Nvidia DGX Quantum system blends CPUs, GPUs with CUDA | TechTarget
A new system combining Nvidia’s Grace Hopper Superchip and Quantum Machines’ OPX platform provides scientific researchers and bleeding edge developers with a way to build apps that work across classical and quantum systems.
The new DGX Quantum system, which Nvidia and Quantum Machines claim is the first GPU-accelerated quantum server, not only lets developers build more powerful apps that run on classical and quantum technologies but also provides capabilities including improved calibration, control system, quantum error correction and hybrid algorithms.
The Nvidia Grace Hopper Superchip is connected to the Quantum Machines OPX+ platform via a PCIe cable, which permits sub-microsecond latency between GPUs and the quantum processing bits (QPUs), a Nvidia spokesperson said.
The Grace Hopper Superchip combines Nvidia’s GPU architecture with its Grace CPU. It is designed for large scale AI and HPC applications, delivering higher performance for applications involved with running terabytes of data needed to solve complex problems.
Quantum Machines’ OPX+ quantum control system bridges classical compute engines and quantum control stacks that enable the added speed and performance of both QPUs and quantum algorithms.
Both can be appropriately scaled to systems ranging from a few qubit QPUs to a quantum-accelerated supercomputer, the companies said.
These systems are of interest to “supercomputer centers and service providers that have already deployed quantum processors and are serious about connecting them up to classical processors,” said Sam Stanwyck, quantum computing product lead at Nvidia.
“Cloud providers, too, have expressed an interest — particularly those a little behind the curve that want to catch up with a more robust cloud-based model,” he said.
Last summer Nvidia delivered Quantum-Optimized Device Architecture, or QODA, to support hybrid quantum-classical system applications. The partnership with Quantum Machines goes a step further.
“Instead of using a classical microprocessor, they are using the GPU to help drive performance that should give them a better response time between the classical and quantum platforms,” said Doug Finke, chief content officer at Global Quantum Intelligence. “Connecting the two platforms with a PCIe cable also helps.”
Another analyst agreed that the timing is right for the arrival of DGX Quantum, given the path being followed by Nvidia competitors such as IBM. It plans to deliver a quantum-centric supercomputer this year.
Doug FinkeAnalyst, Inside Quantum Technology
“It’s a natural step for Nvidia to move its development in that same direction,” said Paul Smith-Goodson, an analyst with Moor Insights and Strategy. “Plus, Quantum Machines has a hardware-agnostic platform that can be used with a quantum computer. I would say that the convergence (with classical and quantum) has begun.”
DGX Quantum also provides developers with NVIDIA CUDA Quantum, a unified software stack now available in open source.
CUDA Quantum is a hybrid quantum-classical platform that permits the integration and programming of QPUs, GPUs and CPUs that exist in a single system.
“The open source version (of CUDA) is a platform that makes it possible for a domain scientist, for example, to program different types of quantum and classical resources to work together and then optimize them to run on the new (DGX Quantum) system,” said Itamar Sivan, co-founder and CEO of Quantum Machines.
Nvidia partners committed to integrating the CUDA platform include Agnostiq and QMware along with several hardware suppliers, such as IonQ, Atom Computing, ORCA Computing and QuEra.
As Editor At Large with TechTarget’s News Group, Ed Scannell is responsible for writing and reporting breaking news, news analysis and features focused on technology issues and trends affecting corporate IT professionals.