Amber20 is available in the clusters of Minotauro and IQTC. The new realeased pmemd.cuda performs faster than in the older version, but we need to take into account the GPU that we use!
In summary, we have in Minotauro the K80, with average performance, but unavailable with 4 GPUs, but with shorter time of pending jobs in the queue. In the other hand, you have the GTX 1070 Ti and RTX 2080 Ti, both of them available with 4 GPUs, but the RTX performs the fastest of all, with 78 ns/day with 4 GPU. The problem is the large pending time of the jobs.
Choose wisely in your calculations!