The graphic card is the most important piece of hardware for using VideoStitch. As its stitching engine leverages the CUDA framework, it requires a CUDA compatible graphic card to run.
CUDA is a technology developed by Nvidia available on most of the GPUs it produces. Nvidia GPU chips are found on various manufacturer's graphic cards. There are three main information to look for when considering a CUDA compatible graphic card for VideoStitch to use: the compute capability, the number of CUDA cores, and the amount of graphic memory offered by the card.
VideoStitch Studio requires 2.0 or better. This number describes the CUDA hardware capabilities. A list of graphic cards and their compute capability can be found on Nvidia's website.
The more, the fastest.
You will be able to stitch as VideoStitch Studio leverages these CUDA cores for processing. The computing power gain grows linearly with the number of CUDA cores. This value can vary greatly from below 100 cores up to 2000 cores and more. We recommend to target cards with over 300 cores. Fewer cores will work as well, but provide for a rather slow processing experience. Cards with over 1000 cores provide faster processing resulting in an even more responsive GUI and stitching experience. High end cards such as the 'GTX Titan' provide over 2000 cores and offer a real-time stitching experience.
The more, the higher resolution you will be able to stitch.
The maximum output size you will be able to stitch depends on various factors such as the number of input videos and how much these inputs overlap, within the limits of the available graphic memory on the graphic card. We recommend at least 3GB of graphic memory. We found that 1GB works fine for stitching 6x Hero2, 2GB fits a Hero3 setup better, while stitching higher resolution images (4K, full resolution time-lapse) or more videos requires 4GB or more for a full resolution workflow - 10000x5000 pixels (or more) equirectangular videos!
Please note that some high end graphics cards come with multiple GPUs.
These cards behave just like a multiple GPU setup : the amount of memory that VideoStitch can use is the ‘per GPU’ graphics memory