Nvidia 2023 Quantum, AI, Chip ...... The future is here!
The AI event GTC 2023 officially unfolded at 11pm on March 21, along with a speech by Nvidia CEO Jen-Hsun Huang.
"AI's 'iPhone moment' has arrived", at the Nvidia GTC 2023 conference, Nvidia CEO Jen-Hsun Huang repeated this idea 4 or 5 times.
Not only Nvidia, last night, Microsoft, Google, Adobe and other vendors launched their own AI services in the same time, you catch up with me as if to convey the same anxiety: "In this big AI era, if you do not want to be subverted, you must first subvert others." Overnight, we witnessed several big breakthroughs in AI from the service layer to the application layer. Our lives are being rewritten by AI on a daily basis.
Now in its 14th year, the 2023 GTC Developer Conference will feature 650 presentations from AI development leaders. NVIDIA's CEO Jen-Hsun Huang's keynote covers how NVIDIA and its partners offer everything from training to cutting-edge service deployments. After launching ChatGPT last year, the CEO said, "We are in the iPhone moment of AI." This year, Jen-Hsun Huang discussed how NVIDIA and its partners are seeking to leverage AI technology to overcome a wide range of challenges.
This time, the new system, built with Israeli company Quantum Machines, offers a revolutionary new architecture for researchers working on high-performance and low-latency quantum classical computing.
As the world's first GPU-accelerated quantum computing system, NVIDIA DGX Quantum combines the world's most powerful accelerated computing platform, enabled by the NVIDIA Grace Hopper superchip and the CUDA Quantum open source programming model, with the world's most advanced quantum control platform, OPX (powered by Quantum Machines) combined with the world's most advanced quantum control platform, OPX (provided by Quantum Machines).
This combination enables researchers to build exceptionally powerful applications that combine quantum computing with state-of-the-art classical computing, enabling calibration, control, quantum error correction and hybrid algorithms.
"Quantum-accelerated supercomputing has the potential to reshape science and industry with capabilities that can serve humanity in tremendous ways," said Tim Costa, NVIDIA's head of HPC and quantum, "NVIDIA DGX Quantum will enable researchers to push the boundaries of quantum-classical computing."
At the heart of DGX Quantum is an NVIDIA Grace Hopper system connected by PCIe to the Quantum Machine OPX+, enabling sub-microsecond latency between the GPU and the quantum processing unit (QPU).
DGX Quantum also equips developers with NVIDIA's hybrid GPU-Quantum programming model, CUDA Quantum, a powerful unified software stack that is now open source.CUDA Quantum is a hybrid quantum-classical computing platform that enables integrate and program QPUs, GPUs and CPUs in a single system.
The new platform is named after CUDA, the software most AI developers use to access NVIDIA's graphics processing units (GPUs), which gives NVIDIA chips a huge competitive advantage; it is also the first platform to build quantum algorithms using the popular classical computer coding languages C++ and Python. The program will help run algorithms on both quantum and classical computers, depending on which system is most effective at solving problems.
"CUDA Quantum will do the same for quantum computing, enabling domain scientists to seamlessly integrate quantum into their applications and gain access to new disruptive computing technologies," said Tim Costa, head of HPC and quantum at Nvidia, adding that while CUDA is proprietary, CUDA Quantum is open source and was developed with input from many quantum computing companies.
Finally, NVIDIA announced a new group of partners to integrate CUDA Quantum into its platform, including quantum hardware companies Anyon Systems, Atom Computing, IonQ, ORCA Computing, Oxford Quantum Circuits and QuEra; quantum software companies Agnostiq and QMware; and supercomputing centers National Institute of Advanced Industrial Science and Technology, Center for Information Technology Sciences (CSC) and National Center for Supercomputing Applications (NCSA).
At the end of the quantum computing panel, Tim Costa said, "All quantum today is research, not production, and that's not going to change in the short term," adding that with DGX Quantum, however, researchers will be able to develop hybrid applications and key methods for the future of quantum computing.
In 1999, NVIDIA defined the GPU, the advent of which is seen by the industry as the beginning of modern computer graphics technology. However, the GPU's microarchitecture is inherently suited for matrix-like parallel computing, and its capabilities are not limited to the realm of graphics cards, so from the early 21st century there were professional computing professionals who wanted to use the GPU to do some parallel computing related to the field of artificial intelligence.
Now, GPUs have become the main driver of AI general-purpose chips, artificial intelligence.
At the GTC conference yesterday, NVIDIA introduced new GPUs: four configurations (L4 Tensor Core GPU, L40 GPU, H100 NVL GPU, Grace Hopper superchip), one architecture, and one software stack for accelerating AI video, image generation, large language model deployment, and recommender systems, respectively.
1) L4: A general-purpose GPU designed for AI video, providing 120 times higher AI video performance than CPU with 99% higher energy efficiency; optimized for video decoding and transcoding, video content review, and video calls, such as background replacement, re-lighting, eye contact, transcription, and real-time translation.
