Topic for Discussion Sessions

Architectures and Algorithms for Accelerators and Specialized Machines

Specialized machines and accelerators have received considerable attention in the past, although technological and economical issues have typically favored general-purpose solutions.

The situation has evolved, in the last decade or so, due to the very significant investments worldwide in computing infrastructures for Artificial Intelligence and, more specifically, for Machine Learning.  In fact, the computations in these areas tend to be very rich in dense as well as sparse linear algebra, making specialized accelerators an attractive avenue to achieve top performance.

The current VLSI roadmap, aiming at 0.2 nm feature size around 2036, and progress toward wafer-scale integration (WSI) make the scenario for the next decade quite interesting.  Continuous re-evaluation of architectural organizations and algorithmic approaches will be essential to achieve the best exploitation of the technological potential.

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

1. Technology outlook in the next decade (conventional and non-conventional)

2. Evolution of ASICs vs. GPUs vs. FPGAs

3. Computational requirements of key application areas (AI, ML, Neuroscience, Biology, Medicine, Climate, Energy, ...)

4. Models of computation for accelerators

5. Models for analysis or architectures

6. Accelerators in supercomputers

7. Graph vs. Matrix computation

8. Low/variable precision arithmetic

9. Quantum accelerators