There are considerable price increases with each quite small increase in
speed-- hundreds of dollars--, but over two or three years I think the
extra dollars would be worth the performance increase... *IF* there is a
noticeable performance increase.
This would depend to a large degree upon the code... specifically, if
the code (OS and apps) makes use of the expanded instruction sets of the
more expensive CPUs. Generally the code doesn't, unless gcc/make is
configured for the particular CPU and then that source is compiled. I've
done this in the (distant) past and noticed a significant increase in
performance over the stock executables provided by the distro.
http://gcc.gnu.org/onlinedocs/gcc-4.6.4/gcc/i386-and-x86_002d64-Options.html#i386-and-x86_002d64-Options
says "mtune" can be set to one of these:
corei7
Intel Core i7 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3,
SSSE3, SSE4.1 and SSE4.2 instruction set support.
corei7-avx
Intel Core i7 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3,
SSSE3, SSE4.1, SSE4.2, AVX, AES and PCLMUL instruction set support.
core-avx-i
Intel Core CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, SSSE3,
SSE4.1, SSE4.2, AVX, AES, PCLMUL, FSGSBASE, RDRND and F16C instruction
set support.
Does 'cpuinfo' tell us about all of these when they're present, or are
we supposed to find out some other way?
These three options wouldn't seem to come close to specifying all the
various core i7 CPUs there are and optimizing for all the features of
each. Getting *some* of the additional instructions offered by i7s
would certainly help performance over what the standard distro offers,
but probably still not enough to warrant the extra expense of the
higher-end CPUs.
http://gcc.gnu.org/onlinedocs/gcc-4.6.4/gcc/i386-and-x86_002d64-Options.html#i386-and-x86_002d64-Options
also offers:
native
This selects the CPU to tune for at compilation time by determining
the processor type of the compiling machine. Using -mtune=native will
produce code optimized for the local machine under the constraints of
the selected instruction set. Using -march=native will enable all
instruction subsets supported by the local machine (hence the result
might not run on different machines).
but, again, does it make a distinction between the three CPUs cited at
top (and yet others)? If the code produced for all three CPUs is the
identical, then there isn't much point in spending for the costlier CPUs.
And does using "native" give better or worse results than specifying one
of the core* options?
Also, when compiling a kernel to run on a VM, should some other gcc
option(s) be used?
At this point I'd just be making wild guesses about how all this
actually works out. So does anyone have experience with, or maybe some
inside knowledge about, any of this?
If so, thanks for any light you can shed.