Work out flops of a processor instantly with clear inputs, formula shown and shareable results.
Peak FLOPS is clock times cores times FLOPs per cycle, where the last term comes from vector width and fused multiply-add: a 512-bit unit doing FMA on doubles gives 16 FLOPs, doubled to 32 with two units. Achieved efficiency of 60-80% on dense linear algebra is good; sparse workloads reach far less.
Peak FLOPS
peak = clock x cores x FLOPs per cycle; FMA counts as 2 FLOPs
Vector lanes times FMA factor times pipelines. AVX-512 with two FMA units gives 8 doubles x 2 x 2 = 32 FLOPs per cycle.
Memory bandwidth. Most codes are bandwidth bound, not compute bound, so arithmetic intensity — FLOPs per byte moved — determines the achievable fraction.