AMD has been telling anyone that will listen that it’s new Piledriver core is faster than that old thing called “Bulldozer” for quite sometime now. Fortunately for AMD, after reviewing both the Piledriver based A10-5800K and the Piledriver infused FX-8350 over the past few months, I can safely say that Piledriver is faster than Bulldozer. But the question that we’re aiming to answer here is how much faster is it? If we go by what AMD promised when the first generation of FX CPUs launched last year, we should be looking at a 10 to 15 percent increase in performance.
Now this 10 to 15 percent figure is a combination of two kinds of enhancements. First off AMD has improved performance by tuning its Piledriver core so that it produces more performance, than Bulldozer did, for every clock cycle that it spends working on a task. Additionally, AMD has improved performance by clocking it’s Piledriver based chips higher. The Bulldozer based FX-8150 had a 3.8Ghz base clock and the Piledriver based FX-8350 has a base clock of a cool 4Ghz. So between an unspecified increase in performance per clock and a five percent boost in raw clock speed, AMD was targeting a 10 to 15 increase.
It’s easy to see how, and to what degree, the raw clock speed boost helped performance; but what’s less clear is exactly how much better AMD’s Piledriver core performs than it’s Bulldozer core. To find out we dusted off an old FX-8150 and warmed up the testing platform from our recent FX-8350 review. We set both of our contenders at a clock speed of 4Ghz, a voltage of 1.45, and a memory speed of 1600Mhz. Additionally, we disabled the Cool ‘n’ Quiet, Application Power Management, and Turbo Core features in our motherboard’s BIOS. The data that we’re looking at below is as close to a direct, core to core, comparison as possible.
Moving to the benchmarks we tested these chips in a modified version of the CPU test suite we used for the Trinity and Vishera reviews. We’ve added a new browser based test, called Chalkboard that should reflect single core performance in web applications.
Starting with Cinebench we can see that Piledriver offers an eight percent advantage in the single threaded test and a smaller six percent advantage in the multi-threaded test. Moving to our Paint.NET benchmark, Piledriver is once again ahead by eight percent. In Truecrypt there is a three percent gain in AES performance and a six percent gain in ATS performance.
Looking at Handbrake, Piledriver outperforms Bulldozer again by about six percent. In 7-zip we have an eight percent gain, and in the SunSpider benchmark we have the largest advantage shown in our testing, a whopping 12 percent. Octane shows Piledriver to be nine percent faster, but wPrime shows only very small performance advantage, one to two percent, in favor of Piledriver. Our final test, Chalkboard, shows a seven percent advantage in favor of Piledriver.
The big take away from our testing here is that the Piledriver core is without question faster than the Bulldozer core. But the Piledriver core shows the biggest gains over Bulldozer in single threaded applications, as opposed to multi-threaded applications, which have traditionally been AMD’s strong suit. Poorly threaded web based tests like the SunSpider and Chalkboard benchmarks showed much larger gains than well threaded tests like our wPrime and Truecrypt benchmarks.
Moving to gaming benchmarks we tested out two contenders in three games today; Crysis 2, Counter-Strike: Global Offensive, and Nexuiz. All of our benchmarks were run in 1080P with all of the in-game quality, filtering, and anti-aliasing settings maxed out. We used AMD’s HD 7970 GHz Edition as our discrete graphics card.
In our Crysis 2 benchmark it’s basically a tie between Bulldozer and Piledriver. There’s not much to say here other than both chips provided totally playable experiences.
Looking at our Counter-Strike: Global Offensive results shown above we can see that Piledriver offers a measurable advantage over Bulldozer. Again both chips provided perfectly playable experiences.
Nexiuz is a game that, even with Vsync disabled has a frame cap at 63 frames per second. But despite this pitfall, it offers us a chance to see the differences between Bulldozer and Piledriver in the context of a frame rate capped engine. Our Bulldozer chip offers significantly less stable performance than our Piledriver chip, with frame rate drops every seven seconds or so, compared to drops every twenty second on the Piledriver system.
Considering the performance numbers that we’ve seen in these three benchmarks it’s clear that Piledriver at worst offers equivalent performance to Bulldozer, and in certain cases can offer a smoother experience.
Now let’s condense everything down into a single average improvement number. Six percent is the average advantage that Piledriver offers over Bulldozer when both are in an eight core configuration, and clocked at 4Ghz.
Thinking back on AMD’s claim of ten to fifteen percent increases in performance I think it’s fair to say that they have achieved that goal. The FX-8350 has a five percent clock speed advantage over the FX-8150, in addition to a six percent average in per clock performance improvement. Thus we have a combined 11 percent performance increase, which is within AMD’s guidance range.
Of course AMD’s Piledriver core performed differently depending upon the application, but I think that it’s fair to say that AMD by and large met it’s promise. Putting this gain in the context of the larger market reminds us that AMD is still in an untenable position. Small performance gains like this one, are not enough to bring AMD back up to parity with Intel’s CPU offerings. AMD can’t hope to survive on ten to fifteen percent performance increases like this one, but they sure don’t hurt.S|A
Latest posts by Thomas Ryan (see all)
- Apple Maintains UX Dominance With NVMe SSDs - Nov 15, 2016
- Nvidia’s Q3 2016 Was A Record Breaker - Nov 14, 2016
- AMD Launches The Radeon Pro WX Series - Nov 7, 2016
- Intel Attacks IoT with the Atom E3900 and A3900 - Oct 25, 2016
- ARM Goes for IoT with the Cortex-M33 and Cortex-M23 Cores - Oct 25, 2016