Products You May Like
Microsoft Ignite is underway in Chicago, and the first big news to come from it involves, as you’d expect, AI. And where there’s AI, there’s often Nvidia, so it’s no surprise that Microsoft is releasing AI tools for developers aimed at what they’ve dubbed “GeForce RTX AI PCs.”
The new tools will help developers create “digital humans” to be the faces of digital agents, assistants, and avatars that perceive the world in a more human-like way. Nvidia believes this approach will give digital humans better context when answering questions from us old-school humans.
To that end, it announced the Nvidia Nemovision-4B-Instruct model. It uses Nvidia Vila and Nvidia NeMo to deliver accurate performance on RTX GPUs. The tool, Nvidia says, helps digital humans see images on the screen (and in the world) the same way we do.
“Multimodality serves at the foundation for agentic workflows and offers a sneak peek into a future where digital humans can reason and take action with minimal assistance from a user,” Nvidia said in a statement.
Mistral NeMo and TensorRT Updates
Nvidia also announced the Mistral NeMo Minitron 128k Instruct suite of small language models. The large-context model will be available in 2, 4, and 8 billion parameter versions. The GGUF format models are designed to accommodate a variety of PCs, including low-power devices, to create “digital human interactions.”
The TensorRT Model Optimizer received new capabilities. “With the latest updates, TensorRT ModelOpt enables models to be optimized into an ONNX checkpoint for deploying the model with ONNX runtime environments—using GPU execution providers such as CUDA, TensorRT, and DirectML,” Nvidia said in a statement.
Nvidia expects the updated TensorRT ModelOpt to reduce memory footprints during deployment by as much as 2.6x compared with FP 16 format. This would be on RTX GPUs.
As chipmakers have been finding their footing in the AI market, Nvidia has been stomping its way to success with excellent performance from its GPU lineup and AI accelerators. Last month, Intel acknowledged that it won’t try to compete with Nvidia on high-end performance but rather woo customers with cost and overall value with its Gaudi 3 accelerators. Conversely, AMD has high-performance chips in its 5th Gen Epyc CPUs and AI accelerators (with MI325X leading the way).
Microsoft Ignite started this morning (not counting the pre-game fun yesterday afternoon). It runs until Friday, so expect more AI-related announcements this week.