Xilinx®, the leader in adaptive computing, today announced the Alveo U55C accelerator card, the most powerful accelerator card for HPC and big data workloads, enabling powerful FPGA-based HPC clustering on existing data center infrastructure and networks.

Xilinx held a press conference on the 16th.

Nathan Chang, HPC product manager for Xilinx's Data Center Group, who was in charge of the press conference presentation, said that high performance computing (HPC) will face power consumption issues as the next challenge, and that typical HPC architectures will struggle to provide appropriate performance per watt, pointing out issues of wasted clock cycles, reduced workload, and increased power consumption.

To this end, Xilinx is introducing a new standard-based API-enabled clustering solution that enables large-scale FPGA deployment and the Alveo U55C (Alveo&trad) at the 'SC21 Supercomputing Conference'.e; U55C) data center accelerator card announced.

Purpose-built for HPC and big data workloads, the new Alveo U55C card is the most powerful Alveo accelerator card in the Xilinx Alveo accelerator portfolio, delivering the highest compute density and HBM capacity.

Additionally, a broad range of customers addressing large-scale compute workloads can leverage the new Xilinx RoCE v2-based clustering solution to implement powerful FPGA-based HPC clustering on their existing data center infrastructure and networks.

The Alveo U55C card is implemented in a single-slot Full Height, Half Length (FHHL) form factor with a low maximum power of 150W.

It offers superior compute density with 16GB of HBM2, twice that of its predecessor, the dual-slot Alveo U280 card. The U55C delivers greater compute performance in a smaller form factor, enabling the creation of high-density Alveo accelerator-based clusters.

It can solve big computing that requires scalability, such as big data analytics and AI applications, as well as mathematical problems involving high-density streaming data and high IO.

API-based clustering solutions leveraging RoCE v2 and data center bridging and 200Gbps bandwidth can deliver Alveo networks that are vendor-independent and offer InfiniBand-like performance and latency.

Additionally, HPC developers can scale their Alveo data pipelines out on the Xilinx Vitis™ unified software platform via integrated MPI.

Now you can scale across hundreds of Alveo cards, regardless of server platform and network infrastructure, and across shared workloads and memory, leveraging existing open standards and frameworks.

Software developers and data scientists can leverage the Vitis platform to leverage Alveo and adaptive computing benefits with high-level programmability for applications and clusters.

Xilinx has invested significantly in its Vitis development platform and tool flows to make adaptive computing more accessible to software developers and data scientists without hardware expertise.

Support for major AI frameworks such as Pytorch and Tensorflow, as well as high-level programming languages such as C, C++, and Python, enables developers to implement domain solutions using specific APIs and libraries, or easily accelerate key HPC workloads within existing data centers by leveraging Xilinx software development kits.
/> Nathan Chang, HPC product manager for Xilinx’s Data Center Group, said, “The new Xilinx HPC clustering solutions enable massive scale-out across existing customer infrastructures and networks, while enabling software developers and data scientists to leverage the benefits of Xilinx adaptive compute with high-level programmability of applications and clusters.”