Your search did not match any results.
We suggest you try the following to help find what you’re looking for:
By Sasha Banks-Louie | September 2020
Oracle’s new high-performance computing on Oracle Cloud Infrastructure is helping solve some the world’s toughest problems, from helping with modeling to combat diseases to simulating car crashes to help reduce car-crash fatalities. By running complex simulations and analyzing mathematical models in the cloud, Oracle opens the door for more scientists, engineers, and analysts to access high-performance computing, since it costs a fraction of running HPC using on-premises servers.
As Oracle upgrades its first-generation cloud HPC offering with faster and more powerful processors, “we expect to offer 30% better per-core performance for crash simulations, computational fluid dynamics, and electronic design automation workloads,” said Clay Magouyrk, Oracle executive vice president of Oracle Cloud Infrastructure, during a 25-minute Oracle Live event on September 22.
Oracle’s new level of performance for such processing-intensive workloads is now possible in part because of Oracle’s decades-long partnerships with top innovators in the microprocessor world. The Oracle Live event featured insights from the CEOs of AMD, Ampere, Intel, and NVIDIA, who demonstrated how they’re working with Oracle and pushing high-performance computing to new limits. Here are their insights from the event:
The 2019 launch of Oracle Exadata X8M with Intel Optane persistent memory gave “Oracle’s customers 10 times faster analytics and two and a half times more throughput” compared to Oracle’s first-generation HPC, said Intel CEO Bob Swan during the Oracle Live event.
Today, customers are turning to Oracle and Intel to help them run performance-intensive HPC jobs on demand, instead of having to buy fixed, on-premises capacity. “Oracle’s bare metal offering combined with Intel’s optimized Xeon silicon allows Oracle applications to handle far more users, operate on larger active datasets, and respond in real time to complex queries,” Swan said.
“We expect to offer 30% better per-core performance for crash simulations, computational fluid dynamics, and electronic design automation workloads.”
As businesses create ever-larger machine learning models to solve their toughest problems, the need for more compute, memory, and networking resources has kick-started a revolution in faster and more powerful graphics processor designs.
NVIDIA’s new A100 GPUs with Mellanox Direct Connect are now available with Oracle’s new HPC offering, helping customers get “accelerated computing” from extremely large networking clusters of up to 512 GPUs.
But accelerated computing requires people to have immediate access to huge libraries of domain-specific software, explained Jensen Huang, founder and CEO of NVIDIA, during the Oracle Live event. For example, NVIDIA’s CUDA-X AI provides software to build deep learning apps for a range of use cases. Whether it’s crash simulations, genomics processing, or deep learning and analytics, each process requires its own domain-specific software stack.
“We’ve created this registry in the cloud that has all of these really complex software stacks, perfectly tuned, updated, and all containerized, so that all you have to do is pull it into the OCI instance and spool up your machine,” Huang said.
Putting all of this technology together to help build Oracle’s “cluster-scale” high-performance computing platform lets companies of all sizes run massive workloads in the cloud. Whether those workloads involve simulating drug molecules or running machine learning algorithms for a human resource application’s chatbot, high-performance computing is now ready for the masses.
“We’re going to put this technology into the hands of enterprise customers all over the world,” Huang said. “It’s a great next adventure for us, and it’s really great to do it with Oracle.”
Through its partnership with Ampere, Oracle will offer the chipmaker’s newly released Altra server processors—80-core CPUs designed specifically for ARM-based workloads, which involve compute-intensive, hyperscale processing.
By early 2021, Oracle will begin coupling these CPUs on a single server, giving customers up to 160 cores of ARM performance at 3.3 GHz per core, a terabyte of memory, and 100 gigabits per second of bandwidth. “Our 80-core approach with a single core per instance allows for a level of isolation and security that people are excited about,” said Renee James, founder, chairman, and CEO of Ampere Computing. James added that these new CPUs also allow for a level of density and scalability that’s oriented around how people sell, use, and develop cloud instances today.
With its bare metal servers, virtual machines, containers, block storage, and a cluster network of more than 20,000 cores, Oracle’s ultra-low latency clocks in at just two microseconds.
To help customers meet their data processing and system memory requirements, Oracle is also going to start offering AMD’s new Milan processors in 2021. The new CPU shape will be offered at the same price as Oracle’s current AMD EPYC E3 shape, but aims to offer much higher performance.
The Milan processor, as AMD’s new E4 instance for Oracle Cloud Infrastructure, “is designed for the cloud and will provide unmatched computing performance and business value to power the world’s most important cloud workloads and services,” said Lisa Su, president and CEO of AMD.