Title says 2 minutes, but it's actually 2:37, which is 30% more. Or actually 3 minutes if you don't disable optimizations, which is 50% more. Those 2x80 cores machines are also quite expensive. For reference, my Threadripper 3970X (32 cores, 64 threads, so much less cores) can build clang with the same command line (without optimization) in less than 2 minutes (1 minute and 42 seconds), or with -DLLVM_TARGETS_TO_BUILD="X86" in 2 minutes and 23 seconds, and that's much cheaper.
I used to run american fuzzy lop on consumer dual-channel motherboards, and got about a 6x speedup from a 16-core Ryzen setup.<p>Are these cloud instances running on 40-channel ram motherboards ? Otherwise isn't it wasteful to have 80 cpu's on a single socket ?