A hardware researcher named Christopher Domas has initiated a project to identify the slowest single x86 instruction, creating a leaderboard to rank these performance bottlenecks. The current top offender is the `fxrstor64` instruction, which took an astonishing 198 billion cycles, or 62 seconds, to execute by restoring a 512-byte SIMD register state from memory. Domas achieved this by manipulating memory-mapped I/O and overwhelming the CPU's PCIe fabric to deliberately slow down the instruction's execution. Future leaderboards are planned for ARM and RISC-V architectures, with specific rules ensuring only single instruction latency is measured. AI
IMPACT Provides insights into CPU architecture and instruction-level performance, potentially informing future hardware and software optimizations.
RANK_REASON This is a research project focused on understanding CPU instruction performance by deliberately slowing them down. [lever_c_demoted from research: ic=1 ai=0.4]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
