Currently, there are two implementations of the AGP architecture, known as the "execute model" and the "DMA model." In the execute model implementation, the display device supports the same features for non-local video memory surfaces and local video memory surfaces. As a result, when you retrieve hardware capabilities by calling the IDirectDraw7::GetCaps method, the blit-related flags in the dwNLVBCaps, dwNLVBCaps2, dwNLVBCKeyCaps, dwNLVBFXCaps, and dwNLVBRops members of the DDCAPS structure will be identical to those for local video memory. Under the execute model, if local video memory is exhausted, DirectDraw will automatically fall back on non-local video memory unless the caller specifically requests otherwise.

In the DMA model implementation, support for blitting and texturing from non-local video memory surfaces is limited. When the display device uses the DMA model, the DDCAPS2_NONLOCALVIDMEMCAPS flag will be set in the dwCaps2 member when you retrieve device capabilities. In the DMA model, the blit-related flags included in the dwNLVBCaps, dwNLVBCaps2, dwNLVBCKeyCaps, dwNLVBFXCaps, and dwNLVBRops members of the DDCAPS structure describe the features that are supported; these features will often be a smaller subset of those supported for local video memory surfaces. Under the DMA model, when local video memory is exhausted, DirectDraw will automatically fall back on non-local video memory for texture surfaces only, unless the caller had explicitly requested local video memory. Texture surfaces are the only types of surfaces that will be treated this way; all other types of surfaces cannot be created in non-local video memory unless the caller explicitly requests it.

DMA model implementations vary in support for texturing from non-local video memory surfaces. If the driver supports texturing from non-local video memory surfaces, the D3DDEVCAPS_TEXTURENONLOCALVIDMEM flag will be set when you retrieve the 3-D device's capabilities by calling the IDirect3DDevice7::GetCaps method.

Currently, there are two implementations of the AGP architecture, known as the "execute model" and the "DMA model." In the execute model implementation, the display device supports the same features for non-local video memory surfaces and local video memory surfaces. As a result, when you retrieve hardware capabilities by calling the DirectDraw7.GetCaps method, the blit-related flags in the lNLVBCaps, lNLVBCaps2, lNLVBCKeyCaps, lNLVBFXCaps, and lNLVBRops members of the DDCAPS type will be identical to those for local video memory. Under the execute model, if local video memory is exhausted, DirectDraw will automatically fall back on non-local video memory unless the caller specifically requests otherwise.

In the DMA model implementation, support for blitting and texturing from non-local video memory surfaces is limited. When the display device uses the DMA model, the DDCAPS2_NONLOCALVIDMEMCAPS flag will be set in the lCaps2 member when you retrieve device capabilities. In the DMA model, the blit-related flags included in the lNLVBCaps, lNLVBCaps2, lNLVBCKeyCaps, lNLVBFXCaps, and lNLVBRops members of the DDCAPS type describe the features that are supported; these features will often be a smaller subset of those supported for local video memory surfaces. Under the DMA model, when local video memory is exhausted, DirectDraw will automatically fall back on non-local video memory for texture surfaces only, unless the caller had explicitly requested local video memory. Texture surfaces are the only types of surfaces that will be treated this way; all other types of surfaces cannot be created in non-local video memory unless the caller explicitly requests it.

DMA model implementations vary in support for texturing from non-local video memory surfaces. If the driver supports texturing from non-local video memory surfaces, the D3DDEVCAPS_TEXTURENONLOCALVIDMEM flag will be set when you retrieve the 3-D device's capabilities by calling the Direct3DDevice7.GetCaps method.