You are trying to allocate 88MB. ~130MB are in the cache, but are not a contiguous block, so cannot be used to store the needed 88MB. 0B are free, which shows that the rest of your device memory is used to store the CUDA context, memory for other applications etc.
1 Like