AMD Bulldozer Review: FX-8150 Gets Tested

Overclocking FX-8150 (On Air)

During its press briefing in Austin, AMD’s architects suggested that Bulldozer was designed to hold IPC steady and facilitate significantly higher clock rates. Regardless of the IPC story, we're still expecting solid overclocking headroom.

AMD gave us a guide on what to expect, based on its own experiences, mitigating our suspense. On air, its team had seen a sustainable 4.6 GHz across all cores. Liquid cooling pushed that ceiling up to about 4.9 GHz. Nitrogen and helium—both for the sake of exhibition on an architecture with no cold bug—achieved 7.7 and 8 GHz, respectively, with all cores enabled.

The FX-8150’s base frequency is 3.6 GHz. Its intermediate Turbo Core setting (the clock at which all eight cores can operate simultaneously) is 3.9 GHz. With at least four cores (two modules) idle, Turbo Core is able to push up to 4.2 GHz. Now, using 1.4125 V, I was able to achieve stability at 4.5 GHz on all cores. That frequency was good enough to bring down our SolidWorks 2010 workload from 3:24 to 2:57. But that was also right at the brink (I’m not used to seeing >66°C from an AMD chip; check out the screen capture below).

A more granular approach would be to set a lower base clock rate in AMD’s Overdrive software, and then configure aggressive P1 and P0 states for Turbo Core as modules fall idle. Your threaded apps won't realize as much benefit, but you also stand a better chance of hitting higher clocks at lower voltages in programs like iTunes and WinZip, which only tax one core.

Memory Scaling

Thanks to the changes made to Socket AM3+, Bulldozer enables a much more flexible memory controller than its predecessor. Officially, AMD supports data rates of up to 1866 MT/s. However, you can push even higher than that using the right motherboard.

At 1600 MT/s, the FX-8150 is able to match Intel’s Sandy Bridge architecture. That’s an impressive feat, given the bandwidth Phenom II X6 extracted from two channels of DDR3 memory.

Is there a reason to buy higher-end memory kits for an FX-oriented build? I yanked the high-density kit out of our test bench and installed a smaller (2 x 2 GB) DDR-2133 kit from Kingston to measure scaling in WinRAR, which we’ve seen exhibit sensitivity to memory performance in the past.

SiSoft Sandra 2011 shows that bandwidth continues scaling right up to the processor’s official ceiling, while WinRAR demonstrates that seemingly big throughput gains don’t always translate to the same real-world advantage. Even still, it’s good to see that there is additional performance available if you’re willing to tweak your memory subsystem.

Stay Cool, My Friends

At launch, AMD plans to ship its FX CPUs with the same heat sinks you find in a Phenom II X4 or X6 Black Edition processor-in-box package.

Perhaps you’ve also heard rumors that AMD might sell certain models with water cooling kits. This is, in fact, going to happen. However, they’ll be limited to specific markets, starting with Japan. The information we have from AMD right now suggests that the US should see its own closed-loop water cooling kits at some point. But the “when” hasn’t been determined yet.

AMD let us know it shipped samples of this kit to our offices just before launch. However, as of this writing, they haven't arrived.