Blog

In this blog post, we’ll continue our journey to build a state-of-the-art (SOTA) matmul kernel on NVIDIA Blackwell by exploring the cluster launch control (CLC) optimization. At the end of the post we’ll improve our performance by another 15% and achieve 1772 TFLOPs, exceeding that of the current SOTA.

The Modular community has been buzzing this month, from our Los Altos Meetup talks and fresh engineering docs to big wins with Inworld and Oracle. Catch the highlights, new tutorials, and open-source contributions in this edition of Modverse.

In this post, we continue on this journey and discuss how to leverage the 2SM technique along with pipelining to increase our performance about 5x and get within 85% of state-of-the-art (SOTA).

In this post we are going to continue our journey and improve our performance by more than 50x our initial kernel benchmark. Along the way we are going to explain more GPU programming concepts and leverage novel Blackwell features.

This series of blog posts will showcase how one can: 1. Write a high-performance GPU kernel on Blackwell that offers performance competitive to that of NVIDIA's cuBLAS implementation. 2. Shows how one can leverage Mojo's special features to make the kernel as simple as possible.

This past month brought a wave of community projects and milestones across the Modular ecosystem!Modular Platform 25.5 landed with Large Scale Batch Inference, leaner packages, and new integrations that make scaling AI easier than ever. It’s already powering production deployments like SF Compute’s Large Scale Inference Batch API, cutting costs by up to 80% while supporting more than 15 leading models.

Modular Platform 25.5 is here, and introduces Large Scale Batch Inference: a highly asynchronous, at-scale batch API built on open standards and powered by Mammoth. We're launching this new capability through our partner SF Compute, enabling high-volume AI performance with a fast, accurate, and efficient platform that seamlessly scales workloads across any hardware.

Modular has partnered with SF Compute to address a fundamental asymmetry in the AI ecosystem: while model capabilities advance exponentially, the economic structures governing compute costs remain anchored in legacy paradigms. 

Modular Inc. announces MAX High-Performance GenAI Serving and MAX Code Repo Agent now available in AWS Marketplace's new AI Agents and Tools category, delivering 10x performance improvements and streamlined AI deployment for enterprises.

Between a global hackathon, a major release, and standout community projects, last month was full of progress across the Modular ecosystem!Modular Platform 25.4 launched on June 18th, alongside the announcement of our official partnership with AMD, bringing full support for AMD Instinct™ MI300X and MI325X GPUs. You can now deploy the same container across both AMD and NVIDIA hardware with no code changes, no vendor lock-in, and no additional configuration!