Help running a Tesla P40
Help running a Tesla P40
Tesla P40 reached its maximum output on an HP Z620 workstation setup. The GPU used was a 24GB NVIDIA Tesla P40 with headless operation. Display side featured a Quadro P400 capable of managing all visual outputs. The system ran on Xeon E5-2620 v2 CPUs with PCIe 3.0 support. Memory consisted of 128GB DDR3 ECC RAM, which provided reliable performance. The BIOS was updated to a non-OEM version supporting WDDM. Drivers were re-installed to unlock compatibility for the P40 (version 30.0.15.1179).
Power delivery came from the official HP Z620 PSU via a standard dual-cable configuration. Adapters were used to connect the GPU to the Y-aggregator, then to a single 8-pin Tesla input. The hardware itself was original OEM except for the RAM upgrade from 16GB to 128GB.
Technical challenges emerged: despite software indicating a 250W capacity, the card operated at low speeds (Gen 1, 2.5GT/s). Performance remained constrained even under heavy workloads like Blender or Octane. The PCIe link speed was stuck at Gen 1 (2.5GT/s), while HWiNFO64 showed a higher theoretical rate (PCIe v3.0 x16). Power consumption stayed within a hard limit of 43–50W during intensive tasks. Clock speeds were fixed at 202MHz (TCC) or 405MHz (WDDM), with no increase under load.
Throttling was minimal—NVIDIA diagnostics reported no active issues, yet the hardware failed to scale. Steps taken included forcing Gen 3 PCIe in BIOS and adjusting registry settings to enable WDDM support. Driver mode switched between TCC and WDDM, but behavior stayed consistent. Physical maintenance involved cleaning the interface and reseating the card. ECC was disabled via Nvidia Control Panel, which slightly improved rendering times but increased power draw.
Thermal performance was acceptable; cooling wasn’t a concern. The main hurdle was achieving true WDDM compatibility. I’m seeking guidance on how to modify the BIOS or registry further to unlock higher performance without compromising stability.
Would you like to proceed with the conversation in the current thread?