anybody using lib.rs/crates/cubecl in prod?
it says "alpha" which in rust parlance means "ship it" right?
it says "alpha" which in rust parlance means "ship it" right?
cubecl
Multi-platform high-performance compute language extension for Rust
lib.rs
June 15, 2026 at 9:53 AM
anybody using lib.rs/crates/cubecl in prod?
it says "alpha" which in rust parlance means "ship it" right?
it says "alpha" which in rust parlance means "ship it" right?
☑️ Apple via Metal shader language
☑️ nVidia via CUDA
☑️ misc via Vulkan
☑️ land of slow via CPU
☑️ WASM via WebGPU
alrighty then
☑️ nVidia via CUDA
☑️ misc via Vulkan
☑️ land of slow via CPU
☑️ WASM via WebGPU
alrighty then
![CubeCL kernel running on WebGPU
initializing WebGPU…\nruntime: "wgpu<wgsl>"
input : [1.0, 2.0, 3.0, 4.0, 5.0]
square: [1.0, 4.0, 9.0, 16.0, 25.0]
(if you see [1, 4, 9, 16, 25], a CubeCL kernel just ran on WebGPU)](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:dqrcz4py2hrv4lzcs6ygoagj/bafkreianvsc4gigvhvf3wyhr5qlxkr52hcxnqlqur6srzbivheh4loust4)
June 15, 2026 at 10:17 AM
☑️ Apple via Metal shader language
☑️ nVidia via CUDA
☑️ misc via Vulkan
☑️ land of slow via CPU
☑️ WASM via WebGPU
alrighty then
☑️ nVidia via CUDA
☑️ misc via Vulkan
☑️ land of slow via CPU
☑️ WASM via WebGPU
alrighty then
I'll be speaking at #rustconf26I in three weeks!
If you want to know what it takes to build a ONNX runtime from scratch, how to get the GPU to perform math from Rust, and if you should use cubecl, rust-gpu, or wgpu, be sure to tune in!
Grab your seat TODAY: bit.ly/4wwduG6
#rustconf #rustlang
If you want to know what it takes to build a ONNX runtime from scratch, how to get the GPU to perform math from Rust, and if you should use cubecl, rust-gpu, or wgpu, be sure to tune in!
Grab your seat TODAY: bit.ly/4wwduG6
#rustconf #rustlang
Register - RustConf 2026
RustConf tickets (Employer-paid & individual) are available for our 10th annual event dedicated to the Rust programming language.
bit.ly
August 17, 2026 at 7:30 AM
I'll be speaking at #rustconf26I in three weeks!
If you want to know what it takes to build a ONNX runtime from scratch, how to get the GPU to perform math from Rust, and if you should use cubecl, rust-gpu, or wgpu, be sure to tune in!
Grab your seat TODAY: bit.ly/4wwduG6
#rustconf #rustlang
If you want to know what it takes to build a ONNX runtime from scratch, how to get the GPU to perform math from Rust, and if you should use cubecl, rust-gpu, or wgpu, be sure to tune in!
Grab your seat TODAY: bit.ly/4wwduG6
#rustconf #rustlang
Very interesting interview. Tracel uses Rust to build Burn, an inference and model training framework, and CubeCL, a compute language. Why Rust? Because its static guarantees are key for performance optimization, and its portability allows targeting any device and platform, large or small.
CUDA, CUDA, CUDA... What if there was a different, open way to write highly-efficient GPU kernels?
This is the big swing that Tracel AI is taking. And, guess what... Rust is key to their approach. Check out our podcast with Tracel CEO Nathaniel Simard (linked below):
This is the big swing that Tracel AI is taking. And, guess what... Rust is key to their approach. Check out our podcast with Tracel CEO Nathaniel Simard (linked below):
April 1, 2026 at 6:19 AM
Very interesting interview. Tracel uses Rust to build Burn, an inference and model training framework, and CubeCL, a compute language. Why Rust? Because its static guarantees are key for performance optimization, and its portability allows targeting any device and platform, large or small.
Big ideas, bold experiments, and lessons learned—#rustconf25 kicks off next week! 🦀
Featured talks: AI with Burn & CubeCL, #rustlang vs Carbon, cheaper Rust CI, and 10 years of Redox OS.
Register: ti.to/rustconf/202...
Details: rustconf.com
#rustconf #opensource
Featured talks: AI with Burn & CubeCL, #rustlang vs Carbon, cheaper Rust CI, and 10 years of Redox OS.
Register: ti.to/rustconf/202...
Details: rustconf.com
#rustconf #opensource
RustConf 2025
RustConf 2025 | Seattle, WA & Online, Sept 2-5, 2025
ti.to
August 28, 2025 at 4:45 PM
Big ideas, bold experiments, and lessons learned—#rustconf25 kicks off next week! 🦀
Featured talks: AI with Burn & CubeCL, #rustlang vs Carbon, cheaper Rust CI, and 10 years of Redox OS.
Register: ti.to/rustconf/202...
