Tesla K40C Compatibility

Accelerated Computing CUDA CUDA Setup and Installation
#1

Hello,

I just built my first workstation for a cancer research project that I’m planning on using with a Tesla K40c, unfortunately there weren’t many resources I could find out there for hardware compatibility. I got everything setup and ready to install my Tesla but when I plugged the card in all I got was a blank screens, not even BIOS, no audible post codes from my motherboard but I did get lights and fans. I was hoping those of you with more knowledge and experience can look over my build at let me know if there are any compatibility issues?

MB: Asus Prime X570-P
CPU: AMD Ryzen 7 2700X
Ram: 4 x 16gb DDR4 3000Mhz
GPU: PNY Geforce GT 710 (for display output)
PSU: EVGA 550W

I’m leaning towards a PSU issue or an incompatibility with the GT 710, I’m keeping my fingers crossed it’s not a CPU/MB issue since it’s not a work project and it’s the best I could afford at this time. I was also thinking if it’s a K40c issue I could switch to a Qaudro K6000 but since they are very similar I’m worried I’d spend the money on the Qaudro and have the same issues with that. Thank you in advance for any help.

#2

My guess would be an incompatibility between the motherboard (system BIOS) and the K40c GPU.

Check for any BIOS updates for your motherboard. If not (or if there is no change after updating system BIOS), there may not be anything you can do to get it to work.

I think its unlikely to be a PSU issue. I’m not saying your PSU is ideal (although at first glance it seems probably OK, wattage-wise), I’m simply saying I don’t think it is the source of a no-boot problem, if the system boots correctly without the K40c. Make sure you have provided the proper power connections to the K40c, although getting that wrong should not prevent a system from booting.

If you do get it to work (i.e. boot) at some point, you may find that the GT 710 is not a good choice for a display GPU. If it is a Fermi variant of GT 710, you will not be able to use any CUDA version later than CUDA 8 with that setup.

https://www.techpowerup.com/gpu-specs/geforce-gt-710.c2614

However having a fermi GPU sitting alongside your Kepler K40c in and of itself is not going to prevent your system from booting.

An additional thing to check would be BIOS settings. You would need to do this when the K40c is not plugged in (obviously) so it will be tedious to do trial-and-error, but you may find BIOS settings that allow it to boot with the K40c. I’m not familiar with this ASUS BIOS directly, but the things I would be on the lookout for would be options to enable legacy behavior in the UEFI BIOS, an option to switch to legacy instead of UEFI mode, and/or options to enable modified addressing (e.g. enable mapping above 4G boundary). The K40c has a not-terribly-large PCI BAR compared to K40m, but it may still present an unusual mapping scenario for the BIOS.

Finally, make sure you have a K40c (has a fan) vs. a K40m (has no fan). If you have a K40m by chance, then this is a more difficult endeavor because:

  1. K40m has a huge (~16G) PCI BAR that has to be mapped by the BIOS (although even this usually does not present a no-boot scenario, just a resource mapping failure)
  2. K40m requires server flow-through cooling. It will not keep itself cool. Providing the necessary cooling in a non-server environment is usually fairly challenging.

https://devtalk.nvidia.com/default/topic/1039289/cuda-setup-and-installation/k40-setup-on-lenovo-p510/

#3

Hi Robert,

Thank you so much, that was hugely helpful! I looked over the BIOS documentation and identified a few settings that I can change. I’ll post back with my results once I give that a try.