want to check out.

Regards

Ben