I go into detailed discussion with tromp in my thread:
https://bitcointalk.org/index.php?topic=557732.0Also you can find some discussion between him and myself in the MemoryCoin 2.0 PoW thread which I linked to upthread.
In short, I don't believe it is CPU only currently, but (and I think we agreed this, but ask him) it might be the appropriate algorithm for mobile later if CPUs move to extremely high number of cores. Our discussion concluded with more testing is needed and I would try to help him get a TileGX in future if I can.
I'm much more interested in your thoughts on HoneyPenny's PoW, as it's claimed to be an improved and future-proofed version of the CryptoNote algo: