GLM 5.2 on AMD MI355X: 2,626 tok/s at Over 2x Lower Cost Than Blackwell

Wafer.ai has announced that the GLM 5.2 language model running on AMD MI355X processors achieves 2,626 tokens per second per node at over twice lower cost than on Nvidia's Blackwell GPUs. The results suggest AMD hardware could be a serious alternative to the currently dominant Nvidia chips for AI workloads. This is a significant signal for the AI infrastructure market, where compute costs are a critical factor.

More on this topic

💻 Technology AMD Helios outpaces Nvidia on memory per rack but trails on rack-level compute TechRadar · 1d ago 💻 Technology Nvidia plans $250bn push to bolster OpenAI's AI infrastructure Al Jazeera · 2d ago 💻 Technology Nvidia and OpenAI in Talks for $250 Billion Credit Backstop for Ohio Data Center CNBC · 2d ago

More on this topic

💻 Technology AMD Helios outpaces Nvidia on memory per rack but trails on rack-level compute TechRadar · 1d ago 💻 Technology Nvidia plans $250bn push to bolster OpenAI's AI infrastructure Al Jazeera · 2d ago 💻 Technology Nvidia and OpenAI in Talks for $250 Billion Credit Backstop for Ohio Data Center CNBC · 2d ago