When they first hit the scene this summer, we thought AMD’s new A-series APUs—more commonly referred to in certain circles by their code-name, Llano—were a nice fit for laptops. Llano silicon combines four relatively low-speed CPU cores with an integrated Radeon graphics processor, and AMD focused lots of attention on making sure the chip only sips power at idle, to prolong battery life. Intel’s competing Core i3 processors have two fast cores, and Llano’s quad cores give it nearly comparable CPU performance, while its Radeon IGP runs circles around Intel’s rather anemic graphics. In all, the mobile A-series APUs are an attractive alternative to the latest from the Intel juggernaut—no minor achievement these days.
When AMD attempted to migrate Llano’s competitive formula onto the desktop, though, the road got bumpy. The CPU cores needed a big clock frequency boost in order to compete with Intel’s desktop Core i3 processors, and to get there, AMD had to raise the chip’s operating voltage substantially. As a result, the first desktop Llano, the A8-3850 APU, had an outsized power envelope of 100W—5W higher than the Phenom II X4 840 it ostensibly replaced (which was made on an old 45-nm fabrication process) and worlds apart from the competition, the 65W Core i3-2100. Worse, the A8-3850’s four CPU cores were still notably slower than the Core i3’s dual cores, especially in single-threaded applications.
Llano retained its advantage in graphics horsepower over the Intel IGP, but we had trouble seeing the value proposition of a 100W chip whose primary attraction was somewhat nicer integrated graphics than the 65W competition—especially when a cheap video card would provide better graphics than any IGP. In short, Llano wasn’t terribly attractive when it moved too far from its original mission as a low-power solution for highly integrated systems like laptops. Fortunately, AMD also promised to release 65W versions of Llano for the desktop eventually, and we looked forward to those as a potentially smarter choice with more of a natural fit in some parts of the market.
Although they are currently in short supply due to manufacturing problems, the 65W desktop variants of Llano have intermittently popped up in stock at various online retailers in recent weeks. They are out there, even if they’re a little scarce right now—and fortunately, we’ve gotten our hands on one of ’em, the A8-3800 APU.
The A8-3800 differs from the A8-3850 we’ve already reviewed in just a few respects, the most obvious being its 65W thermal design power (TDP) rating. Also unlike the A8-3850, the A8-3800 makes use of AMD’s Turbo Core dynamic clock frequency tech, which allows the chip to range up to higher clock speeds temporarily when there’s thermal headroom available—that is, when not all of the CPU cores are heavily burdened at once. In this case, the A8-3800 can stray from its 2.4GHz base clock up to 2.7GHz. Those frequencies leave the A8-3800 a bit behind the A8-3850, whose four cores regularly run at 2.9GHz. Happily, though, AMD didn’t have to compromise on the 3800’s graphics in order to fit into the smaller power envelope. The two products share the same Radeon HD 6550D IGP with 400 shader ALUs at 600MHz.
The A8-3800’s list price is $129, ten bucks less than the A8-3850’s. That means its primary competition from Intel is similar: the Core i3-2100 at $117, or perhaps more appropriately, the Core i3-2105 at $134. The difference between those two Core i3 models is simple: the i3-2105 has full-on Intel HD 3000 graphics, not the chopped-in-half HD 2000 variant in the i3-2100. Their CPU performance should be the same. Thus, we’ve tested both chips, but we’ve confined the Core i3-2105 to our integrated graphics tests alone.
As if the competiton weren’t formidable enough, Intel has recently released a few new models, some at overlapping prices. For instance, the Core i3-2125 lists for $134, runs at 3.3GHz (200MHz faster than the i3-2105), and also has an HD 3000 IGP. The Core i3-2130 runs at 3.4GHz and sells for $138, but has HD 2000 graphics. We haven’t tested them yet, but the slight CPU clock speed boosts should make these newer models even tougher competition for the A8-3800. Still, they’re not likely to alter any key dynamics fundamentally, as you’ll understand once we get into the performance results.
Oh, and the results on the following pages were obtained with the same configurations detailed on this page of our A8-3850 review. We’ve just dropped the A8-3800 into our Llano test rig and added it to the mix.
Power consumption and efficiency
Since the A8-3800’s primary distinction is its 65W power envelope, we’ll begin with a look at power consumption. There’s a lot going on in the graphs below, I’ll admit. Most of the systems are configured as similarly as possible, with the same power supply, storage, memory, and graphics card. Only the motherboards and processors vary.
However, for Llano and friends, we also wanted to take a look at power draw with only integrated graphics, so the results parenthetically marked “IGP” don’t include a separate video card.
