Different NBANDS value from two computer resources

Message

pnita · #1 Post by **pnita** » Wed Apr 25, 2012 9:09 pm

Dear All,

I have calculated the DOS on the clean vanadium surface using two computer resources (school's cluster VS TeraGrid). When I plotted the LDOS of d-orbital, I found dissimilarity between two results I obtained and I figured out that the NBANDS value is the reason. I didn't explicitly set NBANDS in the INCAR so I assumed that the value should be default.

According to VASP manual, the NBANDS for the non-spinpolarized calculation is NELECT/2 + NIONS/2. In my system, NELECT = 140 and NIONS = 28. Therefore, I should expect the NBANDS is 140/2 + 28/2 = 84. However, the NBANDS value from two calculations are

School's cluster --> 88
TeraGrid --> 128

So, here is my questions
1. Why is the real NBANDS value different from the expected one?
2. Why do two computer clusters give different NBANDS value?

Both calculations were done with vasp.4.6.36.

Thank you very much.

alex · #2 Post by **alex** » Thu Apr 26, 2012 9:53 am

As far as I know the numer is a multiple of either NPAR or the number of processor cores?!

Cheers,

alex

pnita · #3 Post by **pnita** » Thu Apr 26, 2012 7:57 pm

Dear Alex,

Thank you for your response. So does it partly has anything to do with NELECT and NIONS? Or it's relied only with NPAR or number of processors?

Thank you.

#4 Post by **admin** » Fri Apr 27, 2012 12:33 pm

NBANDS= N x #CPUs overrules this setting

pnita · #5 Post by **pnita** » Mon Apr 30, 2012 12:10 am

Dear Admin,

Thank you for the clarification. By the way, what does N refer to here in NBANDS = N x #CPUs?

Thank you.

gevolo · #6 Post by **gevolo** » Sun Jun 10, 2012 4:33 pm

N should be just an integer, so your two clusters have probably diff number of CPUs. In order NBANDS to be a multiple of the cores used, VASP modifies its value. If you check the core number of each case, it should make sense.