Blog

Democratizing AI Compute Series
Go behind the scenes of the AI industry with Chris Lattner

Modular at LLVM Developers' Meeting 2026
As GPUs get faster, the cost of CPU-to-accelerator communication grows until it takes up much of the inference time. Device graphs address that by submitting a whole sequence of work to the device at once. MAX uses them, and so does every widely used serving stack. The usual way to build one is record and replay: capture a stream of commands at runtime, then replay it. CUDA and HIP both offer this.

LLVM Dev Meeting 2026: Durable Infrastructure in the Agentic Era
Kristof’s exploration of the question also revealed a lot about the future of open source software projects. Kristof argued that durable open source projects depend on three things: communicating intent, maintaining cognition, and supporting contributors. If contributions are to maintain the architectural vision and standard of quality that the core maintainers expect, projects need to document and share that intent more deliberately.

Modverse #55: Mojo 1.0 Beta, Community Mojo Libraries, and Real-Time Patient Conversations Powered by MAX
This edition captures everything happening across the Modular ecosystem, from developers building with MAX and Mojo🔥 to the broader impact Modular is having across AI infrastructure. Here's a look at what's been happening lately.

Modverse #54: AMD AI DevDay, New Modular Offices, and a Community That Keeps Shipping
There was a lot to celebrate in April: the community shipped GPU renderers, FFmpeg bindings, raylib wrappers, BLAS routines, and a 2D graphics API, just to name a few. The team connected with tons of developers at AMD AI DevDay and our joint meetup with AMD, two new Modular offices opened on two different continents, and Gemma 4 launched with same-day support on NVIDIA and AMD. Here’s the April roundup.

Modverse #53: Community Builds, Research Milestones, and a Growing Ecosystem
This edition captures everything happening across the Modular ecosystem, from developers building with MAX and Mojo🔥 to the broader impact Modular is having across AI infrastructure. Here's a look at what's been happening lately.

How to Beat Unsloth's CUDA Kernel Using Mojo—With Zero GPU Experience
Traditional GPU programming has a steep learning curve. The performance gains are massive, but the path to get there (CUDA, PTX, memory hierarchies, occupancy tuning) stops most developers before they start. Mojo aims to flatten that curve: Python-like syntax, systems-level performance, no interop gymnastics, and the same performance gains.

🔥 Modular 2025 Year in Review
Our four-part series documenting the path to record-breaking matrix multiplication performance became essential reading for anyone serious about LLM optimization. The series walks through every optimization step—from baseline implementations to advanced techniques like warp specialization and async copies—showing you exactly how to extract maximum performance from cutting-edge hardware.

Democratizing Compute
Go behind the scenes of the AI industry in this blog series by Chris Lattner. Trace the evolution of AI compute, dissect its current challenges, and discover how Modular is raising the bar with the world’s most open inference stack.

Matrix Multiplication on Blackwell
Learn how to write a high-performance GPU kernel on Blackwell that offers performance competitive to that of NVIDIA's cuBLAS implementation while leveraging Mojo's special features to make the kernel as simple as possible.

Structured Mojo Kernels
Learn how Mojo simplifies GPU programming with modular kernel architecture, compile-time abstractions, and zero-cost performance across modern GPU hardware.

Software Pipelining for GPU Kernels
Explore software pipelining for GPU kernels from first principles. We formalize dependencies as a graph, solve for the optimal schedule with a constraint solver, and show how it all integrates into MAX via pure Mojo.

Why LLM Inference Needs a New Kind of Router
This series walks through why traditional HTTP routing breaks down under LLM workloads and how Modular Cloud solves it with a three-layer architecture built for cache-aware routing.

TileTensor
This series walks through how Modular built TileTensor, a Mojo tensor type that lets kernel authors express complex memory layouts precisely, safely, and efficiently.
No items found within this category
We couldn’t find anything. Try changing or resetting your filters.

Sign up today
Signup to our Cloud Platform today to get started easily.
Sign Up
Browse open models
Browse our model catalog, or deploy your own custom model
Browse models

