AMD's ROCm 7 Fires Up AI With 3.5x Speed

AMD launched ROCm 7 as their newest open software stack technology. The company built this platform to boost AI performance and help developers work faster. ROCm 7 replaces the older ROCm 6 system that AMD used for several years. The new software focuses heavily on AI inference tasks. Developers can expect major speed improvements across different AI models.

The software supports the latest MI350 series graphics cards from AMD. ROCm 7 works with advanced number formats like FP8, FP6, and FP4 for better precision. The platform handles mixed precision calculations seamlessly. These features help AI models run more efficiently on AMD hardware. Performance gains reach up to 3.5 times faster speeds compared to the previous version.

Testing shows impressive results across popular AI models. Llama 3.1 70B runs 3.2 times faster with the new software. Qwen2-72B sees a 3.4 times speed boost. DeepSeek R1 performs 3.8 times better than before. AMD claims their MI355X graphics card beats NVIDIA Blackwell B200 chips when running DeepSeek R1 tests.

AMD plans to bring ROCm 7 to consumer laptops and workstations later this year. The software will support both Linux and Windows operating systems. Enterprise customers will receive complete end-to-end AI solutions. The platform connects GPUs, CPUs, and data processing units together for maximum performance.
 

Attachments

  • AMD's ROCm 7 Fires Up AI With 3.5x Speed.webp
    AMD's ROCm 7 Fires Up AI With 3.5x Speed.webp
    39.7 KB · Views: 104

Trending content

Sponsored

Top