Nvidia NVML does not support non-Tesla product very well. Problems are known with mobile cards and even Quadro cards. (Reported as RFE to Nvidia as Bug ID 2417658.)
Anyway, this can lead to cuda_memtest throwing an [NVML] Error: Not supported (in nvmlDeviceGetSerial) exception which we should catch.