AMD officially releases ROCm 7.0, a competitor to CUDA technology

Teaser image. Source: AMD

AMD has officially released ROCm 7.0, a new version of the open platform for GPU computing. The technology competes with Nvidia's CUDA and allows accelerating computing in applications using a video card. The seventh version received mechanisms for computing artificial intelligence operations on the GPU. This is one of the biggest updates in recent years.

Key innovations:

• Full support for Instinct MI350 on CDNA 4 architecture
• Optimised training and inference in FP4, FP6, FP8
• Support for distributed inference on multi-GPU builds
• New HIP 7.0 for better code portability
• Tools for enterprises: AMD Resource Manager and AI Workbench
• Out-of-the-box Docker images with Llama 3.3 70B, Llama 3.1-405B, DeepSeek R1 models

AMD also introduced the Quark Toolkit, a set of tools for working with large quantified models to reduce costs and speed up deployment.

ROCm 7.0 is now available on the ROCm AI developer portal, with documentation, instructions, and quick-start containers.

Source: www.amd.com