% | $
Quotes you view appear here for quick access.

Intel Corporation Message Board

  • will_amd_yu will_amd_yu Jul 13, 2013 1:07 AM Flag

    Why Jim Mcgrag and eetimes could be wrong about Antutu benchmarks.

    RISC instructions vs CISC. The Intel compiler probably optimizes for CISC, by combining multiple RISC instructions into their CISC equivalent. That could be why many of the instructions excecuted by ARM benchmarks are not executed by Atom. The point is not how many instructions get executed, but is the task done.

    Until we get Intel's side of the story, my bet is the Intel Atom and baytrail kick butt, and eetimes can go sniff Mr Weaver!!

    SortNewest  |  Oldest  |  Most Replied Expand all replies
    • It is way early to be putting too much stock into the benchmarks but I think what most of us have been saying is that on paper at least, Baytrail should easily outperform an equivalent ARM chip. I don't think you are ever going to get any ARM biased writers to ever admit that ARM will under perform an INTEL chip. The common mantra of these people is that ARM chips are inherently more efficient then an equivalent INTEL chip. They are wrong of course, but they will continue to cling to that view regardless dismissing any evidence to the contrary.

    • The problem is that this is about perception and not reality. If the perception is that Intel somehow 'cheated' it's in for some stock price hurt.

    • It's got nothing to do with RISC v CISC.

      Intel replaced a GCC generated binary with an ICC one
      The ICC binary used a different set of compile switches than GCC (which turned on the SIMD instructions for x86)
      The ICC binary 'skipped' parts of the benchmark

      Doing any one of the above invalidates the benchmark score.

      • 2 Replies to theblueredmonk
      • "Intel replaced a GCC generated binary with an ICC one"
        Do you know if Intel did the recompile?

        "The ICC binary used a different set of compile switches than GCC (which turned on the SIMD instructions for x86)"
        Do you know what switches were used? I suspect that a "target Bay Trail" switch was added so the compiler would generate out-of-order Atom code. My bet is that WHOEVER used the compiler on the AnTuTu benchmark source code, used the standard gcc switches.

        I cannot find where AnTuTu has even released their source code, either public domain or for sale. Do you have information about getting it?

      • The compile switches are turning on optimizations. Skipping parts of the benchmarks is your interpretation of what happened. Mcgrag clearly said skipped "instructions".

35.26-0.01(-0.03%)Oct 24 4:00 PMEDT