#cuTile
Optimize GPU kernels with CUDA Tile: a Python-based DSL with tile-centric primitives targeting Tensor Cores.

Fuse PyTorch ops into single tiled kernels kept in registers for peak performance. See examples, docs, and code:

https://bit.ly/3VigiKf
Accelerating Mamba-2 with High-Performance Forward Pass Kernels in NVIDIA CUDA Tile | NVIDIA Technical Blog
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…
developer.nvidia.com
November 10, 2026 at 3:00 PM
Rust + CUDA are coming to @hf.co kernel-builder. We just landed support for cuda-oxide and cutile-rs is up next!

Some example Rust kernels, including a flash-attn2 port to Rust and cuda-oxide:

github.com/huggingface/...

#rust #cuda #gpu #kernels
GitHub - huggingface/kernels-community-rs
Contribute to huggingface/kernels-community-rs development by creating an account on GitHub.
github.com
November 4, 2026 at 4:00 PM
📦 NVIDIA / cuda-rust
⭐ 3,867 (+32)
🗒 Rust

NVIDIA's CUDA platform for Rust. Host runtime crates plus Tile (cutile-rs) and SIMT (cuda-oxide) kernel programming models in idiomatic Rust.
GitHub - NVIDIA/cuda-rust: NVIDIA's CUDA platform for Rust. Host runtime crates plus Tile (cutile-rs) and SIMT (cuda-oxide) kernel programming models in idiomatic Rust.
NVIDIA's CUDA platform for Rust. Host runtime crates plus Tile (cutile-rs) and SIMT (cuda-oxide) kernel programming models in idiomatic Rust. - NVIDIA/cuda-rust
github.com
October 29, 2026 at 3:17 PM
Jane Street 및 NVIDIA, Mamba-2 순전파를 Triton보다 최대 1.82배 빠르게 구현한 cuTile 커널 공개

핵심 요약

• Jane Street와 NVIDIA 엔지니어들이 NVIDIA 기술 블로그 글에서 Mamba-2 순전파를 cuTile Python 커널 3개(chunk_cumsum, fwd_pass_state, compute_out)로 구현한 과정을 수식과 코드로 설명했습니다.
• NVIDIA B300 GPU 1장에서 mamba_ssm의 Triton 커널과 비교했을 때 1.14~1.82배 빨랐고, 상…
Jane Street 및 NVIDIA, Mamba-2 순전파를 Triton보다 최대 1.82배 빠르게 구현한 cuTile 커널 공개
핵심 요약 • Jane Street와 NVIDIA 엔지니어들이 NVIDIA 기술 블로그 글에서 Mamba-2 순전파를 cuTile Python 커널 3개(chunk_cumsum, fwd_pass_state, compute_out)로 구현한 과정을 수식과 코드로 설명했습니다. • NVIDIA B300 GPU 1장에서 mamba_ssm의 Triton 커널과 비교했을 때 1.14~1.82배 빨랐고, 상태 크기가 클수록 격차가 컸습니다. 측정 범위는 커널 단위 순전파뿐이며, 역전파와 학습 전체 속도는 측정하지 않았습니다. • 구현은 TileGym 저장소의 tilegym....
discuss.pytorch.kr
October 26, 2026 at 11:05 AM
NVIDIA and Jane Street built Mamba-2 kernels in cuTile, so developers can write GPU kernels in Python while the compiler handles Tensor Cores. Run your state-space models at https://massedcompute.com
October 24, 2026 at 12:04 AM
Jane Street and NVIDIA optimize three Mamba-2 forward pass kernels in cuTile. The Python-based cuTile DSL targets Tensor Cores automatically, letting engineers skip thread-level scheduling for tiled GPU work.
Jane Street and NVIDIA optimize three Mamba-2 forward pass kernels in cuTile - GenAInews
The Python-based cuTile DSL targets Tensor Cores automatically, letting engineers skip thread-level scheduling for tiled GPU work.
genainews.tech
October 19, 2026 at 5:29 PM
🚀 New benchmarks show NVIDIA CUDA’s cuTile primitives slashing exp2 latency in Mamba‑2, boosting Tensor‑Core powered sequence modeling. Jane Street’s tricks make attention faster than ever. Dive in to see the numbers! #NVIDIACUDA #Mamba2 #cuTile

🔗 aidailypost.com/news/nvidia-...
October 20, 2026 at 2:09 AM
NVIDIA CUDA TileによるMamba-2フォワードパスカーネルの高速化と実装詳解

NVIDIAのCUDA Tile(cuTile)を活用してMamba-2のフォワードパスを高速化するカスタムカーネルの実装手法と最適化を解説。

#CUDA #Mamba-2 #GPU #DeepLearning #NVIDIA
NVIDIA CUDA TileによるMamba-2フォワードパスカーネルの高速化と実装詳解
NVIDIAのCUDA Tile(cuTile)を活用してMamba-2のフォワードパスを高速化するカスタムカーネルの実装手法と最適化を解説。
ai.warp-studio.com
October 19, 2026 at 8:55 PM
Welcome Melih, Senior Research Scientist at NVIDIA, as a #EuroRust26 speaker! 🦀 He’ll introduce cuTile Rust, bringing Rust’s ownership model to GPU programming. If it compiles, it has no data races. ✨

👉 eurorust.eu/talks/fearle...

