Overclock.net › Forums › Industry News › Hardware News › [Phys.org] ARM chip makers set to reach 3GHz next year
New Posts  All Forums:Forum Nav:

[Phys.org] ARM chip makers set to reach 3GHz next year - Page 7

post #61 of 62
Quote:
Originally Posted by sdlvx View Post

More Intel Compiled benchmarks which don't represent real world performance but instead show exceptional situations which rarely happen in the real world.

Nice to know Intel Compiler gains performance by flat out skipping instructions. Very good!

I've learned from OCN that you're suppose to design hardware around software...
post #62 of 62
Quote:
Originally Posted by MrJava View Post

Anyone seen this yet?

http://www.theregister.co.uk/2013/07/12/intel_atom_didnt_beat_arm/

The "corrected" version that supposedly prevents Intel's cheating reduces their score by 20-50%. I say supposedly because while some of the problems are corrected, There's nothing to keep Intel's proprietary compiler from cheating again and it's still unknown if there are other shortcuts being taken (this is the same compiler that went out of its way to keep AMD processors from using SSE instructions so Intel's chips would appear faster).

 

It turns out that AnTuTu also doesn't auto-vectorize and doesn't use NEON. NEON gives between a 2x and 8x performance boost (according to ARM's developer whitepapers), so the final ARM scores should be much higher based on that alone.

 

Considering that the leaked Bay Trail benchmarks used the broken benchmark, their scores are now less than ARMS and if you consider the improvements that are available if optimizations were done, then real performance of Bay Trail looks even worse.

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [Phys.org] ARM chip makers set to reach 3GHz next year