It is running on top of hashcat, and at the above-mentioned compute it was benchmarking 45 million bcrypts per second. At that point it is more about the attack plan than the compute.
"X bcrypts per second" is completely meaningless. What was the bcrypt setting? With the right setting, it would not be more than 1 bcrypt per century, or with the wrong setting, an almost equivalent rate to md5. It depends.
More meaningful would be the speedup compared to a single CPU core, which is what the developers (should) benchmark against. They should make it as slow as possible, so if their system can do bcrypt with a cost of 15 in 0.1 seconds, they should set either that or cost 16. (Much more than 0.2s might be annoying to users or be a DOS vector.)
Right was meant as necessary to achieve that effect (sorry, English is not my native language). Obviously this is not a recommendation but just to point out that its configuration ranges from negligible to (on today's computers) forever.
https://imgur.com/a/DXQMsM1
edit: here is the demo video: https://www.youtube.com/watch?v=KnD4f8N1_OE