Eric, <..and maybe even learn something.> If you want to learn,
ask a question, politely, instead of making short-sighted
statements. Now try to comprehend one more time:
<A synthetic benchmark shows the potential of a processor for running applications that are similar to the benchmark.>
Narrow thinking, student. Try to comprehend one more time:
A real-world application consists of variety of
code patterns that are executed mostly sequentially.
A real-world benchmark measures the time to execute
the sum of all these patterns.
A synthetic benchmark usually synthesizes one of
those code structures, or kernels, or "basic blocks":
- control flows, block copy, matrix multiply,
system calls, process forks, etc.
If one processor executes most of the basic kernels
better than another, it has better chance to better execute
an average real-world code. That is why your statement
"Scoring high on a synthetic benchmark DOES NOT correlate
to faster performance running real applications" is
highly inaccurate although formally correct since you
said "a synthetic benchmark", not "benchmarks". FYI,
the Norton SI benchmark contains not only a single
synthetic pattern but a variety of them, and it
DOES CORRELATE with certain class of real applications.
Scoring good on synthetic benchmarks but poorly
on real applications usually indicates a flaw in
system design, not necessary in CPU itself. Example:
those lame Triton chipsets from Intel, with
look-aside memory-blocked L2 cache.
Have a friendly conversation.
|
