AMD Bulldozer - FX 8150 Performance Review



Product: AMD FX 8150 / Asus Crosshair V
Company: AMD
Author: James Prior
Editor: Charles Oliver
Date: October 11th, 2011

Overclocking

Overclocking the FX-8150 is a piece of cake, thanks to the unlocked multiplier, massive overclocking features in the ASUS RoG Crosshair V Formula, and platform consistency from AMD. Our first look at overclocking, we used AMD Overdrive to see if we can adjust the turbo mode to give us an all-core overclock. Interestingly, we found that AMD Overdrive only allows for a maximum of six cores to be Turbo'd, despite the all-core turbo mode being selected.

Switching to the BIOS, we disabled Turbo and headed for a 22.5 multiplier to see if a 4.5GHz clock would be doable. From AMD Overdrive we saw that the idle voltage was 0.8v, at 3.6GHz it is 1.275v, and at Turbo it is 1.4125v (for both all core turbo 3.9GHz and max turbo 4.2Ghz). Taking that as our guide, we set the vCore to 1.4125 and started Prime95 stability testing. This is where we noticed something interesting, the clock speed would flicker from 3.3Ghz to 4.5Ghz, when a consistent load was applied at all cores. Thinking this was an artifact of Cool'n'Quiet, we disabled CnQ and retested - same behavior. Only after disabling CnQ and C1E as well as C6, would we get a solid stable clock frequency.

At this point our Prime95 test would error randomly after a few minutes, which we traced back to vDroop. Despite setting 1.4125v in the BIOS, under load the core voltage would drop to ~1.38v which was not sufficient for all cores. ASUS implemented Load Line Calibration to compensate for this and, by gradually moving from normal (0%), we found the setting which gave us the vCore we set, under load - Ultra High, which is 75% LLC, the penultimate setting before Extreme and 100%. Upon retesting we found that CPU overcurrent protection was kicking in and hard powering off the system, so we upped that to 140% (Extreme level), which worked. We then dropped vCore to 1.4v and retested, and found a stable 4.5Hz 8-core system.

Scorpious Platform Analysis

Before we benched all our systems, we wanted to see the effect of different parts of the system on performance. After setting our stock baseline performance with the Corsair 2x4GB DDR3-1333 RAM, we dropped in 2x2GB of Mushkin Blackline PC3-16000 (DDR3-2000) so we could explore the effect of different memory clocks.

Test Result
Memory 1333 1600 1866 2000
Timings 9-9-9-24 1T 9-10-9-24 2T 10-11-10-24 2T 11-11-11-24 2T
Dirt 3 (Avg FPS) 116.3 118 118.7 117
Crysis 2 (Avg FPS) 67 69.3 67.8 68.2
X264 Bench (Pass 2, min fps) 37.82 38.23 38.26 38.06

We repeated several of our tests to determine if at stock clocks, the FX processor benefited from increased memory speed by stepping from DDR3-1333 to DDR3-1600, 1866 and finally DDR3-2000. Based on this performance testing (more details later) it seems that DDR3-1600 is the best bang/buck option for FX processors. Given that the FX is apparently slightly hindered at stock speeds with DDR3-1333, this gave us motive to repeat our overclock - up the base clock, and run with higher speed memory.

Using a 215MHz base clock we can use the DDR3-1866 memory speed setting to simply give us PC3-16000/DDR3-2000 speeds. Combining this with our previous overclock of 4.5GHz we can rebench and see if there is an appreciable difference. We attempted higher overclocks, but found the voltage required to hit 4.7GHz was a lot more than 4.5; 1.49375v vCore was required for prime stable 4745MHz operation. This quickly looked to exceed the cooling capability of our Noctua NH-D14, robust and manly as it is, the heat dissipation needs of a 1.5v Bulldozer are likely only going to be met by a liquid cooling system. We then decided to focus on the Northbridge and see if increasing speed would help system performance. We were already at 2.36GHz NB clock with the 215MHz HT reference clock, so we decided to try for 2.580GHz at 1.375v CPU-NB voltage. HT speed was set to 2580MHz, close to the original 2.6GHz default speed.

Repeating our sample benchmarks showed us the difference from our stock platform with ddr3-1600, platform with 4.5GHz CPU at ddr3-1333 and 4.5GHz CPU, 2.6GHz NB and DDR3-2000 platform:

Test Result
System 3.6 / 1600 4.5 / 1333 4.5 / 2000
Timings 9-10-9-24 2T 9-9-9-24 1T 11-11-11-24 2T
Dirt 3 (Avg FPS) 118 118.9 120.7
Crysis 2 (Avg FPS) 69.3 68 75.4
X264 Bench (Pass 2, min fps) 38.23 46.03 47.28

Now stable with memory, northbridge and cpu cores overclocked, we overclocked all our test platforms, detailed thus:

Component Overclock
Processor FX-8150 Phenom II X6 1100T Core i7 920 Core i7 2600K
Enabled Cores/Threads 8/8 6/6 4/8 4/8
Clock 4.5GHz 4.1GHz 4.0GHz 4.5GHz
vCore 1.4v 1.475v 1.425v 1.475v
bClk x Multi 215 x 21 200 x 20.5 200 x 20 100 x 45

Power - Using a Kill-a-watt power meter we recorded total system power draw for three use cases, for different conditions. Idle represents Windows 7 idle with aero enabled, MS Office 2010 Word and Excel running and a PDF document open in Adobe Reader X. Load is under all cores/threads Prime95 torture test of inplace FFT's. Gaming is the peak power recorded during 3DMark11 Combined test at Xtreme settings.

We recorded peak temperatures using AMD Overdrive or CPUID HWMonitor (for the Intel systems) during the above tests.