---
format: "aidr-story-markdown/v1"
id: "be58f6497da33a853d871f7a01a7a18a1f041641d5b45ee5533d45b16cc3a721"
canonical_url: "https://aidr.today/be58f649?lang=en"
title: "Accelerating Mamba-2 with High-Performance Forward Pass Kernels in NVIDIA CUDA Tile"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-19T17:00:00.000Z"
category: "Frameworks"
topics: ["nvidia","inference","framework"]
source_urls: ["https://developer.nvidia.com/blog/accelerating-mamba-2-with-high-performance-forward-pass-kernels-in-nvidia-cuda-tile/"]
summary: "NVIDIA CUDA Tile (cuTile) is a Python-based DSL and parallel programming model for NVIDIA GPUs that automatically leverages advanced hardware capabilities to deliver peak performance for tile-level GPU programming. Machine learning engineers at Jane Street implemented Mamba-2, a state-of-the-art sequence model, in cuTile and partnered with NVIDIA to optimize the kernels. The post describes three custom kernels powering the Mamba-2 forward pass and the cuTile optimizations behind them."
---

# Accelerating Mamba\-2 with High\-Performance Forward Pass Kernels in NVIDIA CUDA Tile

> [Open the canonical story](<https://aidr.today/be58f649?lang=en>)

**Published:** 2026-10-19T17:00:00.000Z
**Category:** Frameworks
**Topics:** nvidia, inference, framework

## Summary

NVIDIA CUDA Tile \(cuTile\) is a Python\-based DSL and parallel programming model for NVIDIA GPUs that automatically leverages advanced hardware capabilities to deliver peak performance for tile\-level GPU programming\. Machine learning engineers at Jane Street implemented Mamba\-2, a state\-of\-the\-art sequence model, in cuTile and partnered with NVIDIA to optimize the kernels\. The post describes three custom kernels powering the Mamba\-2 forward pass and the cuTile optimizations behind them\.

## Sources

- [Story source](<https://developer.nvidia.com/blog/accelerating-mamba-2-with-high-performance-forward-pass-kernels-in-nvidia-cuda-tile/>)