http://www.amdzone.com/phpbb3/viewtopic.php?f=52&t=136697&start=50
by
abinstein » Wed Sep 09, 2009 2:01 am
MU_Engineer wrote:
Lem wrote:Core i5-750 and i7-870 reviews are starting to show up.
Anand has a super glowing Intel love-fest review, which was expected. You can find it
here.
Phoronix has also tested the i5 750 and the i7 870,
here. Unfortunately they used a Phenom II X3 720 to compare with. Amusingly, there are benchmarks where the X3 beats out the Nehalems..
It's very interesting to see the huge difference in results between the Phoronix and the Anandtech tests. I've never really seen a good head-to-head of modern AMD and Intel CPUs under Linux, and here Michael Larabel provides a very nice one, although they should have put in a Phenom II X4 965BE as a price-parity competitor to the i7 920. A bunch of people in the Phoronix forums called Michael out on this, but he used the X3 710 as that's all he had lying around.
The big things that I saw that was teased out of the Anandtech and Phoronix reviews were:
1. The i5 line leans heavily on Turbo Boost for performance. Disabling Turbo Boost cripples i5 performance mightily (the unboosted i5 750 almost always loses to the lower-clocked, half-the-price, Phenom II X3 710 with one fewer core.)
2. HyperThreading does give a performance benefit to the Nehalem line as the HT-less i5 750 absolutely stank compared to the i7 870, which apart from a slightly higher clock speed was completely identical.
3. The triple-channel IMC on the i7 9xx series does help performance as the 2.67 GHz i7 920 always won out over the 2.93 GHz i7 870.
4. There is a massive differential in the performance of typical Windows program binaries and Linux program binaries with respect to AMD vs. Intel CPUs. AMD CPUs do far better in comparison to the Nehalems using GCC on Linux versus MSVCC/ICC/whatever on Windows. You could say this is due to a ton of different things such as Windows programs being optimized much more heavily for Intel CPUs, GCC being naturally tuned much better for AMD CPUs, MSVCC/ICC/whatever poorly supporting AMD CPUs, or GCC poorly supporting Intel CPUs. The one thing you
can't say is that the binaries under Linux are optimized at compile time to favor any one CPU as we
know what compiler and compiler options are used to compile them. Not so with most Windows binaries.
5. Intel is going to have a hell of a time trying to sell any i5 CPUs to Linux users unless they mark down the price a ton or get Turbo Boost to work and give a massive performance boost to the CPUs. This Phoronix article already has gotten a lot of word of mouth in the Linux community, and Linux users tend to be a bunch of cheap bastages. Just about none of them are going to buy a $196 i5 750 just to see it get walloped by a Phenom II X3 710 that costs half as much.
MU_Engineer, I like your post, rational & right on the point.
IMHO, the disparate benchmark results from different applications (or optimization) is expected as these processors have more advanced features to improve performance. It is actually pretty easy for a compiler to generate binaries favoring one microarch over the other. (There are sequences of instructions which would benefit Intel's arch but penalize AMD's. There are also sequences that are indifferent on Intel's but penalize AMD even more.) This benchmarketing game is something that AMD can hardly win, due to Intel's much larger size and deeper involvements (in monetary, software IP and labor).