cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
1441
Views
0
Helpful
5
Replies

Nvidia Grid K2 GPU card not detected by UCS Manager

taborda
Level 1
Level 1

Hello,

I installed a new UCS C240M4 with one Nvidia Grid K2 GPU card, this server is connected to a UCS Manager 3.1(2d). After UCS Manager discovered the server all hardware was detected but not the GPU card.

I already created in UCS Manager a new service profile to the server, enabling the BIOS setting Memory Mapped IO Above 4GB, and powered on the server, but still the GPU card is not detected.

I checked if the riser version is a compatible one, and moved the card from riser 1 slot 2 to riser 2 slot 5, without success.

Does anyone know any guidelines to troubleshoot this issue?

5 Replies 5

Kirk J
Cisco Employee
Cisco Employee

Greetings.

Do you have the required power aux cable hooked up to the card as well?

From 240 m4 spec sheet:

Select GPU Power Cables
Whenever you select a K1/K2/K40 or AMD GPU for this server, you must also select one power
cable for each GPU selected. The available GPU power cables are listed in Table 23

Table 23 Available GPU Power Cables

Product ID (PID) PID Description 

UCSC-GPUCBL-240M4      C240 M4 GPU Power Cable
UCS-300WK-240AMD       300 Watt AMD Cable and Kit for UCS C240 M4 Rack Server

  • Caveats
     NVIDIA GPUs can support only less than 1 TB of total memory in the server. Do not install
    more than fourteen 64-GB DIMMs when using an NVIDIA GPU card in this server. Please note,
    this limitation does not apply to Pascal series GPU (P100)
     NVIDIA GRID K1 and K2 GPUs can be mixed. No other GPU mixing is allowed.
     If you order a K80 GPU, note the following:
    — You cannot mix the K80 with any other GPU
    — You must select 2 CPUs for the server
     Slot 5 on riser card 2 is the required slot for the first GPU.
     Slot 2 on riser card 1 is the secondary slot for a second GPU. The riser card 1 options that
    are compatible with GPUs are:
    — Riser card 1 option A (UCSC-PCI-1A-240M4)

 

Thanks,

Kirk...

Hi Kirk,
Thanks for your answer.
Yes, power cable UCSC-GPUCBL-240M4 is connected between the riser and the card.
I tried the card on slot 5 which as the riser UCSC-PCI-1A-240M4, and also in slot 2 wich has the riser UCSC-PCI-2-C240M4, but the card is not recognized in UCS Manager.

 

Not sure if the card is OK, do you know if there is any tool to check it?

 

Best regards,

 

Can you bring this up in stand alone mode, (disconnect 10Gb links from FIs) and hook up some sort of 1gb cat5 to the cimc mgmt/dedicated port?  Default CIMC web interface user/pass should be admin/password

I am curious if the CIMC, inventory, PCIE adaptors list shows the GPU card...

The other caveats seem to be less than 1 TB of RAM, and required both CPUs.

Can you confirm we have less than 1 TB of RAM, and both CPUs are installed?

 

Thanks,

Kirk...

Hi Kirk,

Many thanks for your answer.

The server has only 128GB of RAM and two CPUs Intel(R) Xeon(R) E5-2640 v4.

 

I'll try to do the test in stand alone mode, and also upgrade the firmware to the latest one by CIMC.

If it works, can be a solution.

 

As soon I have news I'll update you.

Best regards

Hi Kirk,
Unfortunately changing to standalone didn't solve the issue.
I checked again if the board is placed correctly on riser 2 slot 5, I cleaned the server configuration to factory default and after upgrading the firmware to 3.0(3f) the PCI inventory only shows the VIC and Ethernet boards.

it seems that the GPU is damaged.

Many thanks for your help.
Best regards,

Review Cisco Networking for a $25 gift card