I was reading GH200 whitepaper and had a query regarding memory access. Why grace cannot access hopper memory allocated using cudaMalloc if both share the same page table?
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Accessing GPU heap | 1 | 862 | August 12, 2015 | |
| cudaMalloc and sharing between CPU threads | 0 | 4344 | May 20, 2009 | |
| Difference between cudamallocmanaged and malloc/new | 2 | 69 | February 20, 2025 | |
| No pagable memory access on GTX 1050 Ti under Linux? | 1 | 1099 | January 20, 2017 | |
| Problem with cudaMallocHost and structs | 4 | 4902 | August 6, 2008 | |
| Mapping CUDA memory to a GL buffer object | 1 | 1007 | May 13, 2016 | |
| How to access the memory that is allocated using cudaMallocHost from cpu? | 1 | 703 | September 4, 2022 | |
| Performance decrease on Unified GracehopperC | 3 | 90 | February 7, 2025 | |
| Page-Locked Host Memory without using cudaHostAlloc() | 1 | 1014 | February 17, 2011 | |
| Reporting a problem with CUDA memory access in multiple OS threads | 4 | 4895 | April 30, 2007 |