pull down to refresh

fwiw Someone in the lab that does local AI stuff recommends fewer more powerful GPUs because communication ends up being a bottleneck.