AMD A6-3650 Llano APU Review

[MAC]

Power Consumption / Temperature Testing

Power Consumption

For this section, every energy saving feature was enabled in the respective BIOSes and the Windows Vista power plan was changed from High Performance to Balanced.

For our idle test, we let the system idle for 15 minutes and measured the peak wattage through our UPM EM100 power meter.

For our CPU load test, we ran Prime 95 In-place large FFTs on all available threads for 15 minutes, measuring the peak wattage via the UPM EM100 power meter.

For our overall system load test, we ran Prime 95 In-place large FFTs on all available threads for 15 minutes, while simultaneously loading the GPU with OCCT v3.1.0 GPU:OCCT stress test at 1680x1050@60Hz in full screen mode.

As expected, the A6-3650 features very similar power consumption figures as the A8-3850, but given its slightly higher idle and load voltages it does draw a bit more wattage. As we mentioned in the launch article, we expected Llano to have lower power consumption due to its new 32nm manufacturing process, but its die does have a chart-topping 1.45 billion transistors, by far the most of any consumer-oriented processor.

Frankly though, this processor has a huge chunk of its die dedicated to graphics, and it was designed to be used in a simple system without a discrete GPU, so let’s take a look at the power consumption when using the IGP.

First let's restate that the Sandy Bridge chips in this graph have an almost unfair advantage in the form of the Intel DH67BL motherboard. This Intel-manufactured motherboard has unmatched idle power consumption, easily 10W less than comparable motherboards from the big three motherboard manufacturers. While the A6-3650's idle power consumption numbers are very good, they are a bit higher than the A8-3850's due to a higher idle voltage (0.456V vs. 0.444V). When you load up the CPU, power consumption rises greatly, and it’s not really competitive with its main competition, the Core i3-2100 series. Thankfully, the power consumption doesn’t rise much when you also tax the integrated GPU, so overall you end up with a system that should peak at about 150W max.

Temperature Testing

For the temperature testing, since we were not given a default cooler from AMD, we used a Thermalright Ultra-120 Extreme with a Thermalright TR-FDB-1600 fan. The ambient temperature was 22°C/72°F. The application used to monitor temperatures was HWiNFO v3.82-1300. Keep in mind that the thermal sensors in most modern processors are not really accurate at measuring idle temperatures, hence the very small delta between the room temp and the idle results.

Idle CPU + Idle IGP: The system was left to idle for 15 minutes.
Idle GPU + Load IGP: OCCT v3.1.0 GPU stress test was run at 1680x1050 for 15 minutes.
Load CPU + Idle IGP: Prime 95 In-place large FFTs was run for 15 minutes.
Load CPU + Load IGP: Prime 95 In-place large FFTs and OCCT v3.1.0 GPU stress test were run for 15 minutes.

AMD are known for making cool-running processors, and the A-series APUs appear to be no different.
Our A6-3650 sample was downright chilly when idle, and barely warmed up when we were fully loading the integrated GPU. Even when both CPU and GPU portions where loaded, the temperature just barely broke the 40°C mark.

 

[MAC]

Overclocking Results

Overclocking Results

Although this is still a brand new platform, we have gotten more accustomed to it over the last month. First and foremost, the analog VGA display output will start failing very quickly once you start increasing the APU bus, so definitely stick DVI if you plan on doing any serious overclocking. The native USB ports might also drop out, but thankfully if your motherboard has any third-party USB 3.0 controller, those ports should still work. With regard to voltage, at least with our ASUS F1A75-M PRO, we simply had to set the CPU voltage to 1.50V with Load-Line Calibration set to Medium, VDDNB at 1.30V, APU1.2V at 1.30V, and DRAM voltage to 1.66V (or whatever your modules need). Disabling all the CPU power-saving features and onboard devices is also advised.

Highest Stable CPU Overclock

Click on image to enlarge​

As you can see, we were able to overclock our A6-3650 to almost 3.6Ghz. This might not seem that impressive when compared to past and present AM3 chips, but it is an almost 1Ghz overclock. Since this sample has a default voltage of 1.416V, increasing it to 1.50V is really not a big deal despite the potentially more fragile 32nm manufacturing process. The full load temperatures were in the mid-50C range with our Thermalright Ultra-120 Extreme and two 120mm 1600RPM 63.7CFM fans.

Highest Stable APU Bus Overclock

Click on image to enlarge​

We continue to have almost unnatural success when it comes to overclocking the APU bus. While our A8-3850 was able to support a 156Mhz APU bus, we were able to reach 162Mhz on this A6-3650. While 162MHz was fully stable, 163Mhz would not even boot so clearly we were on the ragged edge. At this bus speed, the integrated Radeon HD 6530D GPU core should be running at 718Mhz, a humougous 62% overclock over the stock 443MHz. Our 3DMark scores do reveal a significant improvement in overall performance, so all the APU's components are indeed being overclocked.

Highest Stable Memory Overclock

Click on image to enlarge​

As we mentioned in our Llano launch article, the Lynx platform is capable of some very high memory frequencies. While it natively supports DDR3-1866, which is the highest among desktop platforms, it also has considerable overclocking headroom thanks to the APU's new and improved memory controller. Using a Corsair Dominator GT CMT4GX3M2B2133C9 2x2GB kit, which is rated for DDR3-2133 9-10-9-27 at 1.50V, we were able to able to ramp up the memory speed up to DDR3-2338 9-12-9-27 at 1.66V. What makes this most impressive is that we are still using the IGP and we could run 3D apps at this speed. If we disregarded 3D stability, DDR3-2400 was easily achievable, SuperPI and Prime 95 stable. With an even better memory kit, DDR3-2500 and even DDR3-2600 should be fairly easy barriers to break.

 

[MAC]

Conclusion

Conclusion

Although the $135 A8-3850 is cheap enough in our opinion, and a bargain to boot, AMD will soon be offering a multitude of lower priced APUs as well. At the moment though, the $120 A6-3650 is the only one of the lower-end models that is available on the market. This $15 price reduction comes at the expense of a 300MHz CPU clock decrease and a GPU that is about one-quarter less powerful than the one in the flagship APU - at least on paper. Thankfully, as you will see in the table below, the GPU's performance has generally not declined too much. More importantly though, how does this new processor fare against the competition from Intel?

Let's take a look at these tables:

Now to be fair, the A6-3650 is really meant to compete with the $125 Core i3-2100, but the above situation would not really change much if that processor was included. Intel's mainstream-level dual-core/four-thread Core i3-2100 series simply crushes the native quad-core APUs when it comes to pure CPU performance. As we mentioned in the Llano launch article, the "Stars" K10.5 cores simply cannot compete with the newer Sandy Bridge microarchitecture when it comes to lightly threaded applications. The A6-3650 does occasionally score the odd victory in highly multi-treaded workloads, but even those victories are by a relatively slight margin. To make matters worse, in all of our gaming benchmarks it was a one-sided victory in favour of the Sandy Bridge processor, but only when using a discrete GPU.

Having been very impressed with the Radeon HD 6550D in the A8-3850, we were anxious to see how well the slightly neutered Radeon HD 6350D IGP would run our gaming benchmarks. Surprisingly, despite the 25% reduction in shader cores and 35% decrease in the GPU clock, overall gaming performance was not impacted too greatly. Frame rates were obviously down across the board but they were still significantly higher than what the Core i3-2120 and its HD Graphics 2000 were able to achieve. There were a few cases where the Core i5-2500K and Core i7-2600K posted better frame rates than the A6-3650, but this was arguably less due to their higher-end HD Graphics 3000 and more due to their overwhelming CPU performance advantage. Nevertheless, it would be interesting to compare the A6-3650 to the $140 Core i3-2105, which is the only 'regular' Sandy Brdige desktop chip to feature the HD Graphics 3000 IGP.

Ultimately, our conclusion from the A8-3850 applies here: If you're building a new budget-friendly system from scratch that will be used for medium-level gaming, HD movie watching or a long list of other seemingly benign tasks, this new Lynx platform is a no-brainer. The A6-3650 is a terrific all-in-one chip that provides unmatched value when you consider that the integrated GPU is a whole heck of a lot faster than a $45-50 Radeon HD 6450. Intel's lower-end Sandy Bridge models are absolutely faster in CPU-centric benchmarks, but when you take an overall look it is hard not to come to the conclusion that Llano's graphics capabilities more than makes up for its processing shortcomings. In other words, whereas most users likely will not notice that the A6-3650 is slightly slower than another processor in 2D applications, they will notice that it has much better framerates, decoding abilities and GPU computing performance. To us, that equates to a better overall computing experience.

AMD A6-3650 Llano APU Review Comment Thread​