| |
Intel's 80386 processor used an "Early Start" technique to hide memory latency by beginning the next instruction's address calculation during the final cycle of the current instruction, contributing approximately 9% to overall performance. The z386 FPGA core implementation of the 386 microcode added this Early Start optimization along with other improvements, achieving ao486-class performance with Doom frame rates jumping 39% (from 16.6 to 23.0 FPS) at the same 85 MHz clock speed by reducing cycles per instruction. The optimization works by leveraging a forwarding network to detect register dependencies and overlap instruction execution stages, though Intel's original implementation had a corner-case bug in this forwarding logic that caused the famous POPAD bug.
Read Full Article →
← More Tech news