“NVIDIA GTC: The Most Powerful Chips Explode the AI World – A Comprehensive Guide to the Top Global AI Event”

By admin Mar 19, 2024

The global top artificial intelligence (AI) event – NVIDIA (NVDA-US) GTC Developer Conference took place on Monday (18th). As an annual technology release event, the GTC conference is regarded as a beacon for the AI industry, and founder Jensen Huang delivered a two-hour keynote speech at the event.

In the past year, AI has seen significant growth, with NVIDIA being one of the first companies to benefit from the AI boom, with its stock price soaring by 260% in a year. The current market value of NVIDIA has surpassed 2.2 trillion US dollars. NVIDIA employees becoming wealthy through stocks have been a hot topic in the industry.

Here are the key highlights from the GTC conference:

NVIDIA unveils the latest killer AI chip Blackwell GPU NVIDIA officially introduced the new generation AI platform, the Blackwell GPU, which is hailed as the most successful product in NVIDIA’s history, according to Jensen Huang.

Named after mathematician David Harold Blackwell, the Blackwell GPU continues the Hopper architecture previously introduced by NVIDIA. The Blackwell GPU contains 20.8 billion transistors and can support AI models with up to 10 trillion parameters.

In addition to the chip itself, this architecture incorporates the fifth-generation NVLink high-speed interconnect, the second-generation Transformer engine, and a comprehensive upgrade in multiple aspects. Huang mentioned that this new chip will be launched later this year.

On the commercial front, NVIDIA already has initial plans. Huang mentioned that they plan to introduce Blackwell to AI companies worldwide. Currently, companies like Amazon (AMZN-US), Dell, Google, Meta (META-US), Microsoft (MSFT-US), OpenAI, and Tesla (TSLA-US) have plans to use the Blackwell GPU.

Notably, the rumored B100 did not appear, and NVIDIA instead released the superchip GB200, comprised of one Grace CPU and two sets of Blackwell GPUs.

Huang stated that compared to the H100, the computational power of the GB200 is six times that of the former, and its performance for processing multi-modal specific domain tasks can be up to 30 times faster.

NVIDIA showcases robot product At the end of the speech, Jensen Huang unveiled a humanoid robot product named “GROOT” and had two small robots interact on stage, demonstrating the results of robots trained using NVIDIA’s simulation tools.

The “GROOT” project is a general foundational model for humanoid robot learning, equipped with the latest Blackwell platform for use by humanoid robot manufacturers. It assists robots in understanding natural language and mimicking human movements through observation.

NVIDIA mentioned that they are currently establishing an AI platform for humanoid robot companies such as 1X Technologies, Agility Robotics, Apptronik, Boston Dynamics, Figure AI, Fourier Intelligence, Sanctuary AI, Unitree Robotics, and XPENG Robotics.

NVIDIA updates 25 healthcare microservices At this event, NVIDIA also introduced the new AI inference server NIM (NVIDIA INFERENCE MICROSERVICE), allowing everyone to customize AI models and applications through this form.

In the healthcare field, NVIDIA introduced 25 new microservices based on healthcare scenarios (NIM), helping global healthcare companies improve efficiency using generative AI.

For healthcare industry companies, they can use these microservices to filter through trillions of drug compound options, collect more patient data, and conduct early disease detection.

Specifically, these microservices include a range of drug discovery models, including “MolMIM” for chemical generation, “ESMFold” for protein structure prediction, and “DiffDock” to help researchers understand how drug molecules interact with targets.

NVIDIA releases new “DRIVE Thor” automotive computing platform NVIDIA also unveiled the new “DRIVE Thor” automotive computing platform.

As the successor to the previous generation “DRIVE Orin,” this platform provides enhanced autonomous driving cabin performance – integrating the new NVIDIA Blackwell architecture designed specifically for Transformer, LLM, and generative AI workloads.

Several automakers have already announced collaboration plans. The world’s largest electric vehicle company BYD (002594-CN) will use “DRIVE Thor” to develop next-generation electric vehicles, and GAC and XPeng Motors have also announced plans to use “DRIVE Thor.”

Telecom & Climate Governance & cuLitho w At this event, NVIDIA also announced the launch of the “6G Research Cloud Platform.” Based on this platform, telecom companies can simulate physical terrain and artificial structures to significantly enhance wireless transmission reliability. Nokia and Samsung are partners in this field.

NVIDIA also introduced the “Earth-2 Climate Digital Twin Cloud Platform” for weather simulation and visualization. Earth-2 can provide warnings and updated forecasts within seconds, whereas traditional CPU-driven modeling requires minutes or hours.

The “Earth-2 Climate Digital Twin Cloud Platform” not only excels in speed but also offers climate images with a 12.5 times higher resolution compared to current numerical models, 1,000 times faster, and 3,000 times more energy-efficient.

Jensen Huang also presented practical scenarios for this platform. For instance, the Taiwan Central Weather Bureau plans to use this model to predict the landing position of typhoons for personnel evacuation, reducing casualties.

NVIDIA also announced a collaboration with TSMC and Synopsys to integrate cuLitho computational lithography technology into the chip manufacturing process to accelerate the production of next-generation chips.

Huang Renxun stated: “Computational lithography is the cornerstone of chip manufacturing. We are working with TSMC (2330-TW) and Synopsys to advance cuLitho, using accelerated computing and generative AI to pioneer new dimensions in semiconductor process scaling.”

By admin

Leave a Reply

Your email address will not be published. Required fields are marked *