PyTorch SymmetricMemory: Harnessing NVLink Programmability with Ease

XiaoSX May 25, 2025, 11:56am 3

I think they’re referring to atomicCAS based barrier update.

Topic		Replies	Views
Memcpy based P2P communication for pipeline parallelism instead NCCL distributed	9	2127	September 4, 2024
Meta PyTorch Team 2025 H1 Roadmaps	17	7518	June 24, 2025
Overlapping device to host copy with GPU collectives NVIDIA CUDA	5	890	June 4, 2024
FSDP & CUDACachingAllocator: an outsider newb perspective distributed	10	10398	December 13, 2024
Memory operations on a custom backend hardware-backends	4	1356	July 5, 2022