RISC-V simulator enables detailed AI chip performance analysis
A RISC-V Simulator for the Tenstorrent Tensix Architecture
Distributed, Parallel, and Cluster ComputingHardware Architecture
Summary
Running code on some AI chips usually requires special hardware, which can be hard to get and limits understanding how the chip works inside. The authors created Wolfpine, a software tool that simulates a specific AI chip called Tensix, letting programmers run and study code without the physical device. Wolfpine is open, easy to modify, and helps reveal where the chip might slow down, which is usually hidden when using real hardware. This makes it easier to develop better software for AI and high-performance computing.
What this means in practice
- •For ai hardware developers: Use Wolfpine to test and optimize Tensix AI accelerator code without requiring physical hardware access.
- •For performance engineers: Diagnose and fix performance bottlenecks on Tensix-based RISC-V AI systems using detailed event tracing provided by Wolfpine.
Authors
Nick Brown, Daniyal Arshad
Abstract
RISC-V based AI accelerators have become popular and not only for AI and ML, but also as vehicles for exploring HPC workloads more generally. Tenstorrent are one example, and whilst they document their Tensix architecture in detail and develop their software stack openly, running code has historically required a physical card, and their recently released simulator targets hardware equivalence rather than introspection. In this paper we present Wolfpine, a community developed RISC-V simulator for the Tensix architecture, supporting both the Wormhole and Blackhole generations and integrating seamlessly with the TT-Metal software stack. Constructed entirely from publicly available documentation, open source software and measurements on physical hardware, Wolfpine is written in pure Python for readability and ease of modification. An opt-in, provenance-tracked cost model adds timing fidelity incrementally, and typed event streams, per-component tracing and timeline export enable programmers to identify and diagnose performance bottlenecks that the hardware itself hides.