The memory usage of the same program varies significantly across different GPUs, with the RTX2060S occupying approximately 6.7GB and the A4000 occupying around 11.5GB. It is intriguing why there is such a disparity in memory consumption for identical programs. Additionally, it should be noted that both systems are running on Windows 10 operating system. Has anyone else encountered a similar situation?
That is not surprising, is it? The program could allocate more memory if the used GPU has more memory available.
for example, greedy allocators in tensorflow do that.