3 ms·
There are so many execution ports compared to the throughput of decoding, renaming and retiring, that they are rarely the bottleneck. As ant6n correctly pointed
by gpderetta 10y ago
There are so many execution ports compared to the throughput of decoding, renaming and retiring, that they are rarely the bottleneck. As ant6n correctly pointed, memory latency, branch prediction, dependency chain length, decoding and renaming are, in that order, normally the bottleneck even for hand optimized assembler code.