Technology assessment for high performance computing
SURF aims to get a good understanding of the performance of various scientific applications for high performance computing and artificial intelligence that can be used by specific research communities. We evaluate these workloads on state-of-the-art hardware architectures within the SURF Open Innovation Lab (SOIL).
We focus on performance analysis, collaboration and open sharing of results. It is an objective within SOIL to make benchmarking and analysis around future hardware reproducible, open and more engaging for the SURF community. As an HPC center, we support researchers to solve challenges associated with their applications on HPC infrastructures. In this particular initiative, we focus on evaluating, assessing, and benchmarking modern and emerging computing architectures for several application domains such as computational science, machine learning and data visualisation.
The ultimate goal of this work is not just to get the raw computational performance numbers but moreover capturing the usability of the hardware from the researcher’s perspective. A powerful and robust software ecosystem is as important as the hardware itself and therefore we focus a part of our effort in understanding and benchmarking software libraries and backends as well.
- Proper insight in future hardware and software technology enables the Dutch HPC community to develop the best innovative HPC services, tools and applications.
- Development of expertise in the area of performance analysis and evaluation of various hardware architectures for computing workloads supports proper selection and effective use of HPC resources
- Providing access to the latest compute resources enables collaborative assessment and knowledge development.
- Performing standardized assessments of new hardware and software for HPC.
- Open sharing of assessment results.
- Organizing workshops to share knowledge among the SURF institutions.
- Alignment with European initiatives in the area of modern processor design, high performance computing, high performance data analytics and AI, such as PRACE.
- Collaboration with vendors and technology providers, so new hardware and software can be assessed even before it becomes generally available.
- Open for collaboration with any interested SURF institution.
- Consistent and reproducible benchmarking effort.
- Close collaboration with SURF institutions
- Providing access to experimental compute resources, with easy access and provisioning for external parties.
Access to experimental systems
SURF currently has the following set of nodes in an experimental cluster up and running:
- 2 X ARM Machines
- 1 X FPGA (U250- Xilinx) + NVIDIA GPU + Intel CPU
- AMD (EPYC (Naples) + MI50) : 32 core CPU + 4 x MI50 (GPU)
- AMD (EPYC (Naples): 1P ,1x 32 core CPU
- AMD (EPYC (Naples): 2P ,1x 64 core CPU
- 1 X FPGA (U250 from Xilinx) + NVIDIA GPU + AMD EPYC: to be operational after October
SURF is not striving for increasing the hardware counts in our lab aimlessly but would like to play a role in which researchers and ICT experts from SURF institutions can seamlessly get access to experimental systems in order to adapt their codes, perform benchmarks and follow a co-design approach.
Current collaborations and joint projects:
- TU Delft, UvA, and SURF are jointly working on a project exploring heterogeneous computing with field-programmable gate arrays (FPGAs)
- UvA, SURF and AMD will work together to make HPC codes available on AMD GPU hardware, do performance analysis and benchmarking.
Articles and seminars:
- Blog: Building Affordable and Programmable Exascale Capable Supercomputers
- Seminar: Technology Assessment overview
SURF is very open to collaborations on this topic. Please contact us at: