盘点来自工业界的GPU共享方案 - 知乎

针对深度学习的GPU共享 - 知乎

Microsecond-scale Preemption for Concurrent GPU-accelerated DNN Inferences | USENIX

SJTU-IPADS/reef: REEF is a GPU-accelerated DNN inference serving system that enables instant kernel preemption and biased concurrent execution in GPU scheduling.

Transparent GPU Sharing in Container Clouds for Deep Learning Workloads | USENIX

pkusys/TGS: Artifacts for our NSDI'23 paper TGSmulti

Using CUDA IPC memory handles in pytorch - PyTorch Forums

这里面有IPC方案