U.S. bans sales of high-end GPU chips to China, Nvidia expected quarterly loss of $400 million
This week, the United States imposed new trade restrictions on China, banning Nvidia and AMD from selling cutting-edge high-performance computer and artificial intelligence technology to China. In a filing with the U.S. Securities and Exchange Commission, NVIDIA revealed that it is prohibited from exporting its A100 and upcoming H100 GPUs to China and Russia [1]. NVIDIA said the move would cost the company $400 million.
AMD reported that it also received instructions from U.S. authorities to stop sales of its high-end GPU chip, the Instinct MI250, to China and Russia. a variant of the chip, the MI250X, powers the U.S. Department of Energy's Frontier supercomputer, which is ranked No. 1 in the world.
Nvidia shares fell nearly 8 percent on Sept. 1, and AMD also fell 3 percent.
The new export rules imposed on NVIDIA apply to its entire A100 and H100 product lines, as well as to systems using these technologies, including NVIDIA's own DGX systems and the HGX platform used by many partner system manufacturers. In order to export the restricted technology, NVIDIA needs to apply for a license from the U.S. Department of Commerce.
Documents from NVIDIA show that "DGX or any other system containing the A100 or H100 integrated circuits and the A100X are also included in the new licensing requirements. The licensing requirements also include any future NVIDIA integrated circuits with peak performance and chip-to-chip I/O performance equal to or greater than thresholds roughly equivalent to the A100, and any system containing those circuits."
In this regard, Foreign Ministry spokesman Wang Wenbin said that the U.S. approach is typical of scientific and technological hegemony, the U.S. side has repeatedly generalized the concept of national security, abuse of state power, trying to use their own scientific and technological advantages to contain and suppress the development of emerging markets and developing countries, which violates the rules of the market economy and undermines the international economic and trade order. China is firmly opposed to this.
NVIDIA said it is "engaging with customers in China" and "seeking to meet their planned or future purchases of the company's data center products, which are not subject to the new licensing requirements."
Meanwhile, AMD has been told it cannot ship its MI250 gas pedals to China and Russia, and AMD believes its MI100 chips will not be affected. Reuters quoted an AMD spokesman as saying the move is not expected to have a material impact on the company's business.
However, Nvidia is at greater risk. In a recent filing with the U.S. Securities and Exchange Commission, NVIDIA said that "the company could lose $400 million in potential sales in China if customers do not want to buy the company's replacement products or if the U.S. government does not issue licenses in a timely manner or refuses to issue licenses to important customers." NVIDIA said it has sold $400 million of the affected chips to China this quarter.
Nonetheless, NVIDIA said it will be allowed to continue its H100 development program in mainland China and Hong Kong. "The U.S. government has authorized exports, re-exports and domestic transfers to continue development of the H100 integrated circuit. The authorization allows the company to provide exports required to support U.S. customers of the A100 through March 1, 2023. In addition, the U.S. government has authorized the fulfillment of A100 and H100 orders and logistics through the company's Hong Kong facility through September 1, 2023."
This is another in a series of U.S. measures against China's technology embargo. in 2015, Intel was denied an export license to provide its high-performance processor technology to China's supercomputing program. That restriction prompted China to significantly increase its investment in indigenous high-performance computing technology.
01NVIDIA: A key player in quantum computing
NVIDIA is the inventor of the GPU and a leader in quantum computing.
In 2021, NVIDIA released an SDK called cuQuantum, which provides simulation capabilities on NVIDIA's latest GPUs. The toolkit provides a number of API interfaces that allow users to create quantum programs in common frameworks such as Qiskit, Cirq, ProjectQ, Q#, and others, and then simulate them on platforms that include the latest generation of NVIDIA GPUs.
NVIDIA cuQuantum
SDK currently supports two different simulation methods. The state vector simulator provides high fidelity results, but requires a memory space that grows exponentially with the number of quantum bits, limiting the total number of quantum bits that can be simulated. The tensor network approach trades memory footprint for memory, slightly reducing fidelity to simulate programs with more quantum bits. NVIDIA will continue to develop this SDK and will release additional simulation capabilities in the future.
Today, dozens of quantum organizations are already using the NVIDIA cuQuantum software development kit to accelerate their quantum circuit simulations on GPUs.
In May 2022, AWS announced the availability of cuQuantum in its Braket service. it also demonstrated on Braket how cuQuantum can provide up to 900x acceleration on quantum machine learning workloads. Moreover, cuQuantum now enables accelerated computing on major quantum software frameworks, including Google's qsim, IBM's Qiskit Aer, Xanadu's PennyLane, and Classiq's Quantum Algorithm Design Platform. This means that users of these frameworks can access GPU acceleration without any additional coding.
In Europe, NVIDIA and SiPearl are collaborating to expand the developer ecosystem for building tens of billions of subcomputers on Arm. This work will help users in the region port their applications to systems using SiPearl's Rhea and future Arm-based CPUs and NVIDIA accelerated computing and networking technologies.
The Japan Center for Computational Science is integrating the NVIDIA Quantum-2 InfiniBand platform on NVIDIA H100 Tensor Core GPUs and x86 CPUs. The new supercomputer will address work in areas such as climatology, astrophysics, big data, and artificial intelligence.
HPC users are adopting NVIDIA technologies because they offer the highest application performance for established supercomputing workloads (simulation, machine learning, real-time edge processing) as well as emerging workloads such as quantum simulation and digital twins. However, NVIDIA says there is still room for improvement in this technology.
To program quantum processors (QPUs), researchers are forced to use the quantum equivalent of low-level assembly code, which is impossible for scientists who are not quantum computing experts; furthermore, developers lack a unified programming model and compiler toolchain that would allow them to run jobs on any QPU.
This situation needs to change, and it will change. As quantum systems evolve, the next big leap is to move to hybrid systems: quantum and classical computers working together.
On July 12, 2022, NVIDIA released a quantum version of its unified computing platform, QODA (Quantum Optimized Device Architecture), for accelerating breakthroughs in quantum R&D in artificial intelligence, high-performance computing, health, finance, and other disciplines.
QODA will be used in multiple applications
NVIDIA says QODA is inherently interoperable with existing classical parallel programming models such as CUDA, OpenMP and OpenACC. With QODA, HPC and AI domain experts can easily add quantum computing to existing applications, leveraging today's quantum processors, as well as simulated future quantum machines using NVIDIA DGX systems and the vast array of NVIDIA GPUs available in scientific supercomputing centers and public clouds.
A hybrid solution combining classical and quantum computing will enable scientific breakthroughs in the short term," said Tim Costa, product lead for high performance computing and quantum computing at NVIDIA. qODA will revolutionize quantum computing by providing developers with a powerful and efficient programming model."
Reference link:
[1]https://www.hpcwire.com/2022/09/01/us-bars-nvidia-and-amd-from-selling-top-gpus-into-china/
