1 of 11

Running Molecular Dynamic Simulation Workloads on Oracle Cloud Infrastructure

Gloria Lee – Cloud Engineer at Oracle

2 of 11

Benefits of Running HPC Workloads on Cloud

Scale easily
Eliminate capacity issues
Able to burst
Decrease capital and operational costs
Eliminate need for new hardware
Access to latest hardware

But first, let’s talk about the benefits of running any HPC workload, like molecular dynamic simulations, on the cloud.

HPC is one of the last workloads to move over to the cloud, but there are several benefits to doing so.

With physical hardware, you’re unable to scale up and down easily, due to the cost and time constraint of getting these machines. But by being on the cloud, you’re able to spin up clusters immediately, scale freely, and ensure you have capacity.

This is also great for bursting workloads as well.

In addition, you’re able to keep costs low by only paying for what you use and eliminating the need to buy new hardware.

Plus, you’re able to get access to the latest and greatest hardware from your cloud providers

Even with these benefits, people still need to see firsthand how their workloads perform on the cloud.

So, today I will be showing the performance of NVIDIA Clara Parabricks, GROMACS, and NAMD on OCI

3 of 11

Reference Architecture

BM.GPU3.8 – 8x V100 NVIDIA Tensor Core

BM.GPU4.8 – 8x A100 NVIDIA Tensor Core

*Bare Metal = BM

4 of 11

NVIDIA CLARA PARBARICKS�Germline Pipeline

Data Used – NA12878 (link)

48G ERR194147_1.fastq.gz�49G ERR194147_2.fastq.gz

Parabricks Version

BM3.8 – v3.1.7�BM4.8 – va2.0

https://docs.nvidia.com/clara/parabricks/v3.5/text/germline_pipeline.html

5 of 11

30% increase

6 of 11

28% decrease

BM.GPU3.8 - $2.95 GPU/hr

BM.GPU4.8 - $3.05 GPU/hr

7 of 11

GROMACS

Data Used – benchPEP (link)�benchPEP - 12 M atoms, peptides in water, 2 fs time step, all bonds constrained

GROMACS Version 2020.1

8 of 11

38% increase

9 of 11

NAMD

Data Used – ApoA1 (link)

ApoA1 - 92,224 atoms, periodic, PME

NAMD Version 2.13

https://www.ks.uiuc.edu/Research/namd/2.13/features.html - Contributed by Marcelo Melo, Rafael Bernardi, and Till Rudack

10 of 11

32% increase

11 of 11

Thank you!