Nvidia made significant waves at this year’s GPU Technology Conference (GTC), showcasing their next-gen Blackwell AI platform among other groundbreaking innovations. Let’s delve into the key highlights.
Blackwell: The star of the show was Nvidia’s groundbreaking Blackwell GPU, a colossal AI accelerator comprising two GPUs working synergistically within a single package. It’s akin to a double-chip powerhouse capable of effortlessly handling the most demanding generative AI tasks.
But it’s not merely about raw power. Nvidia claims that Blackwell can achieve up to 30 times more inference throughput and four times better training performance than the previous H100 GPU, all while consuming less energy. This is attributed to cutting-edge technologies such as a second-gen Transformer Engine with FP4 precision support and a lightning-fast decompression engine.
But Nvidia didn’t stop there. They integrated the Blackwell GPU into a three-die “Superchip” called the GB200, combining two Blackwell GPUs with a Grace Arm CPU, resulting in a remarkably potent AI processor.
These GB200 Superchips are housed within the NVL72 rack, an AI juggernaut housing 72 Blackwell GPUs and 36 Grace CPUs. Nvidia claims that this single rack can train models with up to 27 trillion parameters, enabling companies to tackle the most formidable generative AI challenges.
The potential impact of Blackwell’s performance is monumental. Nvidia envisions breakthroughs in diverse fields ranging from data processing and engineering simulations to electronic design automation, computer-aided drug design, and quantum computing.
Also Read: South Korea’s President Yoon Warns of Tech Threats to Democracy at Global Summit
Furthermore, it’s not just Nvidia that’s enthusiastic about Blackwell. Major tech players like Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and Tesla are already throwing their weight behind it.
Nvidia has also unveiled new iterations of its DGX AI systems and SuperPOD rack-scale architecture to leverage the power of Blackwell fully. The DGX GB200 systems, utilising GB200 Superchips, promise a 15-fold increase in inference performance and triple the training speed compared to previous DGX H100 systems. Inference involves running live data through a machine-learning algorithm to obtain an output.
The new liquid-cooled SuperPOD architecture, built with DGX GB200 systems, boasts a staggering 11.5 exaflops of AI computing power at FP4 precision, along with a massive 240 terabytes of fast memory per rack. This level of performance is tailored to meet the demands of the most complex AI models and workloads.
Also Read: Top 10 Future Tech: Brain-Computer Interfaces to Eco-Friendly AI
Deepened Cloud AI Collaboration with Google and AWS: At GTC, Nvidia also emphasised its deepening partnerships with cloud giants Google and Amazon Web Services (AWS). Google announced its adoption of the Grace Blackwell AI computing platform and the NVIDIA DGX Cloud service, enhancing its generative AI capabilities.
The two companies will collaborate on optimising open models like Gemma, supporting JAX on Nvidia GPUs, and leveraging Nvidia’s NIM inference microservices to provide developers with a flexible, open platform for training and deploying AI models.
NIM Concept Announced: One of the most intriguing announcements at GTC was Nvidia’s Inference Microservices (NIM) concept. NIM aims to facilitate customers in deploying their generative AI applications securely, stably, and scalably. Part of Nvidia AI Enterprise, NIM includes pre-packaged AI models, integration code, and a pre-configured Kubernetes Helm chart for deployment.
NIM could accelerate the time-to-market for AI deployments, enabling developers to get models up and running in as little as 10 minutes.
Also Read: Samsung Launches Galaxy Ring: Your Health and Your Style in 2024
Project GR00T: Nvidia also unveiled Project GR00T (Generalist Robot 00 Technology), an ambitious initiative aimed at developing general-purpose foundation models for humanoid robots. Project GR00T aims to equip robots with the ability to comprehend natural language, mimic human movements through observation, and rapidly acquire skills like coordination, dexterity, and real-world navigation. Nvidia CEO Jensen Huang showcased several GR00T-powered robots performing various tasks, demonstrating the project’s potential.
Accompanying Project GR00T is Jetson Thor, a new computer specifically designed for humanoid robots, built on Nvidia’s Thor system-on-chip (SoC).
6G Research Platform: Nvidia also underscored its commitment to advancing cutting-edge technologies like 6G wireless and industrial digital twins. The company unveiled a 6G Research Cloud platform, intended to aid researchers in developing the next phase of wireless technology and laying the groundwork for a highly intelligent world.
Omniverse Gets New APIs: Additionally, Nvidia announced the availability of Omniverse Cloud APIs, expanding the reach of its Omniverse platform for creating industrial digital twin applications and workflows. This will enable software developers to seamlessly integrate Omniverse technologies into their existing design and automation software, accelerating the adoption of digital twins across industries.
Thor SoC Expands Footprint: Lastly, Nvidia’s GTC event showcased the company’s burgeoning presence in the automotive industry. Several automakers, including BYD, Hyper (under GAC Aion), and Xpeng, announced plans to utilise Nvidia’s Drive Thor system-on-chip (SoC) for their future electric vehicle fleets.
Drive Thor promises to deliver an unprecedented 1,000 teraflops of performance while reducing overall system costs, allowing automakers to integrate autonomous driving, in-cabin AI, and infotainment systems on a single platform.
