At stock settings the quad-core A72 ARM chip at 1.5GHz is rather slow. It can easily handle an overclock to roughly 2.2GHz with minimal active cooling. But at stock settings, I would probably put the Pi-4 on par with a Core 2 Duo T8100 running at 2.1GHz.
In a more current context I’d guess that a single Ryzen 7 3700X would have the compute power of twelve Raspberry Pi-4B single-board computers. With that said, the most ambitious Raspberry Pi “supercomputer” project to date was built in 2019 by engineers from Oracle Switzerland originally slated for 1024 Pies, they had room to spare and thus added 36 more for a total of 1060.
I would estimate this conglomerate to have the computer power of four or five dual-EPYC (128-core) server blades—at least in raw compute power, but likely much slower in its actual I/O speeds due to bandwidth bottlenecks. The mass of spaghetti interconnects is reminiscent of the early CRAY supercomputers that routed its circuit boards using individual wires.
Given that my “expertise” involves actually building systems I probably should elaborate by saying that a 2GB Pi-4 is about $45 so a thousand of them would be $45,000. By comparison one 64-core EPYC-Rome is $7500, making ten of them (for a 5-blade server) $75,000 for the processors alone, not to mention 2TB of RAM and the motherboards.
It is easy to see the appeal of the SO-DIMM sized RaspberryPi 4 Compute Modules with their built-in 8GB eMMC flash for your own “supercomputer” projects without the tangle of Ethernet cables and power cables.
24.9K viewsView UpvotersView Sharers