We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 – 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!
Today, at the Q2B conference in Tokyo, GPU and AI kingpin Nvidia is announcing QODA — its Quantum Optimized Device Architecture, designed to create a single programming environment for hybrid classical-quantum computing.
Similar in overall aim (and name) to Nvidia’s CUDA (Compute Unified Device Architecture) platform for parallel computing development, QODA takes the highly specialized quantum development discipline and makes it accessible to a broader range of software developers. But the plotline for Nvidia GPUs in the quantum world is more nuanced than it is even in AI, and QODA’s goal is to make it straightforward.
Brave new quantum world
“It’s a very different world than it was a decade ago,” Timothy Costa, Nvidia’s director of HPC and quantum computing products, told VentureBeat. Costa explained what’s behind the progress the quantum industry has made: “What we see is the industry going from one- or two-qubit systems, most of them in academia, up to today, to systems with 200+ qubits based in the cloud.”
Qubits are the rough equivalent of bits in classical computing, but while they can be read as having a value of zero or one, qubits can have multiple values simultaneously, making them and the hardware that instantiates them the essence of quantum computers.
QODA welcomes all developers aboard
QODA’s credo is helping nonquantum-specialized developers take advantage of this industry progress. Specifically, it’s aimed at developers focused on particular domains, including drug discovery, chemistry, finance and optimization (as a general computing technique), where quantum can accelerate things and make it feasible to attack problems that would otherwise be computationally impractical to address. These areas benefit best from a combination of classical computing (albeit in the powerful form of HPC — high performance computing) and quantum.
Nvidia’s GPU technology is already a dominant platform in the HPC world, of course. But it turns out to have specific applicability on the quantum side as well. That’s because, while GPUs aren’t quantum hardware, they may serve as a more effective medium for quantum circuit emulation than CPUs, since GPUs can implement state vector and tensor network methods, which accelerate quantum circuit simulations.
In effect, this means a big GPU system, like Nvidia’s DGX platform, may be able to handle hybrid scenarios especially well, since it offers one physical infrastructure layer that can service both classical and quantum computing workloads.
QODA addresses this new “split personality” potential of GPUs by offering a single platform for hybrid development. Underlying this is Nvidia’s cuQuantum SDK and its DGX Quantum Appliance. The cuQuantum SDK allows developers to simulate quantum circuits on GPUs. It includes integration with quantum computing frameworks Cirq, Qiskit and Pennylane. The DGX Quantum Appliance is a software container that integrates the frameworks with cuQuantum and runs on any Nvidia hardware.
With these technologies underlying it, QODA provides two things to help make quantum computing more accessible to conventional developers:
- A kernel-based programming model for quantum computing development with interfaces for common programming languages, such as C++ and Python,
- A compiler that can accommodate quantum and classical computing-oriented instructions comingled in the same source code, as seen in the figure below.
Hybrid coding example with block of quantum code on top and GPU-oriented code below.
Combining virtual and physical
QODA and cuQuantum work with emulated QPUs (quantum processor units) on GPU hardware, but they work with physical QPUs as well, so code written on the platform is portable between emulated and physical environments.
In fact, QODA and cuQuatum were developed in partnership with numerous vendors in the quantum space, including hardware partners like IQM Quantum Computers, Pasqal, Quantinuum, Quantum Brilliance and Xanadu; software/algorithm partners like QC Ware and Zapata Computing; and supercomputing centers including Forschungszentrum Julich, NERSC/Lawrence Berkeley National Laboratory, and Oak Ridge National Laboratory.
The diversity of hardware partners involved means QODA also works across a variety of qubit “modalities,” including superconducting, neutral atom, trapped ion, diamond processors and photonics.
What’s ahead for Nvidia and quantum
Costa told VentureBeat that with QODA, Nvidia hopes to provide developers with access to disruptive compute technology and allow domain scientists to leverage quantum acceleration, tightly coupled with the best of GPU supercomputing.
Nvidia sees QODA’s mission as getting developers who are focused on a class of applications (rather than on quantum computing itself) to use quantum and to see it as a technology that can accelerate what they’re already doing. This is a pragmatic approach to adoption of quantum computing, which could be the biggest change in computing since the introduction of the microcomputer — or maybe even the mainframe.
Nvidia’s goal with its partnering strategy with QODA is to bring together many startups with the likely effect of promoting cohesion and an ecosystem in the quantum arena. Doing so is key to helping the space mature and be more attractive for adoption by enterprise customers.
Just as Nvidia has helped make AI and autonomous cars actionable to large customers, the QODA announcement should help make quantum computing more industrialized and commercially viable.