arXiv:2504.15465v1 [cs.OS] 21 Apr 2025.
Patrick H. Coppock, Brian Zhang, Eliot H. Solomon, Vasilis Kypriotis, Leon Yang† , Bikash Sharma† , Dan Schatzberg† , Todd C. Mowry, Dimitrios Skarlatos
Carnegie Mellon University
† Meta
The surging demand for GPUs in datacenters for machine learning (ML) workloads has made efficient GPU utilization crucial. However, meeting the diverse needs of individual ML models while optimizing resource usage is challenging. To enable transparent, fine-grained management of GPU resources that maximizes GPU utilization and energy efficiency while maintaining strong isolation, an operating systems (OS) approach is needed. Hence this paper introduces LithOS, a first step towards a GPU OS.
LithOS includes the following new abstractions and mechanisms for efficient GPU resource management: (i) a novel TPC Scheduler that supports spatial scheduling at the granularity of individual TPCs, unlocking efficient TPC stealing between workloads; (ii) transparent kernel atomization to reduce head-of-line blocking and allow dynamic resource reallocation mid-execution; (iii) a lightweight hardware rightsizing mechanism that dynamically determines the minimal TPC resources needed per atom; and (iv) a transparent power management mechanism that reduces power consumption based upon in-flight work characteristics.
We implement LithOS in Rust and evaluate its performance across a broad set of deep learning environments, comparing it to state-of-the-art solutions from NVIDIA and prior research. For inference stacking, LithOS reduces tail latencies by 13× compared to MPS; compared to the bestperforming SotA, it reduces tail latencies by 3× while improving aggregate throughput by 1.6×. Furthermore, in hybrid inference-training stacking, LithOS reduces tail latencies by 4.7× compared to MPS; compared to the best-performing SotA, it reduces tail latencies by 1.18× while improving aggregate throughput by 1.35×. Finally, for a modest performance hit under 4%, LithOS’s hardware right-sizing provides a quarter of GPU capacity savings on average, while for a 7% hit, LithOS’s transparent power management delivers a quarter of a GPU total energy savings on average. Overall, LithOS transparently increases GPU efficiency, establishing a foundation for future OS research on GPUs.
FULL TR: pdf