I actually spent some time at Control Data Institute in a Washington DC suburb after I got out of the Army in 1971, working on and around some of that big iron from the likes of Sperry (Utah / Rand) that were all about the size of my three door refrigerator or larger.
Input was done via 80 column keypunch cards and data was stored on reels of magnetic tape. Not sure how that ranks by way of throughput or I/Os, but......
Conceptually that would be like comparing a hand crank to start early model A or T ford with a Tesla