I was pleasantly surprised to find out that Cisco and nVidia partnered to deliver a GPU that now fits into a blade server form factor. For those UCS customers out there, this was a much anticipated announcement. Since the inception of UCS, heavy graphic user customers had to rely on C-Series rack servers if they wanted to take advantage of GPU capabilities. This worked well, especially since UCS allows C-series to integrate into UCS Manager, allowing full service profile functionality. But, nevertheless you had UCS customers that had standardized on blades and thus was really longing for a blade based GPU.
Fast forward to the end of January, 2016. Cisco and nVidia partnered to deliver a new Telsa brand of GPU called the M6. This GPU plugs into the available B200 M4 adapter slot 2. Note that if you have a VIC or other device installed in slot 2, it must be removed. Their are some pre-reqs that must be followed:
- UCS B200 M4 is the only supported server (thus far – maybe someday that list will grow).
- UCS Firmware needs to be at 3.1.1e
- If going to use in a ESXi environment it needs to be at v6.0 or greater.
- VMware Hardware needs to be at version 11
Its worth noting that the Tesla M6 GPU requires GRID software for VDI. This is a change from the older rack based K1/K2 GPU cards. GRID 2.0 leverages a licensing server that then checks in/out the VDI licenses. The licenses are based on Virtual PC, Workstation user, and Workstation +. Below is a nice summary from nVidia on the differences.
I was fortunate to get to test one of the new M6 GPUs in a lab. Since this is brand new technology it was a challenge finding deployment guides, but in the end was able to get the card up and functioning without much effort. The nVidia Quick Start Guide was a great help.
In summary it was as simple as:
- Install Telsa M6 GPU
- Upgrade Firmware on UCS to 3.1.1e
- Change GPU mode from Compute to Graphics (*this is a very important step)
- Install VMware VIB Host Driver
- Attach vGPU profile to VM
- Install nVidia driver to VM
- Configure the VM for the GRID vGPU license
I have just started the performance testing and will report back the findings. I have found that third party “freeware” benchmark tests work pretty good for both OpenGL and DirectX.
This new blade GPU form factor will allow customers to expand into the GPU market with a choice of rack or blade based options. For customers in Oil & Gas, Energy, Remote Sensing/GIS, etc, this is exciting times.