hpconfig

hpconfig provides class-based specifications for different HPC hardware and environments.

Hardware defintions inherit from the CPU_NODE and GPU_NODE classes defined in hpconfig.utils.classes. These are used in cluster descriptions of HPC configurations, referred to as 'specs. hpconfig` contains a number of different pre-defined specs, such as the Pawsey supercomputing facility , and some planned infrastructures, such as the Square Kilometre Science Data Processors.

Defining a hpconfig spec

An example class may be constructed as follows:

from hpconfig.utils.constants import SI
from hpconfig.utils.classes import CPU_NODE, GPU_NODE


class MyStartingSpec:
   IntelXeonSpec = CPU_NODE(
       name="IntelXeonSpec",
       cores=4,
       flops_per_cycle=8,
       ncycles=2.0 * SI.giga,
       bandwidth=1877 * SI.mega
   )  # ~240 giga flops

# ...
# ...

A spec consists of one or more hardware device (CPU or GPU) as class attributes. These are then collated in the __init__() function in the architecture definition:

def __init__(self):

    self.name = 'MyStartingCluster'
    memory_per_cpu_node = 64
    memory_per_gpu_node = 32
    self.architecture = {
        'cpu': {
            self.XeonIvyBridge: 50,
            self.XeonSandyBridge: 100,
        },
        'gpu': {
            self.NvidiaKepler: 64
        }
    }