UCloud, a rapidly growing Shanghai-based cloud service provider with 31 availability zones worldwide, recently released high performance computing (HPC) services in the cloud based on technology from Arm partners.
The new services, which became available in early January, are targeted at the research and education markets. For scientific research projects, complex data-intensive computing tasks such as simulation, data modeling, and rendering require vast amounts of computing power. This can be most easily provided by elastically scalable high-performance computing clusters with high-speed network fabric connections and massive data storage.
The UCloud Kwaijie Lite HPC Cloud instance runs on Ampere Computing’s Altra processor, which is based on Arm’s Neoverse N1 architecture. Taking advantage of 32 of vCPUs, Kwaijie Lite instance achieved a benchmark score of 20.77 GFLOP/s on the Phoronix Test Suite (PTS) for High Performance Conjugate Gradients (HPCG), according to UCloud. The result is 40 percent greater performance than equivalent Kwaijie cloud instances running on traditional processors.
Meanwhile, another newly launched UCloud HPC cloud instance, Kwaijie Cloud HPC bare-metal service, takes advantage of the Nvidia Bluefield Data Processing Units (DPU) for managing networking, security, and storage functions. Bluefield combines Nvidia ConnectX-5® Dx network adapter technology with Armv8 A72 CPUs. With the integration of Bluefield DPU into bare-metal services, infrastructure software such as Open vSwitch Kernel is offloaded from CPUs to DPUs. Customers who use the bare-metal service get seamless access to features such as NVGRE overlay virtual networks with high-bandwidth and low-latency performance.
The shift of HPC workloads from supercomputers inside university laboratories and corporate R&D centers is one of cloud computing’s more intriguing trends for 2021. Analysts predict HPC in the cloud services grow by 16.7% annually and reach $11.5 billion by 2026. (Brent Gorda, senior director of HPC at Arm, explored the trend in this blog post earlier this year.)
An overview of UCloud operations
For researchers, flexible computing capacity on call can also mean critical time savings. After experiencing bottlenecks on its local systems, scientists at a university in Harbin shifted its workload to UCloud’s bare-metal services. The time to complete the task dropped from an expected ten days to five.
Similarly, a group at a Shanghai university conducting fluid simulations with terabytes of data and enormous amounts of CPU, storage, and memory capacity reduced processing time from an estimated 20 days to a week. UCloud HPC services have also been employed to conduct material science simulations and life science research involving cryo-electron microscopy data.
Lanwei Technology, a SaaS-based engineering services firm, has managed to reduce its HPC costs by 30 percent while improving performance by 30 percent by shifting from a traditional deployment model to UCloud services.
“Scientific research institutions can flexibly configure according to fit their usage requirements. There is no need to queue during peak demand and they can avoid wasting resources when demand is low,” said UCloud CEO, Mr. Ji Xinhua. “Arm technology takes our HPC strategy one step further by reducing costs, increasing performance, and giving our customers a wider array of options at the same time.”
UCloud highlights the dynamism occurring in cloud market. Founded in 2012, UCloud serves over 10,000 enterprise customers through datacenters located in North America, Europe, South America, Africa, and Asia-Pacific. Its content delivery network encompasses 500 plus nodes offering 6.5 tbps of bandwidth. The company held an IPO in January 2020 and has a current market value of approximately $3 billion.
We look forward to collaborating with UCloud as it expands it customer offerings. For more, please visit the company’s website.
[CTAToken URL = "https://www.arm.com/solutions/infrastructure/high-performance-computing" target="_blank" text="Visit Arm HPC Solutions Page" class ="green"]