and this is naming the fact that this linux software is not even optimized for AVX/FMA/XOR and the other goodies in bulldozer, maybe for average joe wannabe "expert" is "slower than sb dude why waste money?" but for us is geek candy on the cheap.
Originally Posted by chris200x9
for the smart mouths FMA and XOR really make huge diff if used properly and your cache is sane and remember that AVX can process 4 double(very rare is most workloads) or 8 floats(very common). in my case i modified my old SSE idct[Mx'M] code to use avx and FMA and loop it to run 1 million times and is around 35/45% faster in the same bulldozer cpu, so bd have a lot of juice to extract yet.
bd problem is most software outhere is poorly threaded and/or with crappy cache management and/or barely uses SIMD at all(this applies to SB too though)
I just hope that the Catalyst drivers will work fine with Linux. Do you expect any problems with the drivers under Linux?
Originally Posted by chris200x9
Catalyst should already work with Trinity.
Originally Posted by Anarchy
Without SB 2700K this comparison just doesn't make sense, because IB 3770K is its direct competitor in terms of price and CPU features (HT & frequency).
I'm excited to see how the Piledriver and Trinity CPUs work out against this!
i have high hopes on those piledriver cores since they should pack more raw horse power to deal with sandy/ivy bridge and crappy windows apps wich will translate in a huge boost for properly optimized applications and if you mix in there GCN that should provide a nice workhorse CPU for heavy compute apps using opencl.
Originally Posted by fuzz
HardOCP has the review you're looking for. Their conclusion? You can't put a piece of paper between the two.
Originally Posted by birdie
Also, it probably doesn't help to market it as 8-core cpu, since it has worse performance than the previous generation 6-core cpu in most tests. Maybe they should have called it 4-cores with Instruction Hyper Multiplexing Turbo Parallel Power ModeTM or some other fancy marketing name. Look at it now: it's sold as an 8-core CPU at 3.6GHz and it's slower or equivalent to a 4-core i7. In some tests (games mostly) it's slower than i3 and that's dual core.
Originally Posted by jrch2k8
Thanks for the article!
I found some of the article a bit unclear though. Some feedback on how it could have been clearer:
- I think there was an excessive number of CPUs being compared. Given that this site is not focused on over clocking, those variants should have been dropped. The excess number made some of the coloured graphs very difficult to read.
- As noted by others, it would have been good to have a 2700k as it is a more direct generation comparison
- I don't think the 3960X was not useful to have in the benchmarks with it's extra cores as it just confused the sandy bridge vs Ivy bridge results
- I would have liked a table of supported/unsupported CPU/chipset features under Linux to make support status clearer
well i give ya that AMD marketing team aren't the brightest bulb of the box(could be a management or upper dogs issue tho) but like i said when you properly optimize your code to use AVX/FMA4/XOR/aes-SI/etc bulldozer fulfill most of his promises and the performance is outstanding for 200$ or less (i would love bandwith memory fixes and maybe a bigger l2 cache).
Originally Posted by devius
AMD bulldozer is an excellent processor for heavy duty task and was designed for that, the problem is 95% of the apps outthere are barely threaded and almost none of that 95% uses SIMD let alone proper cache usage and in the other 5% most apps use Intel C/C++ compiler witch (intentionally) select the most fucked up trashy unoptimized compile route possible on AMD processors. So AMD is in a pretty bad position here since AMD processor need a lot more horse power to reach a comparative level of performance in the apps it can shine <----(don't ask me links use google you will be impressed of how nasty is the performance hit when using intel compiler showing and AMD cpu and when you hack it to mask you AMD cpu as an intel CPU)
the way i see it BD is a future proof design and i infere AMD is looking to get ride of many slow/legacy crap from the x86 world from their cpu's but since most apps still heavily the 486 parts of the CPU and not the latest much more optimized route due the reasons previously mentioned they need to mantain acceptable levels of performance for both worlds hoping that this trend will change and apps will adapt to use these new much more efficients techs in the mainstream market.
so yes today SB/IB will be faster than BD/PD in most workloads but is not cuz a crappy design but more the reasons i stated, but if you take the time to code some optimized example in SB and BD you will see another picture entirely.
in the case of games you are 150% wrong if you even believe mainstream games are state of art code uber optimized to exploit the latest drop of performance of your PC (CPU/GPU), at the contrary is probably the most hackish crappy code you will see in your life, let alone proper scalability or threading/etc.