Re:Intel’s 8-Core Mobile Tiger Lake-H, at 45 W, to Ship in Q1
by giantrobot on Tuesday January 12, 2021 @02:09PM
(#25751817)
I don't find fault in Intel using insider knowledge to make their compilers work better on their own CPUs. If they make optimizations they know work on their CPUs but can't guarantee AMD CPUs will work exactly the same, I also don't find fault in them only supporting their CPUs with those optimizations.
Where I find fault is bullshit marketing saying their CPUs are x% faster than AMD and then point to code compiled with their compiler. I'd be fine if they wanted to say ICC produced faster results than GCC or LLVM. There's no problem selling your CPU with your compiler to get the best performance possible.
Instead they're saying our chips are faster than AMD. Full stop. Once you wade into the bullshit you see they're using ICC with FMV enabled. This means even if you know what compiler flags they used you still can't be sure what code is running on each CPU. It's bullshit marketing and I think it needs to be called out whenever it happens.
I would argue they actively make sure their compiler optimisation works only on Intel CPU. Which is actually a fair deal. Those optimisation work isn't free.
It is BS tactics? Yes. Do I Like it ? No. It is wrong to do it? That is up to debate.
The point is using an Intel Compiler on Non Intel CPU comparison makes the benchmarks pretty much useless. Something like GCC / LLVM should have been used to be vendor neutral.