OpenPOWER Summit North America 2020

In this talk, the Accelerated Big Data Systems group of Delft University of Technology shares experiences on the POWER9 OpenCAPI platform. A key goal of the group is to make programming accelerators (and specifically FPGAs) easier, more portable and more efficient. To this end, ABS has contributed several open source tools and technologies to the community, and created multiple demonstrators that apply these tools in big data analytics and genomics application domains. The presentation provides an overview of the ABS toolchain, the new tools that were developed, and show results of the use cases that were created.

Continue the conversation in Slack

OpenCAPI Acceleration Framework, abbreviated as OC-Accel, is a platform allowing programmers and computer engineers to quickly create FPGA-based acceleration solutions. Both acceleration action's software and hardware parts coherently share the server host memory data through OpenCAPI interface. Designers can thus quickly design an accelerator and benefit from the bandwidth, latency, coherency and programability advantages of OpenCAPI. Applications can be written in Python facilitating technology discovery.

Continue the conversation in Slack

Macromolecular crystallography is an experimental technique allowing to explore 3D atomic structure of proteins, used by academics for research in biology and by pharmaceutical companies in rational drug design. While up to now development of the technique was limited by scientific instruments performance, recently computing performance becomes a key limitation. In his presentation Filip will present a computing challenge to handle 18 GB/s data stream coming from the new X-ray detector. He will show his experiences in applying conventional hardware for the task and why this attempt failed. He will then present how IC 922 server with OpenCAPI enabled FPGA boards allowed to build a sustainable and scalable solution for high speed data acquisition. Finally, he will give a perspective, how the advancement in hardware development will enable better science by users of the Swiss Light Source.

Continue the conversation in Slack

One of the research topics of the Accelerated Big Data Systems group at TU Delft is transparent acceleration of Spark SQL programs on FPGA. This talk discusses our design, the modifications that were necessary in Spark, FPGA programming tools that were developed, and a fully integrated proof-of-concept on an OpenCAPI platform.

Continue the conversation in Slack

In this talk Dr Pinto will introduce ThymesisFlow, a hardware/software open framework based on OpenCAPI for implementing off-tray peripherals disaggregation via a software defined interconnect. Peripherals can be borrowed from a remote server and interfaced to the local processor, in the same spirit an off-chip peripheral is connected to the processor. ThymesisFlow is based on the OpenCAPI 3.0 specification and relies on a custom MAC that interfaces to a low-latency network. Each ThymesisFlow instance serves as bridge between the local coherence domain of a POWER processor and a remote counterpart, mastering unmodified OpenCAPI transactions towards the correct remote coherent domains based on the target address. During this talk Dr. Pinto will drive the audience through the hardware/software design of a main system memory disaggregation prototype using ThymesisFlow, its full stack evaluation on a set of cloud workloads, as well as a live demonstration on two IBM AC922 POWER9 servers.

Continue the conversation in Slack

Concluding the OpenCAPI track there will be two additional lightning talks followed by the described BoF moderated by Allan Cantle.

FPGA Accelerators including Hybrid Memory Subsystems - BittWare - Craig Petrie

Applications that can benefit from OMI enabled FPGA Acceleration - Maxeler  - Oskar Mencer

As the computing industry shifts towards a Disaggregated, Heterogeneous, Data Centric World, we've seen a myriad of Fabric Technologies being proposed and developed in what has commonly been referred to as Fabric Wars. This BoF reviews the key attributes of OpenCAPI, which is the only Fabric in production that's focused on lowest latency and high bandwidth direct connections with Memory, Accelerators & other Systems. The primary aim of the BoF is to both educate and to debate the merits & shortcomings of OpenCAPI when compared to alternate fabrics that typically have a broader scope. Is OpenCAPI a bus standard for the HPC world where only the highest performance will suffice and can such an open technology significantly help in democratizing the High Performance Computing Community once and for all?

Continue the conversation in Slack