What we work on

One Problem - Two Views

Algorithms, Peak Performance & Optimization

Any kind of hardware has to be evaluated in combination with a certain kind of problem statement and a suitable algorithm to solve the problem. We are interested in the maximum possible performance of such hardware-algorithm combinations. We actively develop numerical kernels and port them to new hardware in order to allow for comparison of runtime costs (such as time-to-solution, performance/$, performance/watt) across different hardware concepts. Through publication, engagement at conferences and community outreach, we hope to inform the community on highly efficient implementations as well as dead ends of our research.

Programming Models, Libraries & Usability

Often not peak performance alone is key to hardware adoption. Instead, usability plays an important role for users of any kind of hardware stack. We investigate the usability of emerging hardware and report on the good examples as well as the shortcomings of individual concepts. We actively (co)develop libraries to ease the usage of hardware for our lab and for other investigators. Through open-source software, panel discussions and research collaborations we give back to the wider HPC and AI community.

Some Example Hardware Concepts We Investigate

SpiNNaker2 from SpiNNcloud

A brain-inspired neuromorphic chip that promises highly performant computation with extremely low energy consumption on sparse networks or other sparse workloads.

Wafer-Scale Engine 3 from Cerebras

A massively parallel AI chip that is made out of a whole wafer. It features a massive number of cores with small, but very fast local memory and an on-chip interconnect.

Photonic Chips

We cooperate with a number of analog photonic chip makers and are currently in the preparation phase to investigate their respective chips.

RISC-V Chips

We cooperate with a number of RISC-V HPC chip makers and are currently in the preparation phase to investigate their respective chips.