3 ms·

And then your own custom allocator that would be fitted for your algorithms and vastly faster than malloc.