Doesen't it smell you of BS when a cpu beats a contemporary one with a 100-200% edge?! Doesen't smell of BS when a 3G P4 gets a 4000 rating while a 3000+ only takes 2800. I work everyday with both of them and can say that they're just head to head.
C'mon, are we STUPID not to see that it's just a hard coded line of programming to determine that, not analyzing, it's just something like
No, do you think the same way if there was two systems that are otherwise identical except one used dual channel memory and other used single channel memory and showed 80% advantage in memory bandwidth?? Do you call that bullshit and conclude dual channel only gives 5% bandwidth increase over single channel?? No, because we are looking at one factor.
by m25
Everything clear from the table. Bandwidth and software performance perfectly make sense.
However, I'd let out those STUPID, 100% synthetic, PRO-Intel values that give intel CPUs (even old P4s) lead of 200-300% over K8s.Sandra2007 gives a performance rating of ~4000 for a 3GHz P4 and only 2800 for my 3000+, not to mention the likes of PC/3D mark. That's just shameful
Err, early benchmark showed some synthetics showing clear advantage of A64 over Core 2 Duo(like memory bandwidth benchmark etc), and people outrightly dismissed C2D because of that. However, it showed in actual benchmarks C2D kicks the crap out of A64 and more.
Actually, the reason Pentium 4's got such advantage in some benchmarks over A64(like SSE2) and doesn't in real world apps is because
synthetics measure only one component.
Say a particular real world program ONLY consisted of SSE2, memory dependency, and legacy x87 FPU code.
Say it consisted of:
30% SSE2
50% memory
20% x87 FPU
Say A64's are:
2x FPU performance
1.8x memory performance
0.5x SSE2 performance
Of course in synthetic apps its gonna show P4 killing A64 in SSE2 and A64 killing P4 in memory/FPU performance.
The particular app would show 45% advantage in performance for A64 over Pentium 4.
If another benchmark was:
60% SSE2
15% memory
25% FPU
Pentium 4 would be 11% faster than A64 in this case due to lots of SSE2.
Of course this is vastly simplified comparison, but you see, if you are able to gather ALL synthetic performance and the code in the program, you should be able to predict real world performance.