Posts by Balazs Toth

GEAK V3: Agent-Driven, Repository-Level GPU Kernel Optimization across HIP, Triton, and FlyDSL on AMD GPUs

In the ever-evolving world of GPU computing, optimizing kernels for performance and efficiency is a critical challenge. Hand-tuning kernels demands deep technical expertise and manual iteration. In this blog, you will read about how GEAK v3, the latest iteration of the agent-driven framework, tackles this problem using enhanced features such as task planning, test-harness discovery, patch-based handling of multi-file kernels, dynamic memory system and expert knowledge database. Our results show improvements across three kernel languages (HIP, Triton, and FlyDSL) and both CDNA and RDNA GPUs.

Read more ...


Wan2.2 Fine-Tuning: Tailoring an Advanced Video Generation Model on a Single GPU

This blog post will guide you through fine-tuning Wan2.2 - a state-of-the-art video generation model - on a single AMD Instinct MI300X GPU. By following this guide, you’ll unlock Wan2.2’s advanced video generation capabilities and customize the output — whether in a unique artistic style or a specialized domain — all while running memory efficiently even on a single GPU. Here are some examples of how you can put this guide into practice:

Read more ...