Newer GPUs for VDI based on P100/V100?

Dear All!

Presumably in 2018 we will have to buy some new servers.
Gladly we would like do deliver 3D applications to all of our three rooms with Dell Wyse zero clients.
At the moment only our smallest room with 51 seats can run 3D apps by means of four Cisco C240 Servers with two Nvidia K2 cards each.

The other two rooms have 78 seats each, 156 seats together, which would mean that we at least would need twelve C240 Servers with 24 K2 cards…
That would become much too expensive.

At the Nvidia website I saw the newer P100 and V100 cards.
There also should be Mezzaine cards with these chips.
Can these little cards be used in Cisco B200 M4 blade servers?
If not: can PCIe versions of these cards be used in Cisco C240 servers?

How much would these cards cost?
How much would Nvidia software licenses cost?
How many cards/servers would we have to use?

Thank You very much in advance!

With kind regards,

Hi Christian,

as usual the answer is: “it depends”. What kind of workload do you use? You only mentioned 3D applications. Most important is the framebuffer usage. P100 doesn’t really make sense but you may consider P40 in the future.
Here the HCL:
Be aware that the latest Pascal boards will only be supported with the Cisco M5 hardware.

Now to be more detailed:
Price of the boards->Need to contact your OEM.
License costs: Explained here:
Depends on the edition, I would assume you need vWS license. See the document above for list prices.
Number of cards:
Example with C240 M5 and 2x P40. Assuming you would need P40-1Q profile you get 48 users on one host. So you would need 4 hosts to cover your <200 users. But be also aware that in most cases the CPU is the bottleneck. In the past there was no CPU available to deliver 48 users on a host with the required performance. Now there is the new Xeon Gold with 18x3Ghz so it might work.

You talked about rooms. Does this mean this is more or less for training purposes?