NVIDIA Platforms Expand Support for DeepSeek, Gemma, and Nemotron Models

NVIDIA offers an AI inference platform for exploring and deploying community-built AI models. [2] NVIDIA Nemotron provides high-efficiency, multimodal, open models designed for long-running AI agents. [1] Another featured option is Gemma, which is a family of lightweight, open models created by Google DeepMind. [2]

The DeepSeek model family uses an open-source mixture-of-experts (MoE) architecture to deliver advanced reasoning capabilities. [2] These models can be optimized for data center deployments utilizing TensorRT-LLM or customized through the NeMo framework. [2] Additionally, the NVIDIA DeepSeek R1 FP4 model is a quantized variant created with the TensorRT Model Optimizer. [2]

NVIDIA is actively expanding its ecosystem to support diverse model architectures, positioning its proprietary Nemotron agent models alongside prominent community models like Google DeepMind's Gemma and DeepSeek's MoE architectures. By providing targeted optimization tools such as TensorRT-LLM, the NeMo framework, and TensorRT Model Optimizer, the company ensures its infrastructure remains the premier deployment layer for both proprietary and open-source models in data centers. What the sources don't address: How the inference efficiency and performance of NVIDIA's Nemotron models compare directly to optimized DeepSeek deployments on identical enterprise hardware.