Research

Paper

AI LLM March 24, 2026

Numerical Kernels on a Spatial Accelerator: A Study of Tenstorrent Wormhole

Authors

Maya Taylor, Carl Pearson, Luc Berger-Vergiat, Giovanni Long, Jan Ciesko

Abstract

As AI accelerators gain prominence, their potential for traditional scientific computing workloads remains unclear. This paper explores Tenstorrent's Wormhole architecture, a spatial computing platform designed for neural network acceleration, by implementing three numerical kernels and composing them into a conjugate gradient solver. We present architecture-specific optimizations for sparse numerical algorithms, evaluate their performance against Nvidia GPUs, and expose both challenges and opportunities in porting numerical methods to spatial architectures. Our results demonstrate that AI accelerators merit consideration for workloads traditionally dominated by CPUs and GPUs, and more work should be invested in understanding the capabilities of these architectures and making them accessible to the scientific computing community.

Metadata

arXiv ID: 2603.23343
Provider: ARXIV
Primary Category: cs.PF
Published: 2026-03-24
Fetched: 2026-03-25 06:02

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.23343v1</id>\n    <title>Numerical Kernels on a Spatial Accelerator: A Study of Tenstorrent Wormhole</title>\n    <updated>2026-03-24T15:42:28Z</updated>\n    <link href='https://arxiv.org/abs/2603.23343v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.23343v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>As AI accelerators gain prominence, their potential for traditional scientific computing workloads remains unclear. This paper explores Tenstorrent's Wormhole architecture, a spatial computing platform designed for neural network acceleration, by implementing three numerical kernels and composing them into a conjugate gradient solver. We present architecture-specific optimizations for sparse numerical algorithms, evaluate their performance against Nvidia GPUs, and expose both challenges and opportunities in porting numerical methods to spatial architectures. Our results demonstrate that AI accelerators merit consideration for workloads traditionally dominated by CPUs and GPUs, and more work should be invested in understanding the capabilities of these architectures and making them accessible to the scientific computing community.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.PF'/>\n    <published>2026-03-24T15:42:28Z</published>\n    <arxiv:comment>12 pages, 13 figures</arxiv:comment>\n    <arxiv:primary_category term='cs.PF'/>\n    <author>\n      <name>Maya Taylor</name>\n    </author>\n    <author>\n      <name>Carl Pearson</name>\n    </author>\n    <author>\n      <name>Luc Berger-Vergiat</name>\n    </author>\n    <author>\n      <name>Giovanni Long</name>\n    </author>\n    <author>\n      <name>Jan Ciesko</name>\n    </author>\n  </entry>"
}