#RustLang
Fearless Concurrency on the GPU | EuroRust 2026
You don't need to be a GPU expert to write high-performance GPU code in safe Rust. This talk introduces cuTile Rust, an open source system that brings Rust's ownership model to GPU programming. Its…
eurorust.eu
October 12, 2026 at 1:17 PM
Java vllm-like framework claims 90% of perfomance of llama.cpp on local inference on NVIDIA GPUs by compiling Java to CUDA and cuTile
Java vllm-like framework claims 90% of perfomance of llama.cpp on local inferenc
A Java‑based framework uses TornadoVM to compile Java code to CUDA and cuTile, achieving up to 90% of the performance of llama.cpp for local LLM inference on NVIDIA GPUs. The system includes the jitLL
techyon.pages.dev
October 8, 2026 at 12:57 PM
ξ
cutile package stack
for #CUDA
in virtual environment
J:\env4xi
with #PyTorch
required by #CUDA
and #cutile
October 8, 2026 at 8:09 AM
NVIDIA just changed the game: Native Rust on GPU kernels.

Compile Rust directly to PTX with cuda-oxide & cutile-rs. No more runtime data races! But heavy compilation demands true bare-metal power.

Read the deep dive on CUDA Rust & GPU servers:👇
www.migservers.com/blogs/nvidia...

#NVIDIA #CUDA #AI
How Bare-Metal GPU Servers Prepare You for CUDA Rust
Learn how CUDA Rust supercharges GPU server speed and memory safety, and why bare-metal dedicated hardware is mandatory for testing.
www.migservers.com
October 8, 2026 at 7:45 AM
NVIDIA now has two native tracks for writing GPU kernels in Rust, compiled straight to PTX. cutile-rs runs on stable and is already used in HuggingFace's Grout and mistral.rs.

We want the talk from whoever has run one against real work. CFP closes 31 October.

cfp.rustnationuk.com

#RustLang #Rust
October 8, 2026 at 9:12 AM
It's insane but now you can just buy access to a frontier model for ~48hrs and have it blast out an optimized impl of another model. Jax/CUDA/whatev. Don't like vLLM/llama.cpp? Claude printed me a Jev-compatible server for Clef in 100% cuTile-python last night. now I can just use it locally forever
October 2, 2026 at 5:15 PM
Resistance is cutile
BORBS!

Two Sanderlings and a Red Knot (I think) borbing on the beach for the #BirdOfTheDay
#borb #birds 🌿
September 26, 2026 at 2:34 AM
Rust sigue ganando: ahora le tocó a NVIDIA

NVIDIA presentó CUDA Rust, con dos caminos: cuda-oxide, que compila Rust directo a PTX y está en alfa temprana, y cutile-rs, que ya está publicado y lo usan el motor de Hugging Face y mistral.rs.

▶ https://youtu.be/lvo5QRdHI0o

#ramyxtec #notiweek #gpu
September 26, 2026 at 3:00 PM
NVIDIA's Vera Rubin NVL72 platform boosts AI deployment efficiency.

The Vera Rubin NVL72 boosts system performance, enhancing tokens-per-dollar and revenue potential. New Rust models like cuTile improve GPU kernel safety. Large-scale AI demands dynamic electricity…

Read more on Kimbodo:
How to Build Cost-Effective, High‑Throughput AI Infrastructure: GPUs, Cloud Services and Deployment Tooling
What Happened Hardware and software advances Recent advances reinforce three practical levers for AI deployment: raw accelerator performance, software that unlocks that performance, and energy-aware infrastructure coordination. NVIDIA continues to…
kimbodo.com
September 23, 2026 at 5:55 PM
TornadoVM 7.0.0 is out!

New TileContext API: Write #GPU kernels from #Java in tiles, not threads, via #NVIDIA #cuTile model.

Now, cuTile methods can be mixed with cuBLAS functions in one TaskGraph, optimizing data transfers.

👉 github.com/beehive-lab/TornadoVM/releases/tag/v7.0.0

#opensource #AI
September 22, 2026 at 2:29 PM
TileGymがcuTile PythonからRustへの変換ツールを公開
エンジニアへの影響:既存のCUDAカーネルをRustへ移行し性能比99.5%を実現可能に
https://developer.nvidia.com/blog/translating-cuda-tile-operations-from-python-to-rust-using-agentic-ai/
Translating CUDA Tile Operations from Python to Rust Using Agentic AI | NVIDIA Technical Blog
cuTile Rust () is a tile-based system for safe, idiomatic GPU kernel authoring in the Rust programming language. Extending the Rust ownership model to tile-based GPU kernels, it splits mutable outputs…
developer.nvidia.com
September 19, 2026 at 9:00 PM
NVIDIA is expanding CUDA with native Rust GPU kernel development through the new cuda-oxide and cutile-rs projects.
linuxiac.com/nvidia-intro...

#NVIDIA #CUDA #Linux
NVIDIA Introduces CUDA Rust for Writing GPU Kernels
NVIDIA is expanding CUDA with native Rust GPU kernel development through the new cuda-oxide and cutile-rs projects.
linuxiac.com
September 18, 2026 at 1:04 PM
NVIDIA is expanding CUDA with native Rust GPU kernel development through the new cuda-oxide and cutile-rs projects.
#Linux
NVIDIA Introduces CUDA Rust for Writing GPU Kernels
NVIDIA is expanding CUDA with native Rust GPU kernel development through the new cuda-oxide and cutile-rs projects.
linuxiac.com
September 18, 2026 at 2:46 PM
HuggingFace shipped Nvidia's GPU safety tool 3 months pre-launch. cutile-rs has adopters; cuda-oxide has shared-memory safety on a roadmap. Nvidia called it one announcement.
Nvidia Shipped Memory-Safe GPU Kernels. One Track Has Adopters. One Doesn’t.
Nvidia's GPU safety launch was actually <span class="highlight">two products</span> at very different stages of readiness.
temperaturezero.com
September 18, 2026 at 9:30 PM