If (top != self) { window.location = 'about:blank'; }
NASA Logo, National Aeronautics and Space Administration
High-End Computing Program

+ Home > User Information > Policies > Standard Billing Units

STANDARD BILLING UNITS

The High-End Computing Capability (HECC) project and the NASA Center for Climate Simulation (NCCS) use a common Standard Billing Unit (SBU) for allocating and tracking computing time usage.

The NASA SBU is a way of standardizing work across dissimilar architectures. Representative codes are run on each architecture and their run times compared to the baseline values, currently calculated for a Pleiades Broadwell node. A Conversion Factor is then calculated to convert the individual system’s accounting values to SBUs. The goal is that users will be charged the same number of SBUs for a given workload regardless of the environment they choose to run in.

Each architecture at HECC and NCCS determines the Minimum Allocatable Unit (MAU). This is the smallest unit of the hardware resources that the scheduling software will allocate to a job. For Pleiades, Electra, and Merope at HECC and Discover at NCCS, the MAU is a node; for Endeavor (HECC), the MAU is 8 cores.

The SBU Conversion Factor represents the difference in efficiency of that particular architecture in performing the baseline workload on its MAU compared to the Pleiades baseline MAU. The larger the conversion factor, the faster the architecture completes the work.

In October 2018, a new set of SBU Conversion Factors went into effect, which uses runtime on a Pleiades Broadwell node as a baseline. In the tables below, SBU1 refers to the retired SBU rate using Westmere as a baseline, and SBU2 refers to the new and current SBU rate using Broadwell as a baseline.

This new SBU rate replaces the retired SBU rate established in May 2011, when the baseline values were calculated for a Pleiades Westmere node.

The SBU Conversion Factors

The hisotrical and current SBU Conversion Factors for HECC and NCCS are below.

Note: 1 SBU1 is worth 0.254 SBU2.

HECC SBU Conversion Factors

HECC SBU Conversion Factors

System Pleiades Pleiades Pleiades Pleiades Pleiades Pleiades
Architecture SGI/HPE
ICE X
SGI/HPE ICE X SGI/HPE ICE X SGI/HPE ICE X SGI/HPE
ICE X
SGI/HPE
ICE X
Processor Broadwell Haswell Ivy Bridge Sandy Bridge Sandy Bridge+GPU Skylake+GPU
Cores per MAU 28 24 20 16 16 CPU;
1 K40 GPU
18 CPU;
4 V100 GPU
SBU2 (Current) Conversion Factor 1.00 0.80 0.66 0.47 0.94 9.82
SBU1 Conversion (Retired) Factor 4.04 3.34 2.52 1.82 3.64 N/A

 

HECC SBU Conversion Factors

HECC SBU Conversion Factors

System Pleiades Electra Electra Aitken Aitken Endeavour
Architecture SGI/HPE
ICE X
SGI/HPE
E-Cell 
SGI/HPE
ICE X
HPE
E-Cell 9000
HPE Apollo HPE
Superdome Flex
Processor Cascade Lake+GPU Skylake Broadwell Cascade Lake Rome Cascade Lake
Cores per MAU 48 CPU; 8 V100 GPU 40 28 40 128 28
SBU2 (Current) Conversion Factor 15.55 1.59 1.00 1.64 4.06 1.31
SBU1 (Retired) Conversion Factor N/A 6.43 4.04 N/A N/A N/A

 

NCCS SBU Conversion Factors

NCCS SBU Conversion Factors

System Discover Discover Discover Discover
Architecture Supermicro TwinPro Supermicro TwinPro Supermicro FatTwin SGI Rackable
Processor AMD Milan EPYC Cascade Lake Skylake Haswell (retired July 2023)
Cores per MAU 126 usable 48 40 28
SBU2 (Current) Conversion Factor 3.81 (preliminary) 1.77 1.61 0.81
SBU1 (Retired) Conversion Factor N/A N/A N/A 3.09

 

HECC updates the cost per SBU each October—the beginning of the federal fiscal year (FY)—based on the total SBUs available at that time and the full cost of HECC investment for that fiscal year. The historical cost per SBU for HECC is listed below:

Historical Cost per HECC SBU2

Historical Cost per HECC SBU2

FY2024 $0.29
FY2023 $0.30
FY2022 $0.44
FY2021 $0.47
FY2020 $0.47
FY2019 $0.55
FY2018 $0.65
FY2017 $0.97
FY2016 $1.05
FY2015 $1.13
FY2014 $1.54
FY2013 $2.26
FY2012 $3.23

 

The Charging Formula

The formula for SBUs charged to a job is as follows:

SBUs charged = number of MAUs x number of wall clock hours x SBU Conversion Factor

Given that an SBU is a representation of an amount of useful work, a job should be charged similar amounts of SBUs whether it is running on a more efficient or less efficient system.

Requesting Allocations and Checking Usage

When using the HEC RMS online submission systems to apply for computing time, principal investigators should continue to estimate the number of runs on each processor type of a system, the number of processors per run, and the wall clock hours per run. Calculators within RMS total the processor-hours from these estimates and convert the processor-hours to the common SBUs.

Use acct_ytd for HECC projects and allocation_check at NCCS for checking the amounts allocated and used.

 

USA.gov NASA Logo - nasa.gov