Extra local memory needed for incore

[Sibo Lin]

I am trying to do single point dft calculations on a rather large system (500 atoms). I get the calculation is very slow and the output says "Extra Local Memory (stack+heap) needed for incore: 7242 Mbytes". My current input file specifies "heap 200 mb stack 1000 mb global 2800 mb" and "direct" in the dft section (which I thought would override incore...). The node has 128 gb memory, and I'm running the calculation across all 24 cpus of the node. I chose the 4gb memory specification based on a recommendation to only use 75% of the total memory (128 /24 × 0.75 = 4). 

Do you have recommendations for how to shift the memory allocation to make this calculation run faster?

Thank you!

[Edoardo Aprà]

You need to remove the "direct" keyword to enable the incore algorithm when CD fitting is used in DFT. The memory will be used out of the sum of heap+stack. Therefore I would recommend decreasing global to something around 500 mb. However, you don't seem to have enough aggregate memory to use the incore algorithm based on the numbers you reported