Details: rustconf.com
#rustconf #opensource
Featured talks: AI with Burn & CubeCL, #rustlang vs Carbon, cheaper Rust CI, and 10 years of Redox OS.
Register: ti.to/rustconf/202...
Details: rustconf.com
#rustconf #opensource
This is really cool. It's kind of like a cross-platform CUDA for Rust:
https://github.com/tracel-ai/cubecl
https://github.com/tracel-ai/cubecl
GitHub - tracel-ai/cubecl: Multi-platform high-performance compute language extension for Rust.
Multi-platform high-performance compute language extension for Rust. - tracel-ai/cubecl
github.com
December 23, 2024 at 9:28 PM
This is really cool. It's kind of like a cross-platform CUDA for Rust:
https://github.com/tracel-ai/cubecl
https://github.com/tracel-ai/cubecl
CubeCL: GPU Kernels in Rust for CUDA, ROCm, and WGPU Discussion
GitHub - tracel-ai/cubecl: Multi-platform high-performance compute language extension for Rust.
Multi-platform high-performance compute language extension for Rust. - tracel-ai/cubecl
github.com
April 23, 2025 at 11:40 PM
CubeCL: GPU Kernels in Rust for CUDA, ROCm, and WGPU Discussion
CubeCL looks like the most interesting implementation of Rust-on-GPU so far. Write normal-looking Rust functions and have it compiled to a variety of different backends.
GitHub - tracel-ai/cubecl: Multi-platform high-performance compute language extension for Rust.
Multi-platform high-performance compute language extension for Rust. - tracel-ai/cubecl
github.com
February 27, 2025 at 3:13 PM
CubeCL looks like the most interesting implementation of Rust-on-GPU so far. Write normal-looking Rust functions and have it compiled to a variety of different backends.
Cubecl is great! Depends on your use case but I used it to parallelize lie series computation on the GPU in pure Rust. This was last year so presumably it has fewer rough edges now.
June 15, 2026 at 2:19 PM
Cubecl is great! Depends on your use case but I used it to parallelize lie series computation on the GPU in pure Rust. This was last year so presumably it has fewer rough edges now.
If you ever wanted to write some high-performance multi-platform compute stuff in #rust but didn't have the language extension, fret no more
github.com/tracel-ai/cu...
github.com/tracel-ai/cu...
GitHub - tracel-ai/cubecl
Contribute to tracel-ai/cubecl development by creating an account on GitHub.
github.com
July 19, 2024 at 8:37 PM
If you ever wanted to write some high-performance multi-platform compute stuff in #rust but didn't have the language extension, fret no more
github.com/tracel-ai/cu...
github.com/tracel-ai/cu...
I've only played with CubeCL for an hour or so, but it seems to do what it says on the tin. Rust on the GPU has had a very scattered story so far, but CubeCL looks like they're making real progress towards something portable.
But glad to hear you're liking Mojo. Maybe I'll give it another look.
But glad to hear you're liking Mojo. Maybe I'll give it another look.
June 5, 2025 at 5:55 PM
I've only played with CubeCL for an hour or so, but it seems to do what it says on the tin. Rust on the GPU has had a very scattered story so far, but CubeCL looks like they're making real progress towards something portable.
But glad to hear you're liking Mojo. Maybe I'll give it another look.
But glad to hear you're liking Mojo. Maybe I'll give it another look.
🎙️Speaker Spotlight: Nathaniel Simard
Burn 0.18.0 brought big performance and reliability boosts. Creator Nathaniel Simard will be at #rustconf25 with a talk on “Rust for AI & Accelerated Computing,” featuring Burn and CubeCL.
Join us Sept 2–5
Register: ti.to/rustconf/202... #rustconf
Burn 0.18.0 brought big performance and reliability boosts. Creator Nathaniel Simard will be at #rustconf25 with a talk on “Rust for AI & Accelerated Computing,” featuring Burn and CubeCL.
Join us Sept 2–5
Register: ti.to/rustconf/202... #rustconf
RustConf 2025
RustConf 2025 | Seattle, WA & Online, Sept 2-5, 2025
ti.to
August 1, 2025 at 12:19 AM
🎙️Speaker Spotlight: Nathaniel Simard
Burn 0.18.0 brought big performance and reliability boosts. Creator Nathaniel Simard will be at #rustconf25 with a talk on “Rust for AI & Accelerated Computing,” featuring Burn and CubeCL.
Join us Sept 2–5
Register: ti.to/rustconf/202... #rustconf
Burn 0.18.0 brought big performance and reliability boosts. Creator Nathaniel Simard will be at #rustconf25 with a talk on “Rust for AI & Accelerated Computing,” featuring Burn and CubeCL.
Join us Sept 2–5
Register: ti.to/rustconf/202... #rustconf
CubeCL and by proxy Burn depend on `internment` 😬
I sure hope no one did cargo update or added burn as a new dep in those 90 minutes
I sure hope no one did cargo update or added burn as a new dep in those 90 minutes
August 20, 2026 at 12:17 PM
CubeCL and by proxy Burn depend on `internment` 😬
I sure hope no one did cargo update or added burn as a new dep in those 90 minutes
I sure hope no one did cargo update or added burn as a new dep in those 90 minutes
6. The duo of Louis Fortier-Dubois and Nathaniel Simard from Burn introducing CubeCL, a crate that churns out portable, high-performance kernels to keep data movement from stealing the show. Don’t skip this!
burn.dev/blog/sota-mu...
burn.dev/blog/sota-mu...
