# GPU RAM fragmentation diagnostics **URL:** \ **Category:** Uncategorized\ **Created:** [January 8, 2019, 1:11am UTC](https://discuss.pytorch.org/t/gpu-ram-fragmentation-diagnostics/34073 "2019-01-08T01:11:29Z")\ **Posts on this page:** 1\ **Showing post:** 9
**Author:** ![stas](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/stas/32/8409_2.png) [@stas](https://discuss.pytorch.org/u/stas)\ **Post date:** [January 23, 2019, 2:32am UTC](https://discuss.pytorch.org/t/gpu-ram-fragmentation-diagnostics/34073/9 "2019-01-23T02:32:15Z")
> [@colesbury](#): > > > But then it won’t let me allocate memory smaller than a page if that page is already partially used, am I correct? > > No – memory smaller than a page may use a partially filled page. It just won’t combine multiple partially filled pages together. OK, then I could write an app that would go and fill out all the little nooks and crannies and then dump a summary. It sounds like it’d be the easiest to do directly in C to avoid any overheads that might get incurred by using pytorch. > [@colesbury](#): > > > How would you recommend I then try to diagnose a situation that lead me to start on this path? My idea was to write a sort of memory mapper which I would bisect/splice my code with and through its printouts find the guilty parts that caused that much fragmentation. > > I would just record the sizes of the malloc and free calls to a log. (Edit THCCachingAllocator.cpp). I’m not sure how could that help with detecting fragmentation. Unless you mean looking at mallocs that are lesser than a page size and then accounting the remaining fragments, but that sounds like it’d be very unreliable/difficult to accomplish since it’d require an exact understanding of how CUDA does things internally. Basically this is what I did in my 2nd post, except on the macro-level. > [@colesbury](#): > > > Does the content of this error come from CUDA or pytorch? > > It comes from PyTorch. Cached in this case is cached but not allocated. I think this is diverging from the topic now, so I will create a new one for that and keep this one focused. --- _[View the full topic](https://discuss.pytorch.org/t/gpu-ram-fragmentation-diagnostics/34073)._