Also, a couple of the results, marked “Brick PSU”, involve truly low-power configurations with only integrated graphics and a more efficient laptop-style brick power supply. We pondered putting the Llano and Core i3-2100 systems on the brick PSU, as well, but their total power draw probably wouldn’t mix well with that power supply’s 80W peak rating, so we stuck with our standard PSU instead. Out of necessity and because it’s only fair, we did install the Core i3-2100 in a smaller microATX motherboard, the Intel DH67BL, for the IGP power tests.
We’ll start by looking at the raw power draw results, and then we’ll parse them in various ways.
The A8-3800 system’s power draw at idle is comparable to the Core i3-2100’s, overall. However, at peak, the Core i3-2100 rig draws substantially less power, even though the two chips ostensibly share the same max power (TDP) rating. The gap is largest without a discrete graphics card in the mix, where the A8-3800 setup pulls over 20W more than the Core i3-2100. That gap shrinks when we involve a discrete GeForce and shift the Core i3-2100 to a larger motherboard.
Still, the contrast between the A8-3850 and the A8-3800 is considerable. Our A8-3800-based system requires about 30 fewer watts, either with our without a discrete GPU.
If we concentrate on the period when each CPU is processing the rendering task, we can get a nice measure of power-efficient performance. When we do that, the A8-3800 isn’t especially impressive—more efficient than the A8-3850, but less so than the Core i3-2100.
The developers of Civ V have cooked up a number of interesting benchmarks, two of which we’ve used here. The first one tests a late-game scenario where the map is richly populated and there’s lots happening at once. As you can see by the setting screen below, we didn’t skimp on our the image quality settings for graphics, either. Doing so wasn’t necessary to tease out clear differences between the CPUs. Civ V also runs the same tests without updating the screen, so we can eliminate any overhead or bottlenecks introduced by the video card and its driver software. We’ve reported those “no render” scores, as well,
The next test populates the screen with a large number of units and animates them all in parallel.
The A8-3800 solidly trails the Core i3-2100 in each of our Civ V tests, generally by fairly large margins. The good news, in my view, is that the A8-3800 isn’t that much slower than the A8-3850. For many purposes, the roughly 30W reduction in peak system power draw would make the move to an A8-3800 a positive tradeoff.
While we’re here, I should pause to point out an unlikely source of competition for the A8-3850 from AMD’s own stable of older processors. Notice that the Athlon II X3 455 performs very much like the A8-3800 in the tests above. The X3 455 is an older, triple core product based on a 45-nm fabrication process, and it has a higher 95W TDP. Still, the X3 455 costs only $76 right now. You’d need to buy a discrete graphics card or a motherboard with integrated graphics to use with the X3 455, but you’d have plenty of money left over to do so. The X3 may offer similar CPU performance for less—something to watch as we look through the rest of the tests.
CodeMasters has done a nice job of building benchmarks into its recent games, and F1 2010 is no exception. We went to some lengths to fiddle with the game’s multithreaded CPU support in order to get it to make the most of each CPU type. That effort eventually involved grabbing a couple of updated config files posted on the CodeMasters forum, one from the developers and another from a user, to get an optimal threading map for the Phenom II X6. What you see below should be the best possible performance out of each processor.
Metro 2033 also offers a nicely scriptable benchmark, and we took advantage by testing at four different combinations of resolution and visual quality.
The Core i3-2100 is nearly 50% faster than the A8-3800 in F1 2010. Fortunately, the gap isn’t so great in Metro 2033, and the A8-3800’s 57 FPS average is respectable. Still, we continue to like the tradeoff of taking the A8-3800 over the 3850. The performance difference between the two is fairly minimal.
Battlefield: Bad Company 2
The best thing we can say for the A8-3800 is that it wull run this game competently. We’re less pleased with the Athlon II X3 455, whose 45 FPS average and 37 FPS minimum doesn’t inspire confidence.
Source engine particle simulation
Next up is a test we picked up during a visit to Valve Software, the developers of the Half-Life games. They had been working to incorporate support for multi-core processors into their Source game engine, and they cooked up some benchmarks to demonstrate the benefits of multithreading.
This test runs a particle simulation inside of the Source engine. Most games today use particle systems to create effects like smoke, steam, and fire, but the realism and interactivity of those effects are limited by the available computing horsepower. Valve’s particle system distributes the load across multiple CPU cores.
Although the Core i3-2100 has only dual cores, each of those cores can track and execute two threads via a feature Intel calls Hyper-Threading. This test is highly multithreaded, and the Core i3-2100 handles it very well, outperforming even the fastest current AMD quad-core processor, the Phenom II X4 980. The A8-3800 trails substantially.
