[GTX TITAN/RTX 2080Ti] Unexpected Memory Consumption when using resized TilePool ( with Reprocase attached )

Gaming and Visualization Technologies APIs DirectX, DXR, DirectCompute
1

Hi NVidia Team,
A couple of months ago we enhance our D3D11.3 based engine by using tiled resources. During testing we experienced strange GPU memory consumptions after resizing those buffers. When mapping tiles from regions beyond the tile pools initial size our gpu memeory footprint explodes.

Here is our repo case:

#include “TiledResourceLeakRepoCase.inc”

bool forceLeak = true;
//
// → Sysinternals ProcessExplorer: dedicated GPU MEM: 4.8 MB ( GeForce RTX 2080Ti )
// → Sysinternals ProcessExplorer: dedicated GPU MEM: 3.6 MB ( GeForce GTX Titan )
//
TiledResourceLeakRepoCase( device, context, forceLeak );
//
//-> Sysinternals ProcessExplorer: dedicated GPU MEM: 708.8 MB ( GeForce RTX 2080Ti )
//-> Sysinternals ProcessExplorer: dedicated GPU MEM: 387.6 MB ( GeForce GTX Titan )

TiledResourceLeakRepoCase.inc.txt (4.0 KB)

We test ist only on GTX Titan and 2080 Ti platforms with various drivers. In all situations the memory behaviour was the same, expect the magnitude of memory consumtion.

Any ideas what happens there?

Cheers,
René

PS: A relatet issue I’ve already posted here. Strange Behaviour of DX11 TiledResources at RTX driver ( RTX 2080 Ti )

2

Hi NVidia Team,

I want get back to you with the results of further investigation of my TiledResourceLeakRepoCase() in NSight Systems. The jump in memory Usage encounters after a MakeResident() event depicted at the screenshot below.

NSightSystemsScreen
NSightSystemsScreen1669×1056 122 KB

Maybe this helps if someone take a look on that.

Best,
René

3

Just test it on Quadro too and it seems to work fine on this platform.

// -> Sysinternals ProcessExplorer: dedicated GPU MEM: 4.8 MB        ( GeForce RTX 2080Ti )
// -> Sysinternals ProcessExplorer: dedicated GPU MEM: 3.6 MB        ( GeForce GTX Titan  )
// -> Sysinternals ProcessExplorer: dedicated GPU MEM: 1.5 MB        ( Quadro  RTX 3000 - Notebook )

TiledResourceLeakRepoCase( device, context, forceLeak );

// -> Sysinternals ProcessExplorer: dedicated GPU MEM: 708.8 MB      ( GeForce RTX 2080Ti )
// -> Sysinternals ProcessExplorer: dedicated GPU MEM: 387.6 MB      ( GeForce GTX Titan )
// -> Sysinternals ProcessExplorer: dedicated GPU MEM: 7.5 MB        ( Quadro  RTX 3000 - Notebook  )
4

Hi NVidia team,

Please find attached a complete compilable code project for my repro case.

TiledResourceReproCase.zip (12.3 KB)

Hope this helps.
René