Nvidia has just announced at their GTC event about the performance of quantum simulators using the DGX A100 and its own custom-cooked quantum development software stack, called cuQuantum.
The thing is, most of the quantum hardware makers have their own tools for quantum circuit simulation. And aside from the commercially available Atos/Bull systems, there are more than 300 quantum simulators in use today on different experimental (generally academic) platforms, according to one of the seminal survey/reviews on the state of quantum simulators.
On the same day the GPU maker announced its own CPU to tackle that highly contested market, CEO Jensen Huang made it clear Nvidia is tracking quantum requirements, albeit on the software and simulation side for now. He touched on recent work making quantum simulations faster. This means the company is putting at least some stock in where quantum is headed, enough to develop a cuDNN-like software stack to support quantum simulations.
cuQuantum aims to do for quantum what cuDNN did for deep learning, Huang said, pointing to early work using the SDK at Caltech where a state vector benchmark that took ten days on a dual-CPU server could be cut to two hours using a DGX A100 with the cuQuantum stack.
As a proof point, Nvidia reported it collaborated with Caltech to develop “a state-of-the-art quantum circuit simulator with cuQuantum running on NVIDIA A100 Tensor Core GPUs. It generated a sample from a full-circuit simulation of the Google Sycamore circuit in 9.3 minutes on Selene, a task that 18 months ago experts thought would take days using millions of CPU cores.”
This new SDK for quantum development is aimed at helping quantum device makers refine their own systems as well as to help quantum algorithm creators design and test their approaches. (TheNextPlatform, HPCWire)
