Originally Posted by Alatar
So what tests actually showed that the bias in R11.5 affected AMD results by 10% or 20%? I'm not aware of any tests like that. Actually I'm not aware of any tests that showed anything about R11.5 performing differently on different instruction sets on AMD hardware. Mainly because as far as I'm aware no such test was ever made.
My conclusion on the bias not affecting the results all that much is mainly based on:
1) The results made sense (8350 very close to 3770K both at stock etc.)
And more importantly:
2) R15 results are relatively close to R11.5 ones when comparing intel and AMD CPUs. R11.5 has bias, R15 doesn't. Therefore it's a pretty reasonable assumption that R11.5 while biased, didn't skew the results all that much.
Not that it really matters in this case, all of us know that cine is perfectly fine for comparing AMD vs. AMD.
There are facts and there are opinions. CB 11.5 is biased, that's a fact. How much is the "value" of that bias is a matter of opinions and guesses. But from the moment we know it is biased, it automatically has to be rejected as a mean to judge relative performance between processors from different vendors, exactly because it artificially hampers non intel products. No way around this. I know it mirrors Matrox Cinema performance but even so it must be limited as a judge of performance for those that are interested in this particular piece of software, not as a general indicator. When we are talking about synthetic benching, we aren't looking for results that "make sense", but for results that represent true performance. FX-8350 trails 3770k on CB11.5 MT test, not by much but it does. Things would have looked very different on reviews If FX actually was faster there, even by a hair. And it wouldn't take a big boost for this to happen. So, yeah, bias plays a big role on impressions made.
CB 15, from what I see (looking at results that is) behaves more or less like CB11.5. I've seen people assuming it is fair now, and I am not saying it isn't, but I just don't quite understand where that notion comes from.
A stock FX-8350 scores 101 ST/643 MT on CB 15. A stock i7-3770k 143/708 (win7 results). The i7 has a ~43% lead for ST and ~10% for MT.
A stock FX-8350 scores 1.1 ST/6.89MT on CB 11.5. A stock i7-3770k does 1.66/7.61. The i7 has a ~50% lead for ST and ~10% for MT. Same ballpark, really. My guess is that CB15 is still running some sort of SSE4 under intel (the gap would have widened if they have used AVX or something) and SSE2 for AMD still. If that is the case, and kaveri is tuned to have higher ipc under newer instructions vs richland, then CB is again a piss poor indicator of performance even when it is AMD vs AMD.Edited by Kuivamaa - 1/8/14 at 6:19am