Speaking of zettascale, Intel ignited the debate back in2021. While the HPC community initially smirked at their plans, Intel continued pushing the zettascale agenda, with claims to achieve 1 zettaflop/s by the end of the decade. Was it still going?
I think a more realistic timeline for zetta is zettaop/s in 2032 at 50 MW, zettaflop/s in 2037 at 200 MW, and zettascale by 2038. Can Intel or anybody else pull it off before then? I guess only time will tell.
Also, with Intel claiming to manage 2x performance improvements which would yield zettaflop/s by 2032 (prediction), this 1,000x in performance comes at the cost of 3–5x in power (power requirement of the entire system of 50–100 MW.).
And the energy efficiency to perform fp64 operations needs to be increased by 200–350x, from ≈50 to over 10.000 Gflop/s Watt .