Monitor GPU memory

Add a way to monitor GPU memory

here is an interesting article on the matter
http://nasutechtips.blogspot.fr/2011/02/how-to-get-gpu-memory-size-and-usage-in.html

---

ATI gives you free memory for meshes, textures, and framebuffers, but NVIDIA gives you free + available memory for VRAM in general.

We can expose them in the "Statistics" class (Renderer.getStatistics()) or as methods on the Renderer interface (Renderer.freeMemory(), Renderer.maxMemory()), or we can do both.

1. Free memory, returns the sum of param[0] for the 3 memory pools on ATI. Returns "GPU_MEMORY_INFO_TOTAL_AVAILABLE_MEMORY_NVX" on NVIDIA.

2. Max memory, returns "GPU_MEMORY_INFO_DEDICATED_VIDMEM_NVX" on NVIDIA and returns -1 on ATI (not available).

---

I think that the rather static information on the maximum available memory is moot and should not be exposed via the API.
What is more interesting to the developer is the information on the currently available free memory.

Perhaps replacing the value for the maximum available free memory with the initially available free memory on application start or a value based on a recurring snapshot type of thingy would allow the developer to monitor overall graphics memory usage better than just comparing it to the maximum available memory. For example, prior to a level load, the developer could instruct the RenderStatistics to make a snapshot of the currently available free memory. After loading all level relevant textures, shaders etc. the developers would then be able to compare this to the previous snapshot, giving them an even more acurate number.

Of course there is the fact that the memory available to the GPU is shared by multiple processes and the information returned by the API is at best random, unless you run your application in a very restricted environment, e.g. single process (your game) plus operating system.

So, in favour of the getAvailableMaxMemory() API call I would recommend to allow the user to make snapshots of the currently available free memory by calling RenderStatistics#snapshotAvailableFreeMemory() and then have the RenderStatistics display and return the last snapshot by a call to for example RenderStatistics#getAvailableFreeMemoryFromSnapshot(), until a new snapshot is being made. Or something similar to that functionality.
In addition I would add the RenderStatistics#getAvailableFreeMemory() method for the currently available free memory.

I don't know why NVIDIA included this functionality into their extension, but repeatedly calling upon the GL API for the same static/unchanging information is somewhat moot, and it definitely gives the user no further value in estimating the memory performance of the application.

And, displaying a -1 for ATI cards does not help to promote these very fine cards 😀

---

Well, I don't think these methods should be on RenderStatistics at all. As you say, it's something you want to check periodically about the "renderer" so no reason to involve this "statistics collection" class. Just ask the renderer when you want the information.

If we want to show it in the debug HUD then that's fine but it doesn't have to be in RenderStatistics to do that.

---

It seems that the maximum available memory can also be queried for ATI cards:

http://www.geeks3d.com/20100531/programming-tips-how-to-know-the-graphics-memory-size-and-usage-in-opengl/

---

RenderStatistics was actually a place holder that came to my mind. It does not necessarily have to be retrieved from that class unless it itself will collect the information or the Renderer writes out the information to the render statistics collection periodically.

Reference on googlecode: https://code.google.com/p/jmonkeyengine/issues/detail?id=580

[shadowislord]

Its time to make a decision on this ... Given that there's no platform-independent way of getting available VRAM and also that it differs from platform to platform, there's no point in implementing this. Closing this issue as well as the PR.