Closed xiongyan21 closed 6 years ago
Deleted.
When "noio" and "direct" are removed from the input, the calculation is much more faster, around 4600 seconds saved.
It sounds like you have enough memory to store all the integrals or your filesystem is fast relative to computing integrals.
Deleted.
I don't have any NVMe SSDs but my understanding is they are very fast relative to spinning disk storage...
In any case, the NWChem log file indicates a good fraction of the integrals are cached:
#quartets = 2.049D+08 #integrals = 4.494D+09 #direct = 24.7% #cached = 75.3%
This job likely doesn't need anywhere near 7 GB of GA memory, by the way. You might try increasing the stack to 10G and reducing the GA allocation to 3 GB to see if that allows all the integrals to be cached.
Deleted.
Deleted
Deleted.