It's been a while since I looked, but I recall GPU prices being around $2 and hour (and thus writing them off).

$2 per hour are cluster compute instances with 2 NVIDIA Tesla M2050 GPUs (“Fermi” GF100) GPUs. g2 instances are smaller with NVIDIA GRID GPUs (“Kepler” GK104).

- G2 popular use cases: Game streaming, 3D application streaming, and other server-side graphics workloads.

- CG1 popular use cases: Computational chemistry, rendering, financial modeling, and engineering design.