Skip to main content

Upgrading R730 with NVIDIA K1 GRID Card

I recently ordered a couple of Dell R730 servers and then got a subsequent request to add a little graphics horsepower for our VDI environment in the form of some K1 GRID cards. Turns out, the process to add these suckers in to an existing server that wasn't specifically built out for them has a few catches - you can't just drop them in to the server and take off. It's easy to do, but there isn't much official documentation from Dell on it, so here's a quick guide:

First, the requirements:

  • Compatible R720 or R730 series server
  • 1100W PSUs
  • Compatible processor - there's a specific list of procs that are certified for use depending on your chassis, which relates to power consumption with the GPUs.

Start by removing the shroud cover:


The first problem we have to fix is the airflow issue - the factory setup for an R730 uses heat sinks that capture all of the air under the shroud and will restrict the amount of fresh air going to the GPUs.:


If you ordered for K1 cards from Dell, you should also have received an "upgrade kit" consisting of a new low-profile heat sink and a 6-pin PCIe power cable. Loosen the four screws on the CPU heat sink (I'm installing two cards here and I have two CPUs, so I'll be replacing both), wiggle them a bit to get them to detach, and carefully clean all of the grey heat sink compound off the CPU:

Make sure you also replace any empty DIMM slots with blanks. Install your new low-profile heat sinks, replace the shroud, and you'll have an unrestricted path for airflow back to the PCIe risers (I removed a fan here):

Before you put in the card(s), connect the 6-pin power cable to the appropriate slot on the riser. If you are installing two cards, connect each to their respective riser - do not try to use one cable/riser to power both, even though they do have dual connectors on the end. Route the power cables through the notch in the shroud (marked in red below). Now remove your expansion slot blanks and carefully drop in your GRID cards. They should just barely clear any obstructions to get in there - it's a tight fit.


Install your second card if applicable and then make sure you press the buttons marked in blue to deploy the supports for the card(s):


If you already have 1100W power supplies equipped, you are all done. If you don't, you'll want to upgrade before the systems go in to production. I've used the K1 cards with 750W, but the system will complain about power usage and may not have enough juice at peak consumption or during a failure of one PSU.

Now just power on your server and follow the appropriate guide from NVIDIA for installing drivers: http://www.nvidia.com/object/virtual-gpus.html. Good luck!

Comments

Popular posts from this blog

Outlook Credential Prompt When Opening Exchange 2013 Public Folder

After completing an Exchange 2007 > 2013 migration recently, I was left with one issue that was preventing us from stamping the project as a roaring success and moving on:

Outlook 2013 users were sometimes receiving a single pop-up prompt for credentials whenever they opened the Public Folder (we have only one). One. Single. Prompt.

Google was frustratingly unhelpful because searching for "outlook prompts for username and password when opening public folders" or something similar just resulted in a lot of folks who were always getting a pop-up that wouldn't go away. It was usually caused by an authentication failure of some sort.

However, we were in a different boat - Users got the prompt once when they first launchedOutlook and opened their public folders, but after entering it they could continue - authentication worked. Next time they logged in to their PC, it would happen again. Not a show stopper, but it definitely generated its share of support calls.

Repairing Mailbox Corruption in Exchange 2010

I recently got through recovering an SBS 2011 server after Active Directory face-planted in the middle of a workday. When I say recover, I mean I repeated the entire migration, using a cleaned up secondary DC - it was a fun weekend (expect another post about that experience). Although I thought we were in the clear, I got a call from the client about 24 hours after we had verified everything was working. He indicated that his iPhone had suddenly stopped receiving mail in the inbox (calendar, contacts, sent items were still fine) and throws up an error after spinning in circles for a few minutes that it "cannot connect to mail server".

SCEP Policy Update Troubleshooting

Because I'm a glutton for punishment, I recently started rolling out System Center Configuration Manager 2012 R2 SP1 and System Center Endpoint Protection across our VDI environment. There are always some considerations to be made in a pooled desktop / gold image type environment when loading software that uniquely identifies devices, but lucky for me SCCM/SCEP handled this just fine without any tweaking. However, there were some nuances to how SCEP policies are applied that caused some serious hair-pulling before I spotted the issues.