Tensorflow2.2.2 and NVidia gpus, GPU_Load 1%

[roark52]

Have recently upgraded to Tensorflow2.2.2, and now the GPUload parameter does not rise above 1-3% while running machine learning models.
GPU Clock climbs to 1515MHz, but no GPU load increase.
Windows performance manager has never worked with my GPUs. But TechPowerUp has always been very reliable, until this upgrade.
Has anyone else seen this? Is this a GPU performance failure? Or is the latest Tensorflow / NVidia Cupti not giving GPU_Load info to the TechPowerUp app?
Tensorflow2 and NVidia Cupti seem a bit unstable, my concern is that GPUs are not being utilised, only CPU. Thx
(Anaconda 3, Python 3.8, Dual RTX2080 with SLlink)

 

[W1zzard]

GPU_Load info to the TechPowerUp app?

GPU-Z gets the GPU load number from the display driver, not from the running application.

Since GPU load is low since you upgraded to TensorFlow 2.2.2 I suspect there's some issue preventing it from using the GPU.

Is your GPU temperature increasing at all? (if the GPU is properly used, temps must be higher)