Quantum computing is a promising technology that offers many scientific leaps, namely simulating molecules, boosting drug discovery efforts. Many teams are now researching quantum computer technology and trying to make it perform precise and consistent calculations. And many of these teams are creating quantum circuit simulations to accelerate these innovations. Namely, Nvidia has announced their cuQuantum software development kit at the GTC 2021. This kit will run all of the simulations on the Nvidia GPUs, the GPU with its speed offers many advantages in running complex algorithms comparing the classical processors.
What is Decoherence?
Quantum is our inevitable future, but there are some issues left to be solved concerning these powerful computers. And one of these issues is decoherence. Decoherence is the falling out of quantum states. This is a limiting factor when it comes to quantum computers. The decoherence simply corrupts the functionality of the quantum circuits.
The quantum computers work with qubits, unlike the bits that we are using today. A bit can be either 0 or 1. A qubit can be both, that’s why these qubits are so special and quantum technology is so desired.
The cuQuantum SDK will try to solve the problem with the decoherence by accelerating the quantum circuit simulators which are going to help the researchers to design better quantum computers and more optimal algorithms.
This SDK will also provide some tools that will offer the developers to apply their own custom methods and will support different approaches. Namely, the state vector method and the tensor network method.
State Vector Method
A group of researchers at the Julich Supercomputing Center have incorporated the state vector method for the purpose of simulation of physical realizations of quantum machines on GPUs. They showed an x25 speedup compared to regular CPU-equipped systems.
QCWare, another company that is working in the quantum sphere has worked together with NVIDIA and they have reported compelling evidence that the use of GPUs makes a significant difference when running the quantum approximate optimization algorithm, simulating 20 qubits in the process.
A single NVIDIA DGX A100 equipped with eight NVIDIA A100 80 Tensor core GPUs is capable to simulate even up to 36 qubits. Delivering quite a lot of advantages when compared with regular CPUs.
Besides Julich and QCWare there are other companies that are running these vector simulators on NVIDIA GPUs. These include IBM, Amazon Web Services, Oxford Nanopore, and the NVIDIA AI centre.
Tensor Network Method
The tensor network simulators are more advanced and use a new method that is using much less memory, and also offers a bigger computation power than the state vector.
NVIDIA and Caltech have created a quantum simulator with cuQuantum that runs on NVIDIA A100 Tensor Core GPUs. This setup generated a sample in 9.3 minutes. This sample was generated on the NVIDIA Selene supercomputer and made a full circuit simulation of the Google Sycamore circuit. This 9.3-minute-long simulation previously was expected to take days.
Density Matrix Sims
Researchers at the Pacific Northwest National Laboratory, Washington State University, and Lehigh University have worked on a project that goal was to create a new multi-GPU methodology. They named this methodology MG-BSP. And they used this tech to build a density matrix quantum simulator.
Their research has shown that the team had demonstrated a simulation of 1 million general gates in a test that lasted 94 minutes and ran on an NVIDIA DGX-2.
This research has shown that the density matrix simulator is around ten times faster than state vector quantum simulators according to the group of researchers.
In recent news, Tesla has also unveiled a new supercomputer powered by A100 NVIDIA GPUs, this computer will boost the company’s efforts into the quantum sphere.