Somewhat surprisingly, the A8-3800 becomes much more competitive as we move into productivity type applications. That’s true in part because, with the exception of SunSpider, the rest of these tests are nicely multithreaded, so the A8’s quad cores are put to good use.
The drop from the A8-3850 to the A8-3800 moves Llano back in the standings only slightly, but it’s enough that the A8-3800 falls behind the Core i3-2100 in x264, interestingly enough.
3D modeling and rendering
Let’s pause here to consider the effects of Turbo Core on the A8. Cinebench is typically a litmus test of sorts for Turbo Core, since it includes both a single-threaded and a multi-threaded component—and since 3D rendering speed tends to scale quite nicely with additional cores and threads. As you know, the A8-3800 has Turbo Core, with a 300MHz potential boost on tap, while the A8-3850 does not.
Yet look at the results here for the two A8 processors. The speedup when going from one thread to four with both of them is almost exactly 4X, and with only one thread active, the 3800 is a fair amount slower than the 3850. Now, one would have expected Turbo Core to grant the A8-3800 a bit of an advantage in the single-threaded scenario, since that one core should be running at 2.7GHz. For the same reason, the A8-3800’s speedup with four threads should have been a little less than 4X. What’s the deal?
Worried that Turbo Core wasn’t working right, I fired up a copy of CPU-Z while the Cinebench single-threaded test ran, in order to see what was happening with the CPU clock frequency. Fortunately, my fears were unfounded; the A8-3800 was indeed running at 2.7GHz some of the time. The trouble was that it spent far more time at 2.4GHz, even though only one core was busy. Remember, Turbo Core is fundamentally conservative, because it doesn’t want to violate the chip’s overall TDP. To avoid such problems, Turbo Core “dithers” between the base and peak P-states as needed. In the case of the A8-3800, that means Llano doesn’t spend much time at its 2.7GHz peak frequency—at least not with this workload. Moreover, since Turbo Core behavior is programmed at the factory, all A8-3800 chips should behave the same. Bottom line: don’t expect to see your A8-3800 spending a lot of time at 2.7GHz, even in single-threaded applications.
Integrated graphics performance
Obviously, integrated graphics performance is where Llano carves out its place in the market, assuming you care about IGP gaming performance in a desktop processor. Neither version of the Core i3, not the 2100 nor the 2105, can hold a candle to the A8-3800—and we haven’t even talked about Llano’s vast edge in image quality. Beyond that, there are few surprises here. As you can see, there’s virtually no drop-off when going from the A8-3850 to the A8-3800 in IGP gaming. Still, there’s vastly more performance to be had by pairing one of these processors with a $99 discrete graphics card like the Radeon HD 6670.
If we want to consider the A8-3800 solely as a desktop processor, we can summarize its value proposition with one of our famous scatter plots.
At $129, the A8-3800 offers only slightly higher performance than the Athlon II X3 455, which is a $76 chip. A host of other processors occupy the A8’s price range with substantially higher overall test scores, including the A8-3850, the Phenom II X4 840, and the Core i3-2100 and i3-2105. On the price-performance front, Llano’s desktop incarnations continue to seem rather unfortunate, especially for brand-new 32-nm silicon.
Among those desktop CPUs, though, only the Core i3 chips share the A8-3800’s 65W TDP rating. That’s important to note because a 65W processor can go places that a 95-100W CPU can’t—into smaller form factors with quieter cooling and more economical power supplies, for instance. Inside of such systems—some of them, at least—integrated graphics is likely to be the solution of choice. And in some cases, in such systems, the A8-3800’s vastly superior graphics may well be more highly prized than the Core i3’s superior CPU power.
All of which takes us back to where we left off with the A8-3850 several months ago, when we had trouble finding a sensible home for a 100W chip with a burly IGP. As a 65W part, the A8-3800 ought to make sense for a certain class of relatively cheap, compact computers. Those computers probably aren’t the sort that would be built by PC enthusiasts—even in a mini-ITX enclosure, we’d prefer a Core i3 and a cheap graphics card—but they exist in various forms, including the increasingly popular all-in-one systems that follow the iMac template. For fairly basic computing needs, Llano’s integration of four slower cores and AMD’s Radeon technology could end up providing a better user experience than Intel’s Sandy Bridge CPU-IGP hybrid. We suspect PC makers who adopt the A8-3800 won’t be paying anything close to AMD’s $129 list price, either.
With that said, we’d still like to see AMD lower prices on its retail boxed A-series APUs or, better yet, raise the performance bar while keeping TDPs steady once those pesky manufacturing issues are resolved. The A8 isn’t far from being more broadly appealing, but some tweaks would alleviate our doubts. Here’s hoping those happen soon, or it may fall to the next-gen Trinity APU to close the gap.