Hey all,
Curious for a 2020’s take on this. I’ve noticed a discrepancy between the numbers reported by GL_NVX_gpu_memory_info and the numbers reported by nvidia-smi, in terms of how much GPU memory my application has allocated. nvidia-smi reports both total usage as well as per-process usage, but neither figure lines up with the figures reported by the GL extension; in particular, the GL extension’s result is larger than the total allocation that nvidia-smi reports.
Is GL_NVX_gpu_memory_info still well-supported? Is it meant to give process-local or system-global values? (It’s not 100% clear from the spec). Or should I just be dispatching to nvidia-smi or fishing around in /proc (I’m on linux)?
Relatedly, should I be looking into GL_NV_query_resource or VK_EXT_device_memory_report instead going forward?
Thanks,
Tom