Topic for Discussion Sessions

Architectures and Algorithms for Accelerators and Specialized Machines

Specialized machines and accelerators have long received considerable attention, although technological and economic factors have typically favored general-purpose solutions.

The situation has evolved over the past decade or so, largely because of major worldwide investments in computing infrastructure for artificial intelligence and, more specifically, machine learning.  Indeed, computations in these areas often involve substantial dense and sparse linear algebra, making specialized accelerators an attractive path to achieving top performance.

The current VLSI roadmap, targeting feature sizes around 0.2 nm by approximately 2036, together with progress toward wafer-scale integration, makes the outlook for the next decade particularly interesting.  Continued reassessment of architectural organizations and algorithmic approaches will be essential to fully exploit the potential of emerging technologies.

In this context, the following is a (non-exhaustive) list of topics for discussion at ScalPerf'26.

1. Technology outlook for the next decade, including conventional and non-conventional technologies

2. Evolution of ASICs, GPUs, and FPGAs

3. Computational requirements of key application areas, including AI, ML, neuroscience, biology, medicine, climate, and energy

4. Models of computation for accelerators

5. Models for analyzing accelerator architectures

6. Accelerators in supercomputers

7. Graph-based versus matrix-based computation

8. Low- and variable-precision arithmetic

9. Quantum accelerators