AMD Unveils ROCm 6.1 Software for Radeon GPUs, Enhancing AI Development

AMD has announced the release of its latest open software, AMD ROCm™ 6.1.3, aimed at extending its AI capabilities across its GPU portfolio, including AMD Radeon™ desktop GPUs. This move marks a significant leap in AMD's strategy to make ROCm software more accessible and versatile for developers working on AI workloads, according to AMD.com.

Expanded Support and New Features

The new AMD ROCm release broadens support for Radeon GPUs, enabling AI research and development on widely available desktop platforms. As highlighted by Andrej Zdravkovic, Senior Vice President at AMD, the latest release bridges the gap between data center and desktop environments, providing functional parity that facilitates scalable AI solutions.

Key enhancements in ROCm 6.1.3 include:

• Multi-GPU support for building scalable AI desktops capable of multi-serving and multi-user solutions.
• Beta-level support for Windows® Subsystem for Linux® (WSL), allowing ROCm to operate on Windows OS-based systems.
• Support for the TensorFlow framework, expanding the choices available for AI development.

Enhanced Multi-GPU Capabilities

ROCm 6.1.3 now supports configurations with up to four qualified Radeon™ RX Series or Radeon™ PRO GPUs. This capability leverages data parallelism, where each GPU computes inference independently, thus facilitating client-based multi-user configurations powered by AMD ROCm software.

Additionally, the new release includes Beta-level support for WSL 2, enabling the use of Linux-based AI tools on Windows systems without the need for a dedicated Linux setup. This feature simplifies the development process by eliminating dual-boot requirements.

Broader AI Framework Support

AMD has also introduced qualification for the TensorFlow framework, adding to the already supported PyTorch and ONNX frameworks. This expansion enhances the flexibility and robustness of AMD's open ecosystem, providing developers with more tools to innovate in AI.

New AI Hardware Solutions

The introduction of the AMD Radeon™ PRO W7900 Dual Slot card is another notable advancement. This card features 192 AI accelerators and 48GB of fast GDDR6 memory, optimized for high-density system configurations. It is particularly suited for fine-tuning and running inference on large language models (LLMs) like LLaMA-2 or LLaMA-3 with 70 billion parameters, which require substantial local GPU memory.

Such generative AI models are crucial for natural language processing (NLP) tasks, aiding enterprises in customer interaction, DevOps, and data management.

Commitment to an Open Ecosystem

AMD remains dedicated to building a scalable and open AI ecosystem with ROCm software. The latest ROCm 6.1 release for Radeon desktop GPUs empowers system builders to create on-premises systems that deliver robust AI performance, ideal for mission-critical and low-latency projects. This approach allows organizations to maintain control over their sensitive data.

With AMD ROCm 6.1 and the new Radeon PRO W7900 Dual Slot card, AI on desktops is now more accessible, opening new opportunities for developers and enterprises to enhance productivity and drive innovation.