Last week we have released GPUBox Web Service that allows to set up a private multi-GPU cloud on Amazon EC2. Blender is of course supported and it works like a charm (which was confirmed both in our internal tests and by users who already launched their GPU-enabled cloud infrastructures).
Blender is installed by default on the virtual machine image that is used within the service. Currently it is version 2.71, but you can download, install, and use any other version.
What might be not so obvious to the new users, is that GPUBox Web Service is not another render farm. It lets you edit on your scene and render it in real time and you do not need anything else than a web browser. You can also adjust the size of the infrastructure according to your current needs - even if you want to use 100 or more GPUs.
At the moment the GPUs that can be used are GRID K520 (performance comparable to GTX 670) and Tesla M2050 (not available in every region).
Using GPUBox technology on Amazon AWS requires GPUHours that can be purchased on our website - http://renegatt.com/webservice.php
There are three packages available: 40, 320 and 2000 GPUHours.
GPUHours does not cover using instances on Amazon. We left this area to user because otherwise it would be impossible to use cheap “Spot” GPU instances that can be obtained for ~0.05€ per instance. So total cost of using GPUBox Web Service - in case of using “Standard” Package (0.30€ for GPUHour) and “Spot” instances on Amazon - is ~0.35€/GPU. It means that - for example - you can use 8 GPUs simultaneously for less than 2.90€ per hour!
If you need any help regarding GPUBox Web Service you can: