AMD has had a rough time keeping up with Intel in the CPU market for the past couple of years. The firm's Bulldozer CPU architecture hasn't worked out as planned, and AMD has been forced to limit the scope of its ambition. Rather than competing with Intel across the entire market, it's had to choose its battles, carefully positioning its products and seeking any possible seams in Intel's armor. That strategy has succeeded in places, producing some potentially solid values for end users in the process.
When we reviewed the desktop version of AMD's Trinity APUs last fall, the A10-5800K and A8-5600K, we found that they were a near-miss on the value front. The price was right compared to the competition. The CPU performance was good in some applications—those that use multiple threads well, like image processing tools or video encoders. And the integrated Radeon graphics simply outclassed what Intel had to offer in its Core i3.
But AMD was asking folks to accept some compromises in other areas. Single-threaded performance was relatively pokey, which could lead to reduced responsiveness in everyday tasks and momentary hiccups while gaming. More worryingly, the A10-5800K and A8-5600K chips consumed quite a bit more power than the competing Core i3—nearly twice as much at peak, 100W versus 55W. More power draw means more noise and heat, higher energy bills, and the need for larger PC enclosures.
In the end, that mix of advantages and drawbacks didn't make a lot of sense to us. We just couldn't envision many home-built PCs or pre-built systems where accepting a 100W power envelope to get somewhat better integrated graphics was a winning combination.
Just half a year later, AMD has refined the formula for its A-series processors under the umbrella of a new code name: Richland. Has the firm made enough progress to earn an unqualified win over the Core i3? We've done entirely too much testing in order to find out.
Trinity becomes Richland
Products based on Richland technology are distinct from the prior-generation Trinity parts, but not in the way you'd expect. Richland and Trinity share the same 32-nm silicon, with four integer cores, two floating-point units, and integrated Radeon graphics. On the desktop, they fit into the same Socket FM2-style motherboards.
The differences have to do with the power management capabilities programmed into the chip's firmware. Richland adds three big things that AMD simply didn't have time to include in the first generation of products based on this chip. All of them have to do with dynamic behavior, how the APU's onboard power management microcontroller directs the CPU and graphics cores to scale their clock speeds in response to different workloads.
First, AMD has spent more time in its labs characterizing what these chips can do—what clock frequencies they can tolerate and how much voltage they need to get there. Thus, Richland-based products have more operating points between their base and peak Turbo Core clock speeds. This finer-grained control should translate into higher efficiency, better performance, or both.
Second, Trinity contained an embedded network of temperature sensors across the chip, but those sensors weren't used in determining how far Turbo Core could push on clock speeds. Instead, Trinity estimated its own power use by looking at activity counters. Because conditions vary in the real world, this method of estimating power use, and thus temperature, must rely on some fairly conservative assumptions. Richland's rebuilt power management algorithm, called Hybrid Boost, takes direct input from the chip's temperature sensors. Armed with better intelligence, Hybrid Boost can be more aggressive about pursuing higher clock speeds, giving Richland chips more frequency headroom.
Finally, AMD has added some smarts to Richland's firmware that attempts to determine when one of the two major components of the chip, either the CPU or the integrated graphics processor (IGP), is the primary bottleneck in the current workload. For instance, if the IGP is the main holdup, Richland can rein in the CPU cores and shift more of its power budget to graphics in order to achieve better overall performance.
As you might imagine, all of this power-saving wizardry should have the most tangible benefits in laptops. Still, Richland delivers some modest improvements on the desktop, as well. CPU base and Turbo frequencies have risen by 200-300MHz, and IGP speeds are between 40 and 84MHz higher, both within the same power envelope. The new A-series lineup looks like so:
|A10-6800K||2/4||4.1 GHz||4.4 GHz||4 MB||384||844 MHz||100 W||$142|
|A10-6700||2/4||3.7 GHz||4.3 GHz||4 MB||384||844 MHz||65 W||$142|
|A8-6600K||2/4||3.9 GHz||4.2 GHz||4 MB||256||844 MHz||100 W||$112|
|A8-6500||2/4||3.5 GHz||4.1 GHz||4 MB||256||800 MHz||65 W||$112|
|A6-6400K||1/2||3.9 GHz||4.1 GHz||1 MB||192||800 MHz||65 W||$69|
The A10-6800K is the new flagship, and it sets the template. Compared to A10-5800K introduced last fall, the 6800K has a 300MHz faster base clock and a 200MHz higher Turbo peak. The IGP clock adds another 44MHz, too, while the max power (TDP) remains steady at 100W.
The model numbers that end in K, like 6800K, indicate unlocked parts whose multipliers can be raised at will for easy overclocking. Unlike Intel, AMD doesn't delete features on its unlocked CPUs, so the 6800K still has all of its virtualization capabilities and advanced instruction support intact. Also, alone among the new Richland parts, the 6800K officially adds support for DDR3-2133 memory.
That said, perhaps the most eye-opening improvement over Trinity comes in the form of the A10-6700. Compared to the A10-5800, the 6700 has similar CPU clocks (100MHz lower base, 100MHz higher Turbo max) and a 44MHz faster IGP speed—and it manages that in a 65W power envelope, 35W less than the 5800K. That's true progress, folks. Better yet, AMD has supplied us with an A10-6700 for testing, so we can see exactly how it matches up against the Core i3.
We've also taken this opportunity to test some chips we've so far neglected: the lower-end models in AMD's FX lineup. Like Richland, their CPU cores are based on the "Piledriver" microarchitecture. Beyond that, the two chips diverge. All of the FX models are based on a chip code-named Vishera, which natively has eight integer cores and, unlike the APUs, an 8MB L3 cache. Vishera doesn't have integrated graphics or PCI Express, so it must rely on external chips (a discrete GPU and the 990FX chipset) to provide those capabilities. Naturally, then, the FX processors make use of a different CPU socket, dubbed Socket AM3+.
The fastest Vishera-based offering is the FX-8350, which we reviewed last year. AMD has since added a couple of cheaper options based on Vishera chips with portions disabled, the FX-6350 and FX-4350.
|FX-8350||4/8||4.0 GHz||4.2 GHz||8 MB||125 W||$195|
|FX-6350||3/6||3.9 GHz||4.2 GHz||8 MB||125 W||$132|
|FX-4350||2/4||4.2 GHz||4.3 GHz||8 MB||125 W||$122|
These CPUs have similar clock speeds with varying core counts. From a computer-nerd standpoint, it'll be interesting to see how these differences affect performance. Also from a computer-nerd standpoint, it's a little disappointing to see that even the quad-core FX-4350 requires 125W to do its thing.
Happily, like the K-series Richland chips, all of these FX processors have unlocked multipliers without having any features disabled. They're also fairly cheap. The six-core FX-6350 costs less than the quad-core A10-6800K.
You know, it wasn't supposed to be this way. AMD's CPU lineup is stacked closely together at fairly modest prices due to competitive pressure. Almost assuredly, the plan was for Socket FM2-based APUs to compete against Intel's Ivy Bridge and Haswell quad-core CPUs with integrated graphics. The FX series, which is derived from server-class Opteron tech, would then go up against Intel's high-end platform based on the X79 chipset, which is based on Xeon server tech. Instead, the eight-core FX-8350 sells for under $200, and everything else must cost less than that.
For now, the closest competition for the A10-6800K and A10-6700 is Intel's Core i3-3225. That's a 22-nm chip with dual cores, four hardware threads, a 3MB L3 cache, and Intel HD 4000 integrated graphics. The i3-3225 requires a bit less power than even the A10-6700, with a 55W TDP, and has a slightly lower list price of $134. That price also sets the Core i3 against the FX-6350, albeit in a much smaller power envelope. The Core i3-3225 is based on Intel's older Ivy Bridge architecture; Haswell hasn't quite made it into this price range yet, although it's sure to get there eventually. When it does, Haswell should bring better graphics performance with it. For now, though, AMD has a bit of an opening. Let's see how well Richland takes advantage of it.