There is apparently some allocation granularity in CUDA. The details of it are not published. Here are some examples of related questions/discussion.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| cudaMallocManaged() not allocating memory in device memory | 4 | 1958 | August 22, 2018 | |
| sth wierd about managed memory and free GPU memeory size | 2 | 644 | November 25, 2019 | |
| CudaMallocmanaged() can not exceed more than 65410 iterartions | 1 | 582 | July 13, 2016 | |
| Difference between host cudaMalloc() and kernel malloc() | 1 | 12576 | April 21, 2011 | |
| Unified Memory Allocation Alignment on Windows | 0 | 548 | August 23, 2020 | |
| Inconsistency in allocating maximum memory using cudaMalloc() | 1 | 3491 | August 3, 2011 | |
| Managed memory vs cudaHostAlloc - TK1 | 6 | 2008 | February 15, 2016 | |
| cudaMalloc() | 0 | 838 | October 9, 2013 | |
| Information about Cuda Memory Consumption on TK1, problem with cudaMemGetInfo() | 3 | 586 | October 18, 2021 | |
| shared memory dynamic allocation ? | 4 | 2896 | December 11, 2009 |