Is it possible to run a code originally created for CPU/GPU on QPU ?
Disclaimer:
This may seem stupid question but this is typical customer voice. Sorry in advance for my dilletant approach.
I am not a quantum computing professional yet. I am not even developer. But I do some coding as a complementary for my marketing job and consider it as a hobby too.
I am about to implement ML to video editing and customer interactions (chatbots). And usually I use open source ML projects to be rebuilt and adapted for my specific usecase. All training models are actually created to be used in GPUs. But it takes days only to train the model, see the result and troubleshoot, measure, rebuild and train again. It`s quite time expensive.
One day I found that DWave democratize the access to quantum computing. Seems promising for all the community
But I found that in order to use the QPU I need to write special code for it, suitable for QPU only
Thats`why I`m wondering if it is possible to run a code originally created for CPU/GPU on QPU?
And how it could be possible?
What logic should be used? and how I need to change my common code to run it on QPU
I think that if such a solution will appear it will boost the adoption even more that it accelerates now
Thanks for your interest answering that question
Comments
Not really. Not today, at least.
I think widespread adoption will come, but only after people get familiar with new ways of solving problems. Keep in mind that the D-Wave (and quantum computing, in general) is not for solving just any problem. Some problems work well, while other problems don't really fit at all.
While it may be possible, in the future, to map GPU algorithms to equivalent QPU algorithms, the ability to do so in principle, and in with any degree of efficiency, will probably remain in the research domain for some time. This is because quantum annealing, the process performed by the D-Wave chip, is a very different animal in terms of compute paradigms. It isn't just a lot different - it is fundamentally different.
Thank you Thomas for the great response! To add a bit more detail:
At the lowest level, the D-Wave QPU is able to solve problems written in Ising or QUBO format. Thus, any problem that runs on the QPU must first be converted to one of these formats.
Many classes of problems can be successfully converted into these formats, although there are some problems that cannot. For example, constraint satisfaction and optimization problems are a good match for the D-Wave QPU as they can be effectively expressed in the right format.
To bridge this gap, D-Wave is actively developing a set of open-source tools to facilitate programming these well-suited applications. The idea is to grow these tools and get lots of input from users like yourselves on new applications.
If you're interested in learning about Ising and QUBO problems, and how more abstract problems can be converted into these formats, I recommend giving our documentation on Problem Formulation a read. You can also check out existing D-Wave Ocean tools like D-Wave Binary CSP and D-Wave NetworkX, which allow users to input problems in a more abstracted format.
Any update on this? My lab has several Nvidia GPUs so I need to use them with your QPU.
Please sign in to leave a comment.