I want to know how data transfers in zero-copy (mapped pinned memory) take place from host to device and vice-versa. The Profilers don’t profile the PCI transactions. Is there any way I can track these PCI bus transactions and know the data movement?
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
load/store host data inside kernel? | 2 | 730 | February 11, 2015 | |
gpu access host memory | 1 | 646 | January 20, 2012 | |
cudaHostAllocWriteCombined vs cudaHostAllocMapped | 5 | 4102 | August 19, 2019 | |
Zero-Copy and cudaMemcpy() | 1 | 2612 | May 18, 2009 | |
zero-copy pinned memory and cuda 4.0 | 1 | 3958 | January 25, 2012 | |
Does unified memory incur double transfer? | 2 | 347 | April 6, 2022 | |
zero copy | 0 | 1782 | February 3, 2012 | |
Improving data transfer performance from host to device | 2 | 2081 | January 28, 2015 | |
"What is the best practice for CUDA data transfer (CUDA 12.5)?" | 4 | 166 | September 2, 2024 | |
CUDA + MPI: HtoD and DtoH in profiling & how to keep communication on the device | 1 | 545 | June 29, 2022 |