qTPU: Hybrid Tensor Networks for Quantum-Classical Acceleration
Talkat 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26) · 2026-07-13..07-15·
summary
qTPU introduces a hybrid tensor network abstraction plus a programming model, compiler and runtime that jointly trade quantum error against classical FLOP cost. It compiles via Qiskit and NVIDIA CUDA-Q to target an A100 GPU and IBM Marrakech (156 qubits), slicing circuits and offloading tensor contractions. Measured wins include 3–4 orders of magnitude lower classical overhead, up to 7.2× lower quantum error, up to 53× faster compilation, over 20× end-to-end speedups and ~9× fidelity improvement on real devices, showing the system works at scale.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:46Hybrid app challenges
- 4:54Hybrid tensor networks
- 7:43QTPU system design
- 9:49Optimizer for trade-offs
- 11:36Runtime execution
- 12:44Evaluation of QTPU
- 15:35Conclusion
related talks
45mQuantum application development for multi-processor systems using CUDA-Q
18mIntegrating Quantum Computers in High-Performance Computing Environments
20mLarge-scale hybrid quantum computing solution of linear systems of equations
21mCUDA-QX and Artificial Intelligence for Quantum Computing
16mSolving Real Industry Problems on Today’s Quantum Computers
21mHybrid Quantum Computing for AI
27mQuantum Machine Learning With TensorFlow Quantum
17mQVM: Quantum Gate Virtualization Machine