Supercomputing was a bit different this year, but it was exciting to see the developments across the world for high-performance computing (HPC). Oracle Cloud is honored to partner with Fugaku (#1 on the TOP500 list) as their cloud HPC partner. This partnership, along with the work that we’re doing at OCTOPUS, helps us demonstrate to our enterprise HPC customers how to run production HPC workloads in the cloud.
We’ve been busy this year at Oracle Cloud HPC. Enterprise customers like Nissan are moving entire workloads to our cloud.
Our seamless integration with on-premises schedulers and file systems make the transition to the cloud transparent for end users. This became possible only when we brought bare metal to the public cloud, and now we’re seeing customers bringing 10, 20, even 30 thousand core workloads onto clusters that run 24/7 in the cloud, and they save millions.
No charge for egress and 100-Gbps MPLS connections make data transfer at the scale of enterprise HPC workloads possible. Architectures for electronic design automation (EDA), like the one in the following figure, enable a groundbreaking approach for chip design that isn’t possible on any other cloud.
Figure 1: EDA architecture on Oracle Cloud
The performance and price of our HPC cloud continues to dominate our discussions with customers, analysts, and journalists. Before we even started doing HPC, the promise that the cloud brought to HPC was massive scale. That’s true on Oracle Cloud, but for enterprise customers, price is frequently just as important. We’ve learned that providing the best overall price-performance is the only way to get customers to move production workloads. Oracle is making this possible by providing the lowest cost and highest performing HPC instances on the cloud with the largest physical HPC footprint and lowest storage costs, too.
Figure 2: Oracle Cloud HPC instances beat other clouds in price-performance
Oracle’s flat pricing means that HPC on Oracle Cloud is the same price anywhere that you run it.
Figure 3: Cost comparison to run a common HPC workload
Oracle Cloud HPC scales by using bare metal remote direct memory access (RDMA) networking, so you don’t need to change your Message Passing Interface (MPI). You can run HPC jobs faster and more efficiently on Oracle Cloud than any other cloud.
Because of our storage performance, HPC file servers deployed on Oracle Cloud are the fastest and cheapest compared to other cloud providers.
Figure 4: Cost of common file servers in the cloud
The next generation is just as exciting. We started rolling out the next generation of HPC hardware. See what Intel, NVIDIA, and AMD have to say about how Oracle Cloud HPC provides excellent benefits.
Being the first to market with NVIDIA’s A100 GPU gave our artificial intelligence (AI) and machine learning (ML) customers a head start on building bigger and faster models. In the first quarter of 2021, we’re launching our Ice Lake HPC instances followed by our AMD Rome and— drum roll, please—ARM instances. These instances combine to give our customers flexibility and performance for their HPC workloads.
Our investments in HPC follow the needs of our customers. We don’t view HPC as a science project; HPC is the way that our customers make better products. Oracle’s investments in HPC are increasing, and I look forward to sharing what we have coming next year at ISC!
Join our other customers by spinning up an A100 GPU or running an EDA workload on Oracle Cloud. With US$500 in cloud credits, you can put us to the test! If you’re interested in trying our cloud, also check out Oracle Cloud Free Tier.
Here is what we’re doing at SC20:
Panel: Intel HPC Panel