2) L40: for image generation, optimized for graphics and AI-enabled 2D, video and 3D image generation, with 10x the inference performance of NVIDIA's most popular cloud inference GPU, the T4.
3) H100 NVL: This new H100 NVL with dual GPU NVLink is based on NVIDIA's Hopper architecture and uses the Transformer Engine design to process models, such as the GPT model that powers ChatGPT. In comparison, the new H100 with dual GPU NVLink is 10x faster than the previous HGX A100 in GPT-3 processing.
4) Grace Hopper superchip: The new superchip Grace Hopper connects the Grace CPU and Hopper GPU via a high-speed 900GB/sec coherent chip-to-chip interface. according to Jen-Hsun Huang, the new superchip is said to be ideal for processing large data sets, such as AI databases for recommender systems and large language models. Grace Hopper is ideal for customers looking to build AI databases several orders of magnitude larger.
In response to the above breakthrough, Jen-Hsun Huang said the NVIDIA DGX H100 is a blueprint for customers seeking to build AI infrastructure on a global scale and is now in full production.
In fact, the H100 AI supercomputer is already online, including the announced Oracle cloud infrastructure. Amazon Web Services has also announced P5 instances of its EC2 UltraClusters, which can scale to 20,000 interconnected H100 GPUs.
To accelerate DGX capabilities to startups and enterprises striving to bring new products and develop AI strategies, Jen-Hsun Huang announced NVIDIA DGX Cloud, a new cloud technology that will bring NVIDIA DGX AI supercomputers "from the browser to every company" through partnerships with Microsoft Azure, Google Cloud and Oracle Cloud Infrastructure. The new cloud technology will bring NVIDIA DGX AI supercomputers "from the browser to every company" through a partnership with Microsoft Azure, Google Cloud and Oracle Cloud Infrastructure.
NVIDIA will be working with cloud service vendors to enable customers to use web browsers to be able to use DGX computers through the NVIDIA DGX Cloud to train and deploy large language models or complete other AI workloads.
NVIDIA has already partnered with Oracle, Microsoft Azure is expected to begin hosting the DGX Cloud next quarter, and Google Cloud will soon join the effort by offering DGX Cloud services on a hosted basis to enterprises that are interested in building new products and developing AI strategies.
According to Jen-Hsun Huang, this partnership brings NVIDIA's ecosystem into the hands of cloud providers, while expanding NVIDIA's market size and reach. Enterprises will be able to rent DGX cloud clusters on a monthly basis, ensuring they can quickly and easily scale large multi-node AI training.
At the same time, Jen-Hsun Huang also launched NVIDIA AI Foundations, which consists of a series of cloud services for customers who need to build, refine and operate custom LLMs and generative AI trained with proprietary data and used for domain-specific tasks. It will also partner with Adobe's to build a set of next-generation AI capabilities for the future of creativity.
AI Foundations' services include NVIDIA NeMo, for building text-to-text generative models; Picasso, a visual language modeling service for users who want to build models trained on licensed content; and BioNeMo, to help biomedical researchers.
For the metaverse space, NVIDIA introduced the third generation of the OVX computing system and a new generation of workstations to power the NVIDIA Omniverse Enterprise based large-scale digital twin.
By combining a dual CPU platform, BlueField-3 DPU, L40 GPU, two ConnectX-7 SmartNICs and NVIDIA Spectrum Ethernet platform, the third-generation OVX servers deliver breakthrough graphics and AI performance to accelerate applications such as large-scale digital twin simulation, which in turn improves operational efficiency and predictive planning capabilities .
Enterprises can leverage OVX performance to collaborate on visualization, virtual workstations and data center processing workflows.
In addition, Jen-Hsun Huang announced updates related to NVIDIA Omniverse, NVIDIA's platform for building and operating metaverse applications, adding a range of generative AI, simulation and emulation-related features to enable developers to more easily deploy industrial metaverse applications.
Platform-as-a-Service (PaaS) NVIDIA Omniverse Cloud is now available to select enterprises, enabling them to unify digitalization across their core products and business processes.
In addition to software, NVIDIA's hardware "highlight" is a technology "nuke" that was secretly developed for four years and flung at the chip manufacturing industry - through the breakthrough lithography computing library cuLitho which accelerates computational lithography by more than 40 times, making it possible to produce 2nm and more advanced chips. TSMC, the world's largest foundry, Asmac, the world's dominant lithographer, and Synopsys, the world's largest EDA giant, are all involved in the collaboration and introduction of this technology.
At the end of the keynote, Jen-Hsun Huang thanked NVIDIA's partners and said, "Generative AI will reshape almost every industry. Together, we are helping the world accomplish the impossible."
NVIDIA is the inventor of the GPU and a leader in artificial intelligence computing. Since this year, the cross research of quantum computing + AI is increasing, and with NVIDIA's GPU update and entry into quantum computing at the GTC conference this time, the integration of the two will be further accelerated.
[1]https://www.nvidia.com/gtc/?ncid=pa-srch-goog-729287-prsp
[2]https://hothardware.com/news/nvidia-gtc-2023-keynote-ai-everywhere-all-at-once
