Quote Originally Posted by zalbard View Post
Hi, and thanks.
The problem about this is, these GFlops are not really relevant to real-life performance. For example, currently non-HT CPUs produce more GFlops than those with HT enabled. Yet if you try some rendering software, HT-enabled CPUs will be way ahead.
Another issue is memory performance. Linpack seems to be quite sensitive to that. It doesn't really translate into real-world advantage, though. For example, a 4GHz CPU with 2GHz RAM may be faster in LinX than a 4.5GHz CPU with 1GHz RAM, but it will be the other way around in pretty much all of the applications.
Not to mention that produced GFlops are often inconsistent.
I think it makes sense the way it is now. LinX is used to measure stability, not performance.
However, it's up to the members to decide. If I see many more people supporting your motion, I can rearrange everything.
I am going to have to pass as using gflops as a metric and as you were saying the numbers are inconsistent, and what really does matter is real world applications, I feel the current methodology is inline