NVIDIA made big news at their GTC 2024 event. Their announcements caused major excitement in the tech world.
“Anything you can digitize so long as there’s structure we can probably learn some patterns from it and if we can learn the patterns from it we can understand its meaning, we might be able to generate it as well and so therefore the generative AI revolution is here.” NVIDIA’s founder Jensen Huang at GTC 2024
Get ready as we break down the 10 most important takeaways NVIDIA revealed. From powerful new AI to lightning-fast performance, NVIDIA unveiled game-changing innovations that disrupted the industry.
🚨Improve performance on your model with the best series on the NVIDIA TAO Toolkit: Part 1, 2 and 3.
1. NVIDIA’s Blackwell new GPU: The 1 Exaflop Monster Upending Computing’s Cosmic Order
NVIDIA unleashed Blackwell GPU architecture — a mind-blowing 1+ exaflop computing behemoth contained in a single rack.
This exponential performance leap, millions of times beyond NVIDIA’s first OpenAI system, obliterates previous boundaries. Hailing the birth of a new “Moore’s Law,” the GB200’s cosmic capabilities foreshadow AI and scientific frontiers once unimaginable, redefining the limits of the possible.
“Over the course of the last eight years, we’ve increased computation by 1,000 times.” NVIDIA’s founder Jensen Huang at GTC 2024
2. Meet GR00T: NVIDIA’s Bid to Grow Self-Aware Humanoid Robots
Figure 2. GR00T: NVIDIA’s robotics bet
NVIDIA unveiled Project GR00T, an ambitious moonshot to birth the foundation model for robot sentience.
By harnessing NVIDIA’s full AI stack — Isaac Lab, OSMO, and Jetson Thor — this initiative aims to crack the code on humanoid robots that comprehend multimodal instructions and perform intricate tasks autonomously. If successful, GR00T could upend robotic capabilities, ushering in a new era of intelligent, versatile android assistants.
“Building foundation models for general humanoid robots is one of the most exciting problems to solve in AI today” NVIDIA’s founder Jensen Huang at GTC 2024
3. NVIDIA NIMs: Herding the Software ‘Bots’ That Will Resurrect Productivity
Productivity is about to soar into the stratosphere with NVIDIA’s game-changing Inference Microservices (NIMs).
These purposely-crafted CUDA-powered models will unite into intelligent agent swarms, harnessing inconceivable computing power. From coding to spacecraft design, from creative tasks to architectural marvels, NIMs’ collaborative neural dexterity foreshadows an unprecedented productivity renaissance that will redefine what’s possible across all sectors.
➡️ Read how LangChain Integrates NVIDIA NIM for GPU-optimized LLM Inference in RAG.
4. NVIDIA’s Jetson Thor Chip to Power Human-Like Robots
Get ready for big changes in robotics from NVIDIA. Their new Jetson Thor chip is made to power human-like robots. It upgrades the older Jetson Orin chip with much more computing power.
As we mentioned before, NVIDIA is also building special software called Project GR00T to help train and control these human-shaped robots. The goal is for robots to better work in our world designed for humans.
“In 2023 generative AI emerged and a new industry begins. Why a new industry? We are now producing software using computers to write software.” NVIDIA’s founder Jensen Huang at GTC 2024
5. NVIDIA NeMo Microservices: Build Your Own Custom AI Models With Ease
Creating custom AI models just got a whole lot simpler thanks to NVIDIA’s NeMo platform. New tools like Curator, Customizer, and Evaluator make every step easier — from organizing data to adjusting models exactly how you want to checking their performance. Whether you’re training an AI model from scratch or fine-tuning an existing one, these NeMo tools put the power of custom AI into anyone’s hands.
➡️ See Github repo on Nemo-Curator.
6. Omniverse APIs: NVIDIA’s Digital Twin Tech for Every Developer
Figure 6. Omniverse: digital twins at the center of NVIDIA’s future
With new APIs, NVIDIA is making its powerful Omniverse Cloud available everywhere. Developers can now build Omniverse’s digital twin and simulation capabilities right into their existing design and automation software.
This allows for easily creating virtual copies of real-world industrial operations and testing autonomous machines like robots or self-driving cars before making them. NVIDIA is opening up huge opportunities across industries.
7. Prepare for Extreme Weather with NVIDIA’s Earth-2 Climate Simulator
Figure 7. NVIDIA’s AI system to predict extreme weather
NVIDIA is tackling climate change with a new AI-driven platform called Earth-2. It creates virtual copies of real-world weather and environmental conditions. Using this “digital twin”, organizations can get highly detailed forecasts and warnings about upcoming storms, floods, and other severe events caused by global warming. With Earth-2’s realistic simulations, better preparations can be made before extreme weather strikes.
8. Real-Time Generative AI Healthcare Agents that are 90% Cheaper than Humans
Figure 8. NVIDIA’s AI Healthcare agents are poised to redefine the economics of healthcare
In a jaw-dropping demo, Nvidia unveiled “Rachel” — a stunningly lifelike AI nurse that interacts with incredible bedside manner. With nurses being America’s 6th largest profession, this AI could disrupt the entire healthcare field.
Rachel and her AI colleagues can provide top-notch medical guidance at a fraction of human costs. Brace for an economic earthquake as these ultra-affordable yet highly-capable AI nurses look to undercut human counterparts earning $90/hour.
9. NVIDIA Brings AI Muscle to Microsoft’s Azure Cloud
NVIDIA and Microsoft join forces to supercharge Azure’s AI offensive. From adopting NVIDIA’s beastly Grace chips to natively integrating NVIDIA’s DGX Cloud, the two tech titans are fusing their strengths.
NVIDIA’s Omniverse will power Azure’s industrial design tools, while its generative AI services will uplift healthcare and enterprise apps. As Microsoft’s Copilot synergizes with NVIDIA’s hardware, this union signals tectonic shifts primed to reshape Cloud AI dominance.
10. Architecture leak of GPT-4
The tech world held its breath as NVIDIA’s Jensen Huang appeared to lift the veil on GPT-4’s tightly-guarded architecture. 1.8 trillion parameters? A Mixture-of-Experts [1] layout?
If true, these specifics could unlock the next evolution of AI capabilities that OpenAI has kept under wraps. Huang’s tantalizing breadcrumbs suggest GPT-4 may unleash an unprecedented scale and novel techniques that could reshape the entire AI landscape as we know it.
🤖 Learn how Foundation Models are creating a new Computer Vision Pipeline v2.0.
References
[1] Towards Understanding Mixture of Experts in Deep Learning
Authors: Jose Gabriel Islas Montero, Dmitry Kazhdan.
If you would like to know more about Tenyks, sign up for a sandbox account.
