Jamming lots of little things into a big thing, quickly. We have lots of images. We need them in a s

Allocating close to the 4GB memory limit as pinned memory would kill the system anyway… External Image One is not supposed to allocate large pinned memory, as it prevents the kernel from doing paging.

Allocating close to the 4GB memory limit as pinned memory would kill the system anyway… External Image One is not supposed to allocate large pinned memory, as it prevents the kernel from doing paging.

500 images of 512 Kb each is not more than 250 Mb, that should still be possible right? (there are not many GPU’s that would support a single memory allocation of 4Gb anyway. I don’t have one and I do have PC’s where I could cudaMalloc 4Gb External Image)

500 images of 512 Kb each is not more than 250 Mb, that should still be possible right? (there are not many GPU’s that would support a single memory allocation of 4Gb anyway. I don’t have one and I do have PC’s where I could cudaMalloc 4Gb External Image)

There could be any number of image stacks in play at a time. That rules out pinning them all. And my example image stack was just a common case, there are some big cases of 1000+ images of 2mb each, that’d be 2gb. (It’s a fun set of requirements…) I’m still working on this, but also got assigned to something else, (multithreaded human needed) - so it might be a while till I report success/failure…

There could be any number of image stacks in play at a time. That rules out pinning them all. And my example image stack was just a common case, there are some big cases of 1000+ images of 2mb each, that’d be 2gb. (It’s a fun set of requirements…) I’m still working on this, but also got assigned to something else, (multithreaded human needed) - so it might be a while till I report success/failure…