Lamini ditches Nvidia in favour of AMD

Lamini, an AI startup, is using AMD GPUs instead of the more popular Nvidia GPUs to run large language models (LLMs) like Llama-2 for customers. Through their new LLM Superstation product, Lamini also makes it easy for any company to build proprietary LLMs with AMD Instinct GPUs.

What's going on here?

Lamini has been secretly running over 100 AMD Instinct MI200 GPUs and says AMD's ROCm software is on par with Nvidia's CUDA for large language models.

What does this mean?

Lamini optimizes running and finetuning large LLMs for enterprise use cases. Their software innovations like fast model switching speed up development. The LLM Superstation bundles Lamini's software with AMD GPUs optimized for LLMs.

This signals major competition for Nvidia, as AMD is taking aim at their dominance in the accelerators that power large AI models. Lamini embracing AMD GPUs proves they can run complex models like Nvidia GPUs.

Why should I care?

Competition between AMD and Nvidia in this space could drive down costs and increase access to the infrastructure needed to run advanced AI. For enterprises looking to deploy private customized AI models, having GPU options beyond just Nvidia is appealing. AMD's ROCm reaching parity with CUDA also opens up the software ecosystem.

Lamini highlights the cost and availability benefits of AMD Instinct GPUs over scarce Nvidia offerings. Tests showed AMD GPUs efficiently utilize throughput for key operations needed to train models. AMD is using Lamini's Superstation platform internally to make their ROCm more attractive to developers. Overall, Lamini's move to AMD for its cloud and on-prem AI platform signals AMD's emergence as an alternative to Nvidia for generative AI, presenting enterprises with more options.