NVIDIA is stepping on the accelerator by aiming for qubits and, no doubt, it’s not without reason. First because quantum computing is the future and then because the shortcomings of these computers must be overcome by other systems. With this in mind, the company has introduced cuQuantumits plan to unite GPU, QPU and software in a high-precision HPC-focused system that may change the future of this industry.
A hybrid solution is NVIDIA’s main approach with its new cuQuantum project. What they are trying to achieve is improved integration which is enhanced by deep learning algorithms and which blends quantum and classical systems through the tensor base units that integrate their GPUs. Innovative? Pretty good if we’re being honest.
NVIDIA cuQuantum, artificial intelligence to unite systems
It will be their current and future GPUs based on Ampere and Ada Lovelace, as well as Hopper of course, which will be the starting point for said unification. It’s not so much the architecture, that too, but the Tensor Cores units that will do the quantum calculations because of their parallelizable capacity so that the units QPUs (Quantum Processing Units) take advantage of work where current GPUs are very good and where Tensor Cores take great advantage as NVIDIA already does in its H100, A100 and other models.
What is sought is that certain workloads are optimized in the GPU, such as circuit optimization, calibration and error correction, where once the work is solved in less time than a computer quantum, the result will be offered to him, reducing the bottleneck that they currently due to their nature.
The goal is for a series of deep learning algorithms to optimize these cross-system workloads, where Tensor Cores logically come into play for this, so that the latency between them is reduced and new levels of performance are achieved.
The project is a reality and has been opened to more companies
Apparently the first to test it seemed to be the servers of Amazon web servicebecause it seems that your service support allows integration with cuQuantum. The data that has been offered is really impressive, since NVIDIA has managed to accelerate up to 900 times quantum machine deep learning workloads.
They have also worked with IBM with Classic and simulation world records have been broken using Huang’s company’s DGX equipment. Performance and optimization are therefore real and this is a new market in which NVIDIA wants to break into as it did in its time with HPC and CUDA, where it is now ahead of Intel and AMD.
cuQuantum is the manufacturer’s bet for the quantum realm and judging by the early data, not only is it coming sooner, but it will be really difficult for Intel and AMD to design a software platform in record time to be a threat immediate for NVIDIA. You just have to remember how much time Lisa Su needed for their GPU Instinct and how the greens have eaten up so much of their ground that now the HPC market is practically theirs. So everything indicates that history is repeating itself with cuQuantum.