State-of-the-Art Multiplatform Matrix Multiplication Kernels
We implemented a sophisticated matrix multiplication engine in CubeCL that rivals the performance of cuBLAS and CUTLASS while supporting a wider range of GPUs. Leveraging double buffering, tensor…
burn.dev
July 28, 2025 at 6:53 AM
6. The duo of Louis Fortier-Dubois and Nathaniel Simard from Burn introducing CubeCL, a crate that churns out portable, high-performance kernels to keep data movement from stealing the show. Don’t skip this!
burn.dev/blog/sota-mu...
burn.dev/blog/sota-mu...
Honestly semver is decorative in ML tooling anyway — half the "1.0" libs break you harder than this thing's alpha does. At least CubeCL is upfront about it
June 15, 2026 at 1:20 PM
Honestly semver is decorative in ML tooling anyway — half the "1.0" libs break you harder than this thing's alpha does. At least CubeCL is upfront about it
Super impressive numbers down the thread! What did you start off with before moving to wgpu? cuda-oxide and CubeCL have been on my list to try for a while but just haven’t had the time.
July 13, 2026 at 1:35 AM
Super impressive numbers down the thread! What did you start off with before moving to wgpu? cuda-oxide and CubeCL have been on my list to try for a while but just haven’t had the time.
Multiplatform Matrix Multiplication Kernels | Discussion
State-of-the-Art Multiplatform Matrix Multiplication Kernels
We implemented a sophisticated matrix multiplication engine in CubeCL that rivals the performance of cuBLAS and CUTLASS while supporting a wider range of GPUs. Leveraging double buffering, tensor cores, and vectorization, it compiles seamlessly to CUDA, ROCm, WebGPU, Metal, and Vulkan backends without relying on proprietary or third-party binaries. Matrix multiplication is central to modern AI workloads, especially transformers, and optimizing it ourselves was essential to enable kernel fusion and achieve state-of-the-art performance across platforms in a deep learning framework.
burn.dev
July 18, 2025 at 9:40 PM
Multiplatform Matrix Multiplication Kernels | Discussion
@jeff.cuchazinteractive.org I wasn't aware there was a good GPU programming story in rust, CubeCL looks like it has similar goals but for the rust ecosystem :-)
I'm really excited about Mojo, as a Python person it's giving me low level control I've never had with minimal mental overhead/complexity
I'm really excited about Mojo, as a Python person it's giving me low level control I've never had with minimal mental overhead/complexity
June 5, 2025 at 4:54 PM
@jeff.cuchazinteractive.org I wasn't aware there was a good GPU programming story in rust, CubeCL looks like it has similar goals but for the rust ecosystem :-)
I'm really excited about Mojo, as a Python person it's giving me low level control I've never had with minimal mental overhead/complexity
I'm really excited about Mojo, as a Python person it's giving me low level control I've never had with minimal mental overhead/complexity
関数をステートレスの形に保ったまま環境を与える方式を編み出した。クロージャの実装みたいな形にするということ。
&selfをとろうとすると、CubeCLに生成されるコードがめちゃくちゃすぎて扱うことが事実上不可能だった。
&selfをとろうとすると、CubeCLに生成されるコードがめちゃくちゃすぎて扱うことが事実上不可能だった。
July 1, 2026 at 11:00 AM
関数をステートレスの形に保ったまま環境を与える方式を編み出した。クロージャの実装みたいな形にするということ。
&selfをとろうとすると、CubeCLに生成されるコードがめちゃくちゃすぎて扱うことが事実上不可能だった。
&selfをとろうとすると、CubeCLに生成されるコードがめちゃくちゃすぎて扱うことが事実上不可能だった。
Multiplatform Matrix Multiplication Kernels https://burn.dev/blog/sota-multiplatform-matmul/ (https://news.ycombinator.com/item?id=44609137)
State-of-the-Art Multiplatform Matrix Multiplication Kernels
We implemented a sophisticated matrix multiplication engine in CubeCL that rivals the performance of cuBLAS and CUTLASS while supporting a wider range of GPUs. Leveraging double buffering, tensor cores, and vectorization, it compiles seamlessly to CUDA, ROCm, WebGPU, Metal, and Vulkan backends without relying on proprietary or third-party binaries. Matrix multiplication is central to modern AI workloads, especially transformers, and optimizing it ourselves was essential to enable kernel fusion and achieve state-of-the-art performance across platforms in a deep learning framework.
burn.dev
July 18, 2025 at 9:49 PM
Multiplatform Matrix Multiplication Kernels https://burn.dev/blog/sota-multiplatform-matmul/ (https://news.ycombinator.com/item?id=44609137)