The Xilinx Alveo U55C data center accelerator card and a new standards-based, API-driven clustering solution for deploying FPGAs at huge scale were unveiled recently at the SC21 supercomputing conference. High-performance computing (HPC) and database workloads would benefit from the Alveo U55C accelerator, which would scale easily thanks to the Xilinx HPC clustering solution.
The new Alveo U55C card, designed specifically for HPC and big data applications, is the company’s most powerful Alveo accelerator card ever, with the greatest compute density and HBM capacity in the Alveo accelerator range. It allows users with large-scale computing workloads to build sophisticated FPGA-based HPC clustering utilizing their current data center infrastructure and network thanks to the new Xilinx RoCE v2-based clustering solution.
“Scaling out Alveo compute capabilities to target HPC workloads is now easier, more efficient and more powerful than ever,” said Salil Raje, Executive VP and General Manager, Data Center Group, Xilinx. “Architecturally, FPGA-based accelerators like Alveo cards provide the highest performance at the lowest cost for many compute-intensive workloads. By introducing a standards-based methodology that enables the creation of Alveo HPC clusters using a customer’s existing infrastructure and network, we’re delivering those key advantages at massive scale to any data center. This is a major leap forward for even broader adoption of Alveo and adaptive computing throughout the data center.”
Built for HPC and Big Data Applications
The Alveo U55C card would combine a number of critical capabilities required by today’s HPC workloads. It has the greatest performance-per-watt in the Alveo portfolio, with increased parallelism of data pipelines, improved memory management, and optimized data transfer across the pipeline. With a modest 150W maximum power, the Alveo U55C card has a single-slot full height, half length (FHHL) form factor.
When compared to its predecessor, the dual-slot Alveo U280 card, it has a higher compute density and doubles the HBM2 memory to 16GB. For dense Alveo accelerator-based clusters, the U55C delivers more computing in a smaller form factor. It’s designed for high-density streaming data, high-IO math, and massive compute workloads like big data analytics and AI applications that demand scale-out.
The API-driven clustering solution utilizes RoCE v2 and data center bridging, together with 200 Gbps capacity, to provide an Alveo network that would compete in performance and latency with InfiniBand networks while avoiding vendor lock-in. HPC developers may use the Xilinx Vitis unified software platform to scale out Alveo data pipelining thanks to MPI interoperability. It’s now feasible to scale out over hundreds of Alveo cards using current open standards and frameworks, agnostic of server platforms or network infrastructure, and with shared workloads and memory.
Through high-level programmability of both the application and cluster using the Vitis platform, software developers and data scientists may harness the benefits of Alveo and adaptive computing. Xilinx has put a lot of money on the Vitis development platform and tools pipeline in order to make adaptive computing more accessible to software developers and data scientists who don’t know much about hardware.
Developers can build domain solutions using specific APIs and libraries, or use Xilinx software development kits to easily accelerate key HPC workloads within an existing data center, using major AI frameworks like Pytorch and Tensorflow, as well as high-level programming languages like C, C++, and Python.
The Alveo U55C card is presently available through Xilinx approved distributors and on Xilinx.com. It’s also accessible for private previews at certain colocation data centers, as well as public cloud-based FPGA-as-a-Service providers. Private previews of clustering are currently available, with wide availability planned in the second quarter of next year.