Tenstorrent and Polymage Labs Announce Strategic Collaboration to Deliver State-of-the-Art AI Compiler for Tenstorrent Hardware#
New solution provides seamless compilation from AI models in PyTorch and JAX to Tenstorrent accelerators, achieving promising usability and performance in preliminary results
Jan 5, 2026 – Polymage Labs and Tenstorrent announced their strategic collaboration on a new, state-of-the-art AI compiler for Tenstorrent’s powerful AI accelerators. Compiler software enables the easy programming of chips by translating high-level programmer directives into optimized code that the hardware can execute. In just a few months, the joint effort has successfully enabled Polymage’s PolyBlocks compiler framework to seamlessly integrate with Tenstorrent hardware, delivering impressive performance from unmodified, high-level PyTorch and JAX code.
This collaboration addresses one of the most significant challenges in the AI hardware industry: the software ecosystem. Many AI accelerator vendors have struggled to provide a seamless, high-performance programming experience. This collaboration demonstrates an innovative solution, leveraging the MLIR-based PolyBlocks and Tenstorrent’s MLIR-based infrastructure to create an effective, reusable, and rapidly developed compiler stack.
The compiler solution features advanced transformation capabilities, including cross-operator fusion to reduce memory bandwidth usage, multi-level tiling to efficiently exploit parallelism, and mapping to Tenstorrent’s powerful and specialized matrix and vector units for accelerated compute.
“The synergy between the Polymage and Tenstorrent teams has been incredible,” said Uday Bondhugula, CTO at Polymage Labs and Professor of Computer Science at the Indian Institute of Science. “Using the modern, MLIR-based compiler infrastructures of both PolyBlocks and Tenstorrent (tt-mlir) allowed us to bring up a complete, end-to-end compiler for both PyTorch and JAX in just a few months, an engineering effort that otherwise would normally take several years when starting from scratch and hard to get right even at that. This collaboration validates our approach and showcases how an effective and reusable compiler stack can be quickly built with the right technologies, layers, concepts, building blocks, and collaboration”.
The preliminary results are highly promising – the compiler effortlessly handles complex operator compositions, including attention layers, compiling them from high-level PyTorch code with no user intervention. Expanding beyond traditional CPUs and GPUs, PolyBlocks from Polymage Labs thus now supports Tenstorrent hardware as a hardware target.
“Tenstorrent is committed to providing an easy-to-use software ecosystem, and this collaboration with Polymage Labs is a giant leap forward,” said Stan Sokorac, Senior Fellow, AI Hardware & Software at Tenstorrent. “Seeing unmodified PyTorch code fully compiled for Tenstorrent devices and run effortlessly with near-peak performance on our hardware is a big win for developers. This work directly addresses one of the hardest problems in AI hardware and makes our platform significantly more accessible.”
Two of the biggest strengths of a compiled approach to executing AI models are: (1) it enables cross-operator optimization and fusion, which is key to effectively using available memory bandwidth, and (2) it alleviates the burden of manually writing optimized kernels each time compute patterns or the hardware change. “The Polymage and Tenstorrent teams built a state-of-the-art AI compiler for Tenstorrent hardware with advanced code-generation and optimization capabilities”, said Nick Smith, Senior Principal Engineer at Tenstorrent.
Both teams are continuing to build on these early successes and will share more results as the collaboration progresses.
About Polymage Labs
Polymage Labs is a deep-technology startup incubated at the Indian Institute of Science, Bangalore, India, that builds cutting-edge compiler systems to enable the effective programming of AI chips for high performance.
info@polymagelabs.com
https://www.polymagelabs.com
About Tenstorrent
At Tenstorrent, we build computers for AI, and the developers that are shaping its future.
Our high-performance RISC-V-based CPUs and AI, modular chiplets, and scalable compute systems give developers full control, at any scale from a single-node experimentation to data center-scale deployment.
We believe in an open future. Our architecture and software are designed to be edited, forked, and owned. Our team of engineers, dreamers, and first-principle thinkers is redefining how hardware and software converge to accelerate innovation.
