Workshop: IA^3 2026 — 16th Workshop on Irregular Applications: Architectures and Algorithms
Authors: Antonino Tumeo, Andrew Chien
Abstract: Making graph algorithms fast and scalable on traditional architectures and GPUs often involves breaking them into large, coarse-grained computational blocks that access similar regions of memory. We show that graph algorithms, which have tremendous vertex- and edge-level parallelism, can be scaled as fine-grained programs on suitable architectures. Specifically, we investigate graph algorithms on the UpDown architecture and show how this many-threaded, cache-less architecture enables simple, fine-grained parallel graph algorithms to be both fast and scalable. Single-node UpDown performance is 28x (geomean) over an H200 GPU, and that same program scales to 1000’s of nodes. We evaluate the system three ways: simulation, analytical projection models, and selected validation studies at scale (simulation of 16,384 UpDown nodes, using 256 supercomputer nodes). These results show the UpDown system can achieve 680K giga-traversed edges per second (GTEPS) for PageRank and 989K GTEPS BFS, exceeding prior results significantly. The realism of UpDown models has been validated by commercial activity at Chicago UpDown Computing (www.chupdown.com).Beyond performance, the insights for programming are significant. Our measurements suggest that not only is programming difficult – for distributed memory and GPU-SIMT – it is unproductive. That is, the resulting programs are not performant. for example, we have measured up to 5x higher total memory traffic for GPU implementations compared with UpDown. And the trend is for increasing UpDown advantage as algorithms grow in complexity suggest that kernel and linear algebra library based approaches face significant challenges.
Back to IA^3 2026 — 16th Workshop on Irregular Applications: Architectures and Algorithms Archive Listing
Back to Full Workshop Archive Listing