Q.ANT’s IONOS pilot puts photonic computing in reach of real workloads

Q.ANT and IONOS are opening a limited pilot that will let selected participants run their own workloads on a second-generation photonic processor through dedicated servers in Germany. It follows the companies’ May collaboration announcement and the release of Q.ANT’s software development kit in September. The pilot brings the hardware into reach of teams that want to assess their own code and data, following an earlier evaluation at the Leibniz Supercomputing Centre and installations at the Jülich Supercomputing Centre.

The processor performs mathematical operations on a thin-film lithium niobate photonic integrated circuit. Q.ANT points to matrix multiplication and nonlinear functions as promising workloads, since some operations can be performed in the optical domain. What the pilot can help establish is how much of a useful job can run there. If an application calls the accelerator only briefly and spends most of its time preparing inputs, moving data or processing the output on the host, a fast optical operation may make little difference to the complete job.

A comparison on a real workload would therefore need to hold output quality constant and time the job from input preparation through to the final result. For energy, the measurement boundary matters just as much: power measured at the accelerator tells a different story from energy measured at the server, where the host and the work of feeding the accelerator are included. Testing more than one workload size would also show whether any benefit depends on sending enough work to the photonic device to offset the cost of using it.

The pilot starts in November and is free for a limited number of approved applicants. Participants must propose a use case and be able to administer a bare-metal server. Q.ANT and IONOS plan to provide onboarding and technical support, with a workshop to review the results. The announcement sets out access and evaluation plans; it does not yet contain results from participants’ workloads.

Sources: Q.ANT and IONOS pilot announcement; Q.ANT company newsroom: pilot announcement.