Leela local vs Leela AWS

[Ophitoxaemia]

I recently installed Leela 0.10 on a 2 month old Surface laptop natively booted into Ubuntu 17.04

I also have an AWS account, where one can rent basically any size or type of computer by the hour.

Has anyone figured out if Leela can make use of a mega machine in the cloud? I wouldn't mind if it ran a little faster.

[apetresc]

P2-class instances would be the optimal choice. I've never tried to run Leela on one, but I run other OpenCL workloads on them all the time and it's never behaved differently from a local GPU, so there's no reason it wouldn't work.

As far as I know, Leela isn't able to take advantage of multiple GPUs, so your results won't be great. Individual P2 GPUs are K80 and not particularly powerful (take a look at these benchmarks). AWS workloads are meant to take advantage of parallelism, which Leela doesn't.

That said, it's a good way to create an even playing field for comparing different engines or engine versions against each other, without worrying about hardware inconsistencies.

[Ophitoxaemia]

Thank you for such an informative reply!