The 4.70 GHz benchmarks are clearly throttled, try decreasing the clock one step
Phoronix: AMD FX-8150 Bulldozer Overclocking On Ubuntu Linux
While in the weeks since the launch of the AMD FX "Bulldozer" processors we have looked at many areas of computing performance for the FX-8150 CPU, from the compiler tuning to multi-core scaling, one area that hasn't yet been covered under Linux is the AMD FX-8150 overclocking. But this article changes that.
http://www.phoronix.com/vr.php?view=16720
The 4.70 GHz benchmarks are clearly throttled, try decreasing the clock one step
Indeed, keeping 4.7ghz was quite useless, obviously with sporadic performance and huge power fluctuations it was clocking itself down. I would be curious to see if there is a setting somewhere between 4.42 and 4.7 that yeilds better results.
Not being very interested in overclocking, I vaguely remember Cool&Quiet being incompatible with BIOS overclocking. Is that still a problem?
So after all the benchmarks, what's the final verdict? Is Bulldozer promising or a flop? Seems like I've seen results that point both ways, even within the same application type. I'm still trying to decide if it's something I want to upgrade to in the near future, and all the benchmarks I've seen have just confused me more.
Coming from a long time amd fan, while bulldozer does well at highly parrallel use cases, its overall performance and efficiency are highly dissapointing, its performance per clock and performance per watt are a step behind their previous generation. I personally wouldn't consider getting a bulldozer over a phenom ii x6
A decent amount of performance boost can be achieved with the stock air cooler, while keeping power usage low.
By slightly raising the "FSB" clock (215MHz), and raising *a lot* the TurboCore multiplier (>4.6GHz) you'll see that the system is perfectly stable.
But raising the main CPU multiplier is relatively harder !
I've spent little time playing with the BIOS, so my granularity is high, but AFAICT I can't go past 4.1GHz with a *completely* stable system.
The "pts/cpu" benchmark suite ran in a loop and I thought the system was stable, then I compiled a kernel (--jobs=10), which yielded issues.
Did it again, got a lot of invalid instructions, segfaults, oopses, depending on whether the CPU decides to crash in user or kernel space...
The reported CPU temperature was less than 70°C, be it for the CPU sensor or the mainboard socket sensor (which conditions the fan speed).
A definite flop. Save your money and buy a Sandy Bridge system.
http://www.extremetech.com/computing...cratch-at-tsmc