- Joined
- Jul 31, 2004
Was reading various stuff regarding compilers and libraries (specifically Intel ICC producing code that disables optimizations if it detects a non-Intel CPU). Found some links to an alternative math library, and am curious what libraries and compilers most software actually use.
http://www.yeppp.info/home/yeppp-performance-numbers
The benchmark is single-threaded with all Turbo features disabled.
Cycles per function:
|log|exp
i7 4770k|2.7|4.3
FX-6300|9.2|5.8
Granted, that's a very limited data set, but if similar numbers follow for other operations (and remember that the FX-6300 is only a third the price of the 4770k, so the numbers being somewhat worse isn't a horrible thing), does the "AMD has crappy single-threaded performance" really hold up? ICC has been proven better than other compilers in several cases, but it is known to be biased. How much common software is actually compiled with ICC or using less-than-optimal math libraries that may make the AMD CPUs look worse than they really are? Everybody knows Intel has quite a bit more money and therefore can afford better design teams and fabs, so I'd rather attempt to learn more about compiler/library usage among various products that are often referred to when talking about performance. Might be hard to find that info for a lot of closed-source software
http://www.yeppp.info/home/yeppp-performance-numbers
The benchmark is single-threaded with all Turbo features disabled.
Cycles per function:
i7 4770k|2.7|4.3
FX-6300|9.2|5.8
Granted, that's a very limited data set, but if similar numbers follow for other operations (and remember that the FX-6300 is only a third the price of the 4770k, so the numbers being somewhat worse isn't a horrible thing), does the "AMD has crappy single-threaded performance" really hold up? ICC has been proven better than other compilers in several cases, but it is known to be biased. How much common software is actually compiled with ICC or using less-than-optimal math libraries that may make the AMD CPUs look worse than they really are? Everybody knows Intel has quite a bit more money and therefore can afford better design teams and fabs, so I'd rather attempt to learn more about compiler/library usage among various products that are often referred to when talking about performance. Might be hard to find that info for a lot of closed-source software