7 hours is certainly long for a volume of that size. Are you setting the step size and accuracy parameters? You can spend a lot of time trying to over-optimise the highest resolution (i.e. smallest step size) at the end of the registration.

See:

* http://flybrain.mrc-lmb.cam.ac.uk/dokuwi...
* http://flybrain.mrc-lmb.cam.ac.uk/dokuwi...

for some discussion. Setting '--accuracy ' to e.g. half of your voxel size may save you a lot of time (e.g. '--accuracy 0.5' if you have 1�m voxels) and you can experiment with larger values. Also if your image is oversample compared with the biological variability of the specimen you can decide what the finest resolution image data you want to use is using the and then add the '--omit-original-data' argument.

Also are you using parallel threads? warp will do quite well with 4-8 threads on most hardware.

Best,

Greg.