"Everyone" uses GPUs to run serious models anyway, so ML tasks are not a problem. Still not a great deal.