I think you answered your own question. It seems you’re trying to load a texture too large for the graphics card’s memory. This means that you either have to add memory modules to it or buy another one with more memory.
I’m curious: why do you need such large texture? To display an image that large you have to resort to some large data management technique such as level-of-detail or multiresolution volume. You never actually need (or are able to) to load all your data in full resolution. I mean, if you’re looking at the entire cube from afar, you only need a low resolution version of it. If you zoom in, then you only need the small portions of the image that happen to intersect the viewing frustrum. That’s how we display 300GB-1TB seismic surveys cubes in petroleum industry without the need for a high-end graphics card.
A simpler techinque for displaying large volumes is by using orthoslices or camera-facing slices. This way you reduce the problem to a number of 2D textures, reducing memory demand dramacticaly: