Intel caught cheating?

[Xentropy]

http://arstechnica.com/reviews/hardware/atom-nano-review.ars/6

"My my. Swap CentaurHauls for AuthenticAMD, and Nano's performance magically jumps about 10 percent. Swap for GenuineIntel, and memory performance goes up no less than 47.4 percent. This is not a test error or random occurrence; I benchmarked each CPUID multiple times across multiple reboots on completely clean Windows XP installations. The gains themselves are not confined to a small group of tests within the memory subsystem evaluation, but stretch across the entire series of read/write tests. Only the memory latency results remain unchanged between the two CPUIDs."

 

[Davros]

futuremark have worried me for a while

 

[ShaidarHaran]

If this carried over to real-world apps it may be relevant, but shenanigans in a synthetic test don't exactly surprise (or even concern) me anymore, outside of the GPU world anyway.

 

[Albuquerque]

The title of this thread is misleading -- this would be erroneous behavior on the part of Futuremark, not Intel. Unless you can somehow link Intel to these scores directly, which I don't see in that article...

 

[Xentropy]

Intel wrote the compiler that was used for PCMark 2005, so it could be either Futuremarks or Intels doing. Thats why the heading ends in a ? not a period.

 

[Albuquerque]

Intel wrote the compiler that was used for PCMark 2005, so it could be either Futuremarks or Intels doing. Thats why the heading ends in a ? not a period.

More than one company uses Intel's compilers, probably nearly every game you play on your computer used an Intel compiler for build. Does every game on your computer have the same issue? There's an easy way for them to find out it seems.

It's highly unlikely that a compiler optimization is the suspect behind this.

 

[ShaidarHaran]

More than one company uses Intel's compilers,

True.

probably nearly every game you play on your computer used an Intel compiler for build.

Most assuredly not true. MS VC++ is still the most widely-used compiler in the business.

 

[Albuquerque]

Oops, yeah probably true on the MSVC front.

Wouldn't be difficult to parse out "compiler enhancements" from "application enhancements" though.

 

[swaaye]

I like this argument. It comes up rather frequently and everyone goes bananas.

My question is: why is it cheating if Intel programs their compiler to favor their hardware? Seems logical to me! If they want to gimp the hell out of the competition in their product that they've spend millions of man hours and serious money on, why is this surprising or even evil!? The software companies are at fault for using a knowingly biased compiler. It should be expected that this compiler will do something like this.

AMD and other companies should just make damn sure that MSVC works awesome for their CPUs and push that. Or, like Intel, make their own biased compilers.

 

[Thorburn]

I don't know for sure, but I'd think the Intel compiler team probably just optimises for the Intel microarchitectures.
The fact that they still outpace the MSVC compilers even running on competitors hardware means that if they were 'gimp'ing in it then they aren't doing a particularly good job.

Cut out the conspiracy theories and think a little.

As for PCMark, dunno the root cause but my gut instinct would be to suggest they were choosing the SSE level to use based on the instruction sets available at the time for the vendor, probably more a poor design choice than a malicious 'cheat'.