The z386 open-source FPGA reimplementation of the Intel 80386 has received significant performance improvements. The key addition is early start memory access, a feature native to the original 386 that begins memory allocation for the next instruction one cycle early. Implementing this in the FPGA required getting the computation fast enough to fit within a single cycle, yielding a ~9% gain. Further optimizations include store queue rewrites and computing branch memory jumps several cycles earlier than the original 386 design, reducing branch cycle time from 9.25 to 6. Microcode decode frontend improvements rounded out the changes. The combined result is approximately a 39% performance increase on the DOOM benchmark. The z386 remains incomplete — it still underperforms the original 386 and cannot boot Windows.
