X

The latest cloud infrastructure announcements, technical solutions, and enterprise cloud insights.

Running computer-aided engineering on Oracle Cloud Infrastructure with price and performance in mind

Gloria Lee
Cloud Engineer

Best performance for your high-performance computing workloads

If you want the best performance for your high-performance computing (HPC) CPU-based workloads, BM.HPC2.36 with cluster networking is the shape to choose. The servers are connected directly to the RDMA switch with no hypervisor, no virtualization, and no jitter, resulting in a high-bandwidth, ultra-low-latency network that can go as low as 1.5 µs. This type of networking is ideal for complex computer-aided engineering (CAE) workloads, such as computational fluid dynamics (CFD), crash simulation, finite elemental analysis (FEA), and weather research and forecasting (WRF), because of their tightly coupled nature. With BM.HPC2.36 and RDMA networking, you can achieve performance comparable or better to on-premises environments. For more information, see Optimize Your HPC with Oracle Cloud on Intel.

Accounting for cost

Sometimes performance isn’t the most important factor, but instead cost. Here, our VM.Standard.E3.Flex.64 comes in. With our flex shapes, you’re billed separately for the OCPU ($0.025/OCPU) and memory resources ($0.0015 per GB). For all our virtual machines (VMs), per-second billing is enabled, compared to that of BM.HPC2.36, which bills per-second with a one-hour minimum. Depending on your memory requirements, you can optimize your costs by lowering the memory to a minimum of 1 GB and 1 OCPU. But low costs are not the only benefits of our flex compute shapes. For more details, see Harnessing the AMD flexible virtual machines for high-performance computing on Oracle Cloud Infrastructure.

To demonstrate the price performance you can achieve with VM.Flex at the lowest memory limit of 1 GB and 1 OCPU, we ran an OpenFOAM 42M model using 2500 iterations and compared it against our HPC shape.  We spun up an instance pool of VM.Flex and a Cluster Network of BM.HPC2.36, then ran the simulation across 36, 72, 144, 288 and 576 cores on each. BM.HPC2.36 exceeded performance by approximately 30–60% because of the highly performant intel 6154 chip and linear scaling from the RDMA network.

When evaluating the cost of the workload, VM.Flex was 40% cheaper on average and consistently cheaper as the cores increased for the workload. The following graph shows the cost of each workload at several core counts, based on the time it took the workload to complete.

Regardless of whether you want the best performance or the best price in the cloud, OCI is the cloud for you and your HPC workloads.

Get started today on a solution that fits your workloads

Start your 30-day free trial and get access to a wide range of Oracle Cloud services for 30 days, including BM.HPC2.36 and VM.Standard.E3.Flex shapes. If you’d like more guidance, get hands-on experience with our cloud on our LiveLabs platform.

Learn more about running HPC workloads on Oracle Cloud Infrastructure with the following documentation:

Join the discussion

Comments ( 3 )
  • Pang Tuesday, December 15, 2020
    Awesome work Gloria!
  • Candice Tuesday, December 15, 2020
    Wow! Thanks for the analysis! Good to know
  • Christopher Lee Tuesday, December 15, 2020
    Great article! Provides practical and economical use case for high-performance workload. As someone in the industry, I can say that Oracle has helped me scale up my business performance with integration of Oracle’s infrastructure. I will continue to follow your blog postings, great work!
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.Captcha