Effective October 1, 2018- For Midway2
The charge for service units (SUs) is calculated as the max of the measurements from all types of resources used for the analysis: CPU-core, memory, and GPU.
SU_Charged = MAX [SU_CPU, SU_Memory, SU_GPU]
Please see below for calculations for SU_CPU, SU_Memory, SU_GPU, and SU_Midway.
The base value of SU is defined as using 1 core for 1 hour on Midway.
SU_CPU = # of requested CPU-cores
SU_Memory = Memory requested by job / Memory available per CPU
Note that Memory available per CPU is going to vary with every partion: Broadwl (56 GB / 28 CPUs) = 2 GB and Caslake (180 GB / 48 CPUs = 3.75 GB). Memory available per CPU will be different for AMD nodes, bigmem nodes, GM4 & Beagle3 nodes. Please contact help@rcc.uchicago.edu if you have any questions.
SU_GPU = # of requested GPUs + 1.5 * max(SU_CPU,SU_Memory)
Note that the SU calculation for jobs completed on the Midway3 cluster are subject to a different algorithm due to the increased computational speed of the resources. The calculation of SUs for a given job on a given cluster can be monitored and discussed with the Research Computing Center. Please contact help@rcc.uchicago.edu if you have any questions.