DigitalOcean and AMD announced a significant performance milestone for Character.ai. The collaboration doubled Character.ai's inference throughput. It also halved the cost per token.
AMD's Instinct MI300X and MI325X GPUs power Character.ai's large language models. These models handle over one billion queries daily.
The performance gains resulted from deep co-optimization across the hardware and software stack. This optimization leveraged AMD's ROCm open-source platform and the vLLM inference library.
The collaboration highlights AMD's growing competitiveness in the AI accelerator market. AMD competes on hardware specifications, software integration, and platform optimization.