This is also do-able with instancing. You store the base verts for a single quad as per-vertex data and the info that is unique to each particle as per-instance data (one instance per-particle), then draw. It doesn't have the advantage of a GS of only needing to do the calculations once per-particle (it's once per-vertex instead) but also doesn't have the overhead of having the GS stage active, so it may run faster on some hardware (I can easily enough simulate 1,000,000 particles at semi-playable framerates on the GPU using a similar setup). Instancing is also available on a wider range of hardware, but - again - YMMV where Intel drivers are concerned.