3 ms·
Even when you get past all the indirection and interpreted languages people use there is STILL usually 12x - 100x the speed left on the table. Even in a native
by CyberDildonics 11y ago
Even when you get past all the indirection and interpreted languages people use there is STILL usually 12x - 100x the speed left on the table.
Even in a native program heap allocations can slow something down to 1/7th.
After that memory ordering for cache locality can still gain 10x - 25x speedups.
After that proper SIMD use (if dealing with bulk numeric computations) can buy another 7x (that's the most I've gotten out of AVX and ISPC.
Then proper parallelism and concurrency are still on the table (but you better believe that the concurrency can be very difficult to make scale).
The divide between how fast software can potentially run and how fast most software actually runs is mind blowing.