Incorrect cost estimate for google deeplearning vm in GCP

The cost estimator left out the cost of the GPU when I selected the T4 option. It estimated around $64 when the actual cost is over $500 a month with that configuration. The K80 is around $300 a month, it should work fine since the training isn’t bottle-necked by GPU. I was able to start training with an instance configured with the K80 but it hasn’t been running long enough to know it it’s any slower.