As AI infrastructure continues to evolve, companies are increasingly turning to industry benchmarks and real-world models to guide their GPU decisions. AMD’s Instinct GPUs have shown strong performance in key benchmarks and AI models, helping drive the adoption of AI technologies.
“AI adoption requires delivering both strong benchmark results and the ability to scale across various models used in production,” said Ronak Shah, product marketing manager of Global AI GPU at AMD. “Our strategy is to offer optimized performance for benchmarks like MLPerf while also supporting the latest generative models like Llama 3.1 405B and DeepSeek-R1.”
AMD recently made strides in the MLPerf Inference 5.0 round, submitting its first-ever results for the AMD Instinct MI325X, launched in October 2024. The company also achieved milestones with multi-node submissions in collaboration with partners.
Open-source model performance
“AMD’s focus on both benchmarks and open-source model performance has made it easier for our partners to leverage Instinct GPUs,” said Shah. “Supermicro, Asus, Gigabyte, and MangoBoost have all submitted MLPerf results using Instinct GPUs, demonstrating their growing adoption across the industry.”
The results from Llama 2 70B and Stable Diffusion XL (SDXL) with the MI325X GPUs highlight the competitive performance of AMD’s products, especially in generative AI workloads. Unique GPU partitioning techniques helped AMD achieve strong results, competing directly with NVIDIA’s H200 in the SDXL submission.
As AI continues to develop, AMD Instinct GPUs aim to meet the performance demands of a range of applications, from generative AI to large-scale language models.