It isn't a matter of "no matter how irrelevant they think it is"...it is FACT that it is irrelevent.
The 1.4ghz Athlon is living proof. It slams the 1.8ghz P4 in business benchmarks and although loses in 3d benchmarks, it doesn't lose by much. A chip that is 400mhz slower. How relevent are clock cycles here?
The performance of a chip does not rely on clock cycles alone anymore than horsepower rating of an engine tells you how fast a car can go. The point here is that many other parts and subsystem components must be carefully matched, or have the capabilities to work with the the other. Mismatch parts and the performance goes down the drain.
Many might disagree with me, but I think the key to system performance, or CPU performance isn't just clock cycles or even how many instructions per second it can perform (certainly a combination of the two would be a good indicator), however what is a couple of the main factors cited in many CPU tests that attributes to lower performance? Latency and narrow bandwidth. Why do you think Intel put a 20 segment trace cache in the P4? To cut down on latency. Granted, even though they've upped the reliability of the prediction unit from around 90% to 94%, that huge cache takes a lot of clock cycles to flush when a branch prediction goes bad. When it's on though, things fly because of the far lower latency. This is just one example of how other CPU sections can be used to optimize performance without even looking at clock cycles.
DDR memory also is a fine example of such performance increase without concern for clock cycles. Two reads/writes per clock cycle instead of one...double the data in the same amount of time. The memory still runs at the same clock speed as the SDR counterpart.
The fact is, clock cycles ARE irrelevent as a single measure of performance, but John Q. Public has been erroneously taught different. I can certainly understand AMD's concern for this and their desire to get away from this stigma.
p.s. AMD at one time did give their chips a "PR" nomenclature. PR stood for "Pentium Rating" and was followed by the number corresponding to the clock speed of the Intel chip it was equivalent to.
Slashdot Mirror
It isn't a matter of "no matter how irrelevant they think it is"...it is FACT that it is irrelevent. The 1.4ghz Athlon is living proof. It slams the 1.8ghz P4 in business benchmarks and although loses in 3d benchmarks, it doesn't lose by much. A chip that is 400mhz slower. How relevent are clock cycles here? The performance of a chip does not rely on clock cycles alone anymore than horsepower rating of an engine tells you how fast a car can go. The point here is that many other parts and subsystem components must be carefully matched, or have the capabilities to work with the the other. Mismatch parts and the performance goes down the drain. Many might disagree with me, but I think the key to system performance, or CPU performance isn't just clock cycles or even how many instructions per second it can perform (certainly a combination of the two would be a good indicator), however what is a couple of the main factors cited in many CPU tests that attributes to lower performance? Latency and narrow bandwidth. Why do you think Intel put a 20 segment trace cache in the P4? To cut down on latency. Granted, even though they've upped the reliability of the prediction unit from around 90% to 94%, that huge cache takes a lot of clock cycles to flush when a branch prediction goes bad. When it's on though, things fly because of the far lower latency. This is just one example of how other CPU sections can be used to optimize performance without even looking at clock cycles. DDR memory also is a fine example of such performance increase without concern for clock cycles. Two reads/writes per clock cycle instead of one...double the data in the same amount of time. The memory still runs at the same clock speed as the SDR counterpart. The fact is, clock cycles ARE irrelevent as a single measure of performance, but John Q. Public has been erroneously taught different. I can certainly understand AMD's concern for this and their desire to get away from this stigma. p.s. AMD at one time did give their chips a "PR" nomenclature. PR stood for "Pentium Rating" and was followed by the number corresponding to the clock speed of the Intel chip it was equivalent to.