Attention: For safety and airflow purposes, if you remove
parts from the system, you must ensure that the PCIe tailstock fillers are present.
As part of the GPU replacement, the cold plate is removed. Each GPU must have thermal interface
material (TIM) between the GPU and the cold plate. When a GPU is replaced, each of the TIMs that
share the coldplate must also be replaced. Ensure that you have spare GPU TIMs available before
proceeding.
Procedure
Attach the electrostatic discharge (ESD) wrist strap.
The ESD wrist strap must be connected to an unpainted metal surface until the service
procedure is completed, and if applicable, until the service access cover is replaced.
Attention:
Attach an electrostatic discharge (ESD)
wrist strap to the front ESD jack, to the rear ESD jack, or to an unpainted metal surface of your
hardware to prevent the electrostatic discharge from damaging your hardware.
When you use an ESD wrist strap, follow all
electrical safety procedures. An ESD wrist strap is used for static control. It does not increase or
decrease your risk of receiving electric shock when using or working on electrical
equipment.
If you do not have an ESD wrist strap, just prior
to removing the product from ESD packaging and installing or replacing hardware, touch an unpainted
metal surface of the system for a minimum of 5 seconds. If at any point in
this service process you move away from the system, it is important to again discharge yourself by
touching an unpainted metal surface for at least 5 seconds before you continue with the service
process.
If you are removing multiple GPUs, label the GPUs, noting their positions and
orientations.
The system has 2 or 3 GPUs under a cold plate, depending on the system's configuration.
Remove the air baffle next to the GPU you are replacing.
Remove the cold plate for the GPU you are replacing.
Loosen the cold plate retainer screw from each of the 2 or 3 GPUs and swing each retainer open
as shown in Figure 1.
Attention: Do not unscrew the four spring-loaded screws that attach the aluminum heat
cover to the GPU.
Lift the cold plate (A) off the GPUs that you are removing and attach
the cold plate to the service hook (B) as shown in Figure 2.
For the GPU you are removing, remove the eight screws shown in Figure 3.
Attention: Do not unscrew the four spring-loaded screws that attach the aluminum heat
cover to the GPU.
Remove the GPU from the system backplane.
Be careful of the pins on the bottom of the GPUs to avoid damaging the pins and the GPU. See
Figure 4.