Hacker News new | ask | show | jobs
by rincebrain 504 days ago
This always seemed to be a very deliberate design choice by them to avoid you being able to use their consumer cards headless versus paying them a large amount on the Quadro or DG cards, since the big problem we saw at $OLDJOB was always that you couldn't use CUDA on them headless.

At said $OLDJOB, we ended up soldering dummy VGA plugs that had resistors across the right pins when we wanted to experiment with building a low-power cluster of NVIDIA Ion boards and seeing how it competed with big cards. Ah, memories.
1 comments
alias_neo 504 days ago
Would bet that this is exactly why. I run Tesla GPUs in my server rack which don't even have display ports, but they run any OS just fine with the vGPU drivers, which Nvidia make an absolute pain to obtain.
link
rincebrain 503 days ago
The _very_ first gen of Tesla cards did have those headers on them, IIRC, and then successive ones had the headers on the board but not connected for another generation or so, IIRC.

You also used to be able to edit the PCI IDs for the drivers to get the Tesla ones to attach to consumer GPUs, but that stopped working at some point.

link