Elmer, I understand that base scores are prefered. However until the recent Intel submissions it was common to see 5-10% higher peak scores. Intel has been getting base and peak scores which are essentially identical, which has always suggested to me that they simply added optimizations within the commercial compiler that detected the test programs and implemented the peak tuning. Obviously that is a level of compiler optimization which is easily available to Intel and not to AMD (which is actually using the Intel compiler for these submissions). Clearly the ability to improve performance through compiler tuning has some real world value, so it is not reasonable to simply ignore the fact that AMD gets higher Peak scores, but I remain a bit unclear on how the spec results relate to real world performance.
Actually I think we had this discussion a few months ago, and agreed that SpecInt and FP were generally pretty curious choices for general measurement of Processors designed primarily for multi processor systems. It is mostly a matter of bragging rights with limited value in evaluating real world performance, which can be directly measured on the actual applications.
E
News 
Market Data 
Discover 
AI
