Yes sure!
A texture of the size of your screen and a buffer are created. The buffer is large enough to store all of the pixels that'll be sent to the texture.
The pixels are stored as RGBA data in that buffer. All of these pixels are set according to Mandelbrot formula (which can tell you wether a point belongs to the fractal), and as pixels don't depend on each other, I can parallelize that work. Once everything is rendered, pixels are sent to the texture and displayed. So actually, the only part of the fractal that is rendered is the one you see on the screen. That's also why rendering must be redone each time you want to move in the fractal.
I can't render the whole fractal except when it's not scaled, because it would take way to much time and memory to compute and store millions of billions of pixels.