PT-3 goes live in London
ORCA Computing, an Oxford spin‑out focused on quantum‑accelerated AI, has launched the PT-3, its latest photonic quantum computer. The first PT-3 is installed at Digital Realty’s Innovation Lab (DRIL) in London, where it operates next to NVIDIA accelerated computing as a hybrid platform.
Key practical specs:
- Runs at room temperature (no cryogenics)
- Draws about 6 kW across two standard racks
- Becomes operational within hours of installation
- Modular and field‑upgradeable, so performance can be increased in place
The system is explicitly aimed at commercial quantum advantage: not lab demos, but workloads where classical methods are already hitting limits.
Quantum + GPUs for AI in science
PT-3 is designed to sit alongside GPU clusters and boost AI for science, especially generative models that struggle with limited training data.
In collaboration with NVIDIA, ORCA extended the NVIDIA Proteína protein‑design model with a hybrid quantum–classical architecture:
- Training used 32 NVIDIA Hopper GPUs plus a PT-3 configured with 64 photons in 128 qumodes
- The hybrid model improved the ratio of useful proteins by about 20% over a classical baseline
- It also generated novel proteins with a different mix of secondary structures
These results build on earlier work with the PT-2 system, which doubled the number of strong peptide binders versus classical AI and produced 49% more valid, unique molecules on the QM9 benchmark.
Industrial‑scale optimisation: energy scheduling
Beyond life sciences, PT-3 targets large optimisation problems such as power‑grid scheduling. In a collaboration with JIJ Inc., bp and the UK’s National Quantum Computing Centre (NQCC), ORCA benchmarked hybrid solvers on a 25,755‑variable unit commitment instance.
On that benchmark:
- PT-3 now runs the hybrid quantum–classical workflow in about one hour
- That is more than 12× faster than its predecessor at the same solution quality
- In white‑paper tests, the hybrid solver’s results were competitive with a leading commercial optimisation solver, and with more runtime it found the best solutions of any method evaluated
Why it matters
PT-3 illustrates a pragmatic route to hybrid quantum supercomputing:
- Integrates directly with NVIDIA GPU infrastructure instead of replacing it
- Targets protein design, energy, logistics and defence use cases where better candidates or schedules have immediate value
- Is delivered as data‑centre hardware that can be incrementally upgraded as photonic technology improves
For engineers and researchers, the message is that quantum is starting to appear not as a black‑box cloud API only, but as a co‑processor in real racks next to classical accelerators, focused on concrete AI and optimisation workloads.
Source: ORCA Computing, Electronics Weekly
