By: Patrick Chase (patrickjchase.delete@this.gmail.com), July 9, 2015 1:08 pm
Room: Moderated Discussions
Tim McCaffrey (timcaffrey.delete@this.aol.com) on July 9, 2015 10:41 am wrote:
> The best way to compare these results is to lock down the clock speed, it would
> also help if you could compare systems that were identical in all aspects
> (cache size, etc) but only varied in clock speed. You could then tell if
> scaling the scores by clock speed is valid or not.
Even if you did that it would *still* be meaningless, because you'd be comparing one uarch running at its optimum/target clock speed to another that's been radically underclocked and that is therefore burdened with either a needlessly long pipeline or over-aggressive circuit design. The most you can possibly say is something along the lines of "Haswell performs about the same as Cyclone, but only if you force it to spend >60% of the time doing absolutely nothing".
Microarchitectures must be compared at the operating regimes for which they are optimized. In many cases those operating points are so far apart that useful comparisons are impossible, and IMO Cyclone-v-Haswell is one of those.
> The best way to compare these results is to lock down the clock speed, it would
> also help if you could compare systems that were identical in all aspects
> (cache size, etc) but only varied in clock speed. You could then tell if
> scaling the scores by clock speed is valid or not.
Even if you did that it would *still* be meaningless, because you'd be comparing one uarch running at its optimum/target clock speed to another that's been radically underclocked and that is therefore burdened with either a needlessly long pipeline or over-aggressive circuit design. The most you can possibly say is something along the lines of "Haswell performs about the same as Cyclone, but only if you force it to spend >60% of the time doing absolutely nothing".
Microarchitectures must be compared at the operating regimes for which they are optimized. In many cases those operating points are so far apart that useful comparisons are impossible, and IMO Cyclone-v-Haswell is one of those.