For years now, those scoping out a system (or just the parts to build one) have had a simple choice to make when it comes to processors: Do you want the best performance (Intel) or do you want the best value (AMD)? With its new FX series of CPUs, AMD is proposing that maybe you don’t have to choose after all. The first desktop member of the family we've seen, the AMD FX-8150, which is due for release later in October, may offer Intel cause for concern down the line, but shouldn't worry the company quite yet. The FX-8150's eight processing cores (the first consumer chip so equipped), performance with multithreaded applications, and $245 list price are solid, the chip's hunger for power and struggles with single-thread workloads keep it from unseating Intel in the midrange space—at least for now.

The FX-8150 is based on AMD’s “Bulldozer” core design, the company’s first major hardware rethink in years. Each individual dual-core Bulldozer module is designed to optimize resources, with functions with high utilization (such as integer pipelines and Level 1 data caches) dedicated in each core and everything else (fetch, decode, floating point pipelines, and the Level 2 cache) shared, which lets it both use a higher-performance function unit and reduce the overall die area on the CPU itself. AMD claims that this design offers improved “scalability and predictability” on multithreading applications, with its threading method superior to one in which two threads share a single core. (This is, ahem, the Intel method.)  AMD has also implemented a number of instruction set extensions, such as for SSE 4.1 and 4.2 and Advanced Vector Extensions (AVX), though not all of these will deliver much benefit until Windows 8 comes out.

But that’s all on the inside; from the consumer standpoint, the FX-8150 appears considerably more conventional. Though it's based on a newer 32nm production process that replaces AMD's 45nm process, on the exterior it's of an identical design to AMD’s previous chips, with its pins on the bottom (as opposed to Intel’s method of putting the pins are in the socket rather than on the chip itself). Just don't assume your existing motherboard will work with it: It must use the relatively new AM3+ socket, though in accordance with AMD’s dedication to backwards compatibility, you’ll still be able to use your older chips in the newer board.

The FX-8150 has a 3.6GHz base clock that can be increased to 3.9GHz when each processing core is taking advantage of AMD’s Turbo Core functionality. If they’re all maxed out, the chip’s speed rockets to 4.2GHz. In case this still isn’t enough speed for you, the FX-8150 has an unlocked multiplier; AMD insists that all chips in the FX family will. (This was a benefit previously reserved for the company’s higher-end “Black Edition” processors.) The L2 and L3 caches each measure 8MB, and the chip can use up to DDR3-1866 RAM. One thing this CPU is missing: integrated graphics, like those you’ll find on its Llano-based APU cousins, like the A8-3850 and A6-3650, so you’ll need at least one discrete card in a Bulldozer system. If you want to use two, the FX-8150 supports 2x16 CrossFireX, so both of your linked video cards will operate at higher speeds.

Performance and Power Usage
All this helps the FX-8150 in terms of performance, even when compared with AMD’s last flagship CPU, the six-core Phenom II X6 1100T, though not as much as you might expect—and certainly not in every situation. When multithreading was involved, the results were nice: The FX-8150 earned 6.01 in our CineBench R11.5 rendering test; the 1100T managed 5.88. When converting a video clip to iPhone format using the open-source program Handbrake, the newer chip finished the task in 1 minute 44 seconds, and the 1100T needed 1:48. The new instructions helped the FX-8150 make a major leap in our TrueCrypt 7.1 cryptography test, rising from 176MBps with the 1100T to 224MBps. We saw a smaller improvement in the Futuremark PCMark 7 full-system benchmark test (2,949 for the 1100T and 3,017 for the FX-8150), and slightly more when running Futuremark 3DMark 11’s CPU-based Physics test (a frame rate of 16.91 on the 1100T rose to 19.66 on the FX-8150).

When we used our FX-8150–based test system to apply a dozen filters to a large image using Adobe Photoshop CS5 Extended, its time of 4 minutes 12 seconds was superior to the 4:16 the 1100T required, but with one wrinkle: the Smart Blur effect. The 1100T actually finished this task 18 seconds faster than the FX-8150, so ignoring that one filter would give the newer chip an even greater lead. We asked our AMD contact about this, and he responded, “There will be some instances where calculations will be faster on the Phenom II X6, to do with threadedness, thread scheduling, types of calculations, etc.” We’re still waiting to hear more specifics, and when we do we’ll report them, but the takeaway is that the FX-8150 will not always be faster than its predecessor, something that's troubling for a CPU at the forefront of its company's line.

These patterns continued against Intel’s current-generation chips. The FX-8150's multithreaded CineBench R11.5 score fell somewhere between the Core i5-2500K’s 5.46 and the Core i7-2600K’s 6.94; but it proved much better at Cryptography in TrueCrypt (the Core i5-2500K earned 143MBps, the Core i7-2600K 196MBps). On the other hand, both Intel chips turned out superior frame rates in the 3DMark 11 Physics test (20.37 for the Core i5-2500K and 26.15 for the Core i7-2600K) and better overall results in PCMark 7 (3,471 and 3,649). In terms of media handling, there was no contest: They plowed through Photoshop in 3 minutes 4 seconds (the Core i5-2500K) and 2 minutes 43 seconds (the Core i7-2600K), and each finished Handbrake in 1 minute 11 seconds—more than half a minute less time than the FX-8150 required. (Note: This is with a discrete video card in all cases.)

But take away the multithreading, and major problems emerge for the FX-8150. When running CineBench R11.5 in single-core mode, its score was pathetic: 0.96, not just well behind both the Core i5-2500K (1.49) and Core i7-2600K (1.54), but even AMD's own last-generation 1100T (1.10). This shows that, unless the software you use is dependent on multithreading, the FX-8150 is not a wise choice.

Nor is it ideal if power usage is a significant concern. We measured the full draw of our test-bench system using an Extech Datalogger and discovered that, at idle, the FX-8150 used 155.9 watts compared with the 1100T's 145.6. That's not too bad, but under load the FX-8150's usage skyrocketed up to 250.2 watts, whereas the 1100T topped out at 178.5 watts.  The FX-8150's performance is simply not remarkable enough against the 1100T's to justify that extreme a difference.

If you're determined to build a midrange-or-better AMD computer, the AMD FX-8150 is in no way a bad choice, particularly if you regularly depend on multithreaded software to wrap up your daily tasks before the beginning of the next century.  But the 1100T is overall a better value if you're not in desperate need of additional threads. If, on the other hand, you're platform agnostic, the choice is easier. Intel's chips put up a fierce fight, particularly in terms of media manipulation and gaming, and won't draw as much power under load, and those qualities will make them a superior choice for many users. The good news is that AMD's new FX line shows the company takes its competition seriously, and is laying the groundwork for more exciting innovations in the years to come.  There's every reason to believe that AMD can make this gambit work and steal away the price-performance crown from Intel, which captured it with its sizzlingly placed Core i5-2500K.  That time may not be far away, but the FX-8150 shows it's not here yet.

More CPU Reviews:
•   Intel's First Dedicated Graphics Cards to Launch in 2020
•   US Poised to Take Back Top Supercomputer Ranking From China
•   New AMD Ryzen Threadripper Has 32 Cores, 64 Threads
•   Nvidia Walks a Tightrope Between GeForce, AI
•   Intel Celebrates x86 40th Anniversary with 5GHz Core i7
•  more