Max M10 boards per server

NVIDIA Virtual GPU Forums Virtual GPU General Discussion
#1

Can someone explain the general limitation of 2 M10’s per server. Is it a BIOS issue, RAM, CPU, what is the limitation. If I have a server that can fit 6 GPU’s per server what is the rational behind only supporting 2 per server?

https://www.nvidia.com/object/grid-certified-servers.html

We are using a SuperMicro SYS-2028GR-TRHT, which have enough slots for 6 GPU’s and currently running 4 K2’s per server without issue.
https://www.supermicro.com/products/system/2u/2028/sys-2028gr-trht.cfm

looking to move to the M10-2Q profile and would like to get better density than 32 users per server (32GB per M10 / 2GB per VM) x 2 M10 boards.

#2

Hi,

M10 has 4GPUs on a single board. Issue is the number of available PCI devices per physical host. 3 M10s = 12 GPUs!!!

regards

Simon

#3

I would think the number of available slots, power, and cooling would be the main limiting factors. Currently,you also cannot/should not mix GPU models (like an M10 and an M60) on the same server.

#4

[b]Don’t forget the M10 and M60 are both Double-Height GPU’s.

I do not believe there is a server on the market that will support 6 of them.[/b]

We use Dell R730’s (They support 2 Cards) and we have a new Dell R740 and it supports 3 Cards.

As suggested do NOT try and mix GPU Cards in a single server.

#5

Most server support 2 x M10 GPU cards only due to running out of PCI Express lanes (PCIe x16 needed).