NVIDIA is making a slew of product and other AI-related announcements this week during its annual GTC conference. But not to be overlooked are several announcements related to their quantum computing efforts which cover new supercomputer projects, academic initiatives, quantum clouds, and PQC acceleration. We will describe all of these announcements in this article.
First, the company announced projects at two supercomputing centers to create large quantum emulation capabilities using their H100 Hopper architecture GPU and CUDA-Q programming platform. The first, called ABCI-Q, will be built by Fujitsu and located at the G-QuAT ABCI Supercomputing Center at Japan’s National Institute of Advanced Industrial Science and Technology in Tsukuba. It will consist of over 2000 H100 GPU connected by Infiniband and will be used to research new quantum algorithms as well as techniques for integrating classical and quantum computers in a hybrid configuration. The size of this installation will make it one of the most powerful systems in the world for using classical technology to emulate quantum processors. The installation is expected to be completed early in 2025. A press release with this announcement can be seen here.
A second installation is will take place at the Danish Centre for AI Innovation with a system called Gefion. This project will be funded by the Novo Nordisk Foundation and the Export and Investment Fund of Denmark (EIFO) with funding of 700 million Danish Kroner ($102M USD). The system will include 191 NVIDIA DGX H100 systems (individual computer systems) with a total of 1,528 NVIDIA H100 Tensor Core GPUs and 382 Intel Xeon Platinum CPUs connected via the NVIDIA’s Infiniband solution. The system will include the NVIDIA CUDA-Q programming platform and be used for quantum emulation as well as other non-quantum HPC application. The installation will be performed by Eviden, an Atos Group company, and the system should be ready for pilot projects by the end of the year. A news release provided by the Novo Nordisk Foundation announcing this system can be accessed here.
The next announcement is NVIDIA’s Quantum Cloud. NVIDIA calls this a microservice that will allow users to program and test their applications using the CUDA-Q programming platform and they run them with several of the major cloud providers. The service will support integrations with third party software including a Generative Quantum Eigensolver (GQE) developed with the University of Toronto, Classiq’s software suite for higher level quantum programming, and QC Ware’s Promethium software for quantum chemistry problems. The service will initially support running program emulations on NVIDIA GPUs located at the cloud providers will support running programs an actual quantum processors in the future. The service is available now and is current free. One can apply for early access at https://developer.nvidia.com/quantum-cloud-early-access-join. An announcement of this new microservice has been posted on the NVIDIA website here.
In order to help accelerate the training of a quantum trained workforce, NVIDIA is partnering over 20 universities to develop courseware to teach quantum concepts and programming to learners in a program they call NVIDIA CUDA-Q Academic. The courseware will include workshops, lectures, exercises, Jupyter notebooks, and training in CUDA-Q. The company has been supplementing this effort by sponsoring hackathons with the latest one being QHack which occurred in February. A brief description of this new effort is contained in a blog post from NVIDIA available here.
A final announcement from NVIDIA deals with a software library for accleration of Post Quantum Cryptography (PQC) called NVIDIA cuPQC. One of the issues with the new Post Quantum Cryptography algorithms being standardized by NIST is that they typically take more computer cycles to process. Although these algorithms can run on a standard microprocessor, they may take a little longer to run than the current asymmetric cryptographer like RSA. This additional latency might cause problems in timing sensitive applications so there could be a need for some acceleration. To provide a solution for this, NVIDIA has developed a software library for implementing the Kyber PQC algorithm being standardized by NIST. The algorithm runs on a single H100 GPU and provides a speedup of approximately 500X over implementations on a classical CPU. A final blog announcing this development has been posted by NVIDIA and can be found here.
So although the NVIDIA’s major focus at the QTC conference is their AI related products, they haven’t neglected quantum technology. They have indicated they are working with over 160 quantum partners including more than 90% of the largest startups. They also indicated that over 78% of the available QPUs are able to support CUDA-Q and are working with 15 of 17 available quantum frameworks.
March 18, 2024
