Our core tenet has always been the absolute best price performance, giving customers the benefits of on-premises with all the flexibility of public cloud. This goal was evident when last September, when we announced a roadmap of Compute hardware instances, including instances based on Intel’s 3rd Generation Xeon scalable processors, which are code named “Ice Lake.”
Today, we’re excited to announce availability of these instances in limited preview, with worldwide general availability on April 28, 2021. The launch of these new instances is the latest step in Oracle’s commitment to providing enterprise-grade capability for compute-intensive and high-performance computing (HPC) workloads. Our portfolio of Compute instances gives our customers the benefits of cutting-edge hardware innovation and the flexibility to run each application on the most suitable platform.
These new instances provide performance gains of up to 42% over the prior generation of HPC instances. Moreover, these Intel-powered instances are the first to feature our flexible Compute capabilities, which enable you to choose any granular number of cores and amount of memory to suit your workload’s needs. You’re no longer limited to fixed virtual machine (VM) sizes with 4, 8, or 16 cores and static memory ratios per core. You can deploy a five-core VM with 25 GB of RAM, if that’s what you need, and pay only for what you use.
These instances are ideal for workloads that require the highest single threaded performance, such as batch processing, video encoding, electronic design automation (EDA), distributed analytics, data science, and AI Inference workloads. Workloads like computational fluid dynamics (CFD), used for aerodynamics, see performance improvements of up to 42%, and electronic-design analysis and Monte Carlo workloads see upwards of 40% gains. Best of all, all this performance comes at the same price per core-hour as previous HPC instances at $0.075 per core hour, regardless of the region you deploy in. This performance gain at flat pricing translates into big cost savings for our customers.
|VM.Optimized3.Flex||1–18||1–256||Up to 1 PB of block storage||Up to 50 Gbps||
|BM.Optimized3.36||36||512||3.2 TB NVMe||100 Gbps RDMA||$0.075/core/hr|
You can launch the new bare metal instances in Oracle Cluster Networking with RDMA support and network latencies as low as 1.5 microseconds across single clusters as large as 20,000 cores. That makes them perfect for traditional HPC workloads that require bare metal, low-latency performance across a large cluster to run large-scale message passing interface (MPI) workloads, such as engineering simulations, computational fluid dynamics, DNA sequencing, and reservoir simulations. The bare metal shape also includes local NVMe SSDs for high-speed, low-latency storage access.
“For over a decade, we have worked with Oracle to deliver solutions for a wide variety of workloads from high performance in the cloud to serving needs for machine learning, simulation, and modeling,” said Navin Shenoy, executive vice president of Data Platforms Group, Intel. “It is wonderful to see how Oracle is achieving greater performance in the cloud on their bare metal instances with the 3rd Gen Xeon and how their customers will be able to leverage Intel’s latest processors on Oracle’s powerful high-performance cloud.”
Oracle Cloud performance derives from Oracle’s rigorous approach to HPC, running the truest bare metal offering with the most-performant, lowest-latency HPC network of any major cloud. This combination allows tightly coupled HPC applications, whose performance hinges on the application’s ability to communicate between nodes with minimal delay, to scale to larger core counts without losing efficiency, saving time and money.
On the other hand, Intel’s 3rd Gen Xeon incorporates many innovations. With the new core micro-architecture, it has larger L1 and L2 cache than previous generations. PCIe Gen 4 improves bandwidth and latency to NVMe storage volumes and accelerators. This generation has more memory lanes and better memory bandwidth, and customers can use AVX-512 instructions. Many HPC applications respond well to these improvements.
Having access to the latest and greatest, with a wide selection of hardware to match each workload, is central to Oracle Cloud’s value proposition, along with providing enterprise-grade production HPC capability and the most cost-effective pricing in the public cloud. These 3rd Gen Intel Xeon processor-based instances make true on that promise, offering our customers a step up in cost-performance of up to 42% over the previous generation of Intel HPC instances.
To get started with these instances, visit the OCI management Console, CLI, and SDKs. To learn more about deploying these Compute or HPC workloads, visit our Getting Started GitHub page or our Architecture Center. As we get closer to general availability, look for more in-depth performance analysis and customer stories later in the month.