Startup Majestic Labs Introduces Prometheus AI Server

Startup Majestic Labs has unveiled its Prometheus AI server, a new system designed to compete with Nvidia's offerings in the artificial intelligence hardware market.

The Prometheus server deviates from traditional AI hardware by not using GPUs. Instead, it employs Majestic's proprietary Ignite AI Processing Unit (AIU), which integrates Arm cores with RISC-V vector and tensor engines. Each Prometheus unit can be equipped with up to 12 AIU modules and a substantial memory capacity ranging from 8TB to 128TB of LPDDR6 RAM.

Designed for scalability, a standard 40U rack can house up to four Prometheus servers, with a combined power draw of 120kW. The system utilizes liquid cooling for thermal management, moving away from conventional air cooling methods. Majestic claims that a single Majestic rack offers the memory capacity equivalent to 25 Nvidia NVL72 Vera Rubin racks, while consuming less power.

The Prometheus server is slated for market release next year. It adheres to Open Compute Project (OCP) standards, ensuring compatibility with popular AI development frameworks such as PyTorch, vLLM, and OpenAI Triton.