Kimi-K3 Breaks Records on AMD Instinct MI355X GPUs with 2.8T Parameters

Moonshot AI's Kimi-K3, a massive 2.8-trillion-parameter large language model, has achieved successful deployment on AMD's high-performance Instinct MI355X GPUs.

The model, which became available for API access on July 16, leverages AMD's tensor parallel (TP8) architecture to enable efficient single-instance operation on an eight-GPU configuration.

Kimi-K3's ability to process up to 1 million tokens positions it as a leader in long-form reasoning and document analysis tasks.

The model introduces several architectural advancements, including Kimi Delta Attention (KDA), Gated Multi-head Latent Attention (MLA), and Stable Latent MoE, which reduce memory overhead and improve efficiency for long-context inference.