In our Applications Team at Aegiq, we work with partners to identify and solve real problems using Quantum Computing. We have recently worked on several projects using quantum algorithms for Computational Fluid Dynamics (CFD) use cases, where future fault tolerant quantum computing can provide an exponential performance increase and quantum inspired methods on conventional computers using tensor networks can provide speed ups and memory compression in transient turbulent problems today.
A tensor network is a highly efficient and structured way to represent enormous quantum states by decomposing them into small interconnected tensors, allowing powerful analysis of entanglement and complex physical systems.
Testing and iterating directly on GPU clusters or HPC can be time consuming, so I recently started using an NVIDIA DGX Spark with GB10 Grace Blackwell Superchip for testing, development and small-scale simulations. This allowed me to quickly integrate with NVIDIA accelerated computing, NVIDIA CUDA and NVIDIA cuQuantum, and importantly get my code into an oven-ready form to submit to a large GPU cluster. A user can connect to the DGX Spark using the NVIDIA Sync programme, which has a very simple GUI allowing me to connect and get coding straight away. To start using the device, all I need to do is click on the NVIDIA Sync icon in the Windows system tray, click connect and away I go. It is very user friendly.
Once up and running, I can use Jupyter notebooks which is my preferred way to code in Python and I could easily forget I was even connected to it, as it didn’t change my coding experience at all. This enabled me to perform polished lived demos using Jupyter notebooks that are accessible to clients. I have also been using it remotely from home and had no trouble connecting through a VPN. The box looks cool too – it fits on my desk nicely and makes me want to use it!
As an example of the flexibility and power of the DGX Spark, I have been using it to run testing of quantum computing tasks for fundamental condensed matter models for ~100 qubits. This utilises acceleration from NVIDIAs cuTensorNet and Aegiq’s proprietary libraries. In this context, I used the DGX Spark for fast and efficient cycles of testing and debugging which then inform larger implementations with ~1000 qubits on cloud-based data centre NVIDIA accelerated computing.