Sign in with
Sign up | Sign in
Better With Time? The A8-3870 And Pentium G630, One Year Later
By ,
1. Quo Vadis, Llano? A Look Back and Ahead

More than one year ago, we published AMD A8-3850 Review: Llano Rocks Entry-Level Desktops to coincide with the company's client-oriented APU introduction. It was the first design we'd seen with more emphasis on graphics hardware than x86 cores. Up until that point, Intel's efforts to serve up integrated graphics took a back seat to its processor cores. Back then, editor-in-chief Chris Angelini concluded that, "Llano, as a package, is more balanced in a world where 3D pervades."

That wasn't meant to suggest Llano-based APUs could compete with entry-level discrete graphics. However, AMD's first desktop APU did give us a taste of what it'd be like to play a mainstream game at modest quality and resolution settings without the need for an add-in card. It was a small step forward, but a step nonetheless. The original vision for the Fusion initiative, which we covered in AMD Fusion: How It Started, Where It’s Going, And What It Means, was officially on its way.

The timing of AMD's emphasis on graphics couldn't be any better. Over the course of a few years, the company has lost ground to Intel in most comparisons of x86 performance, core for core, clock for clock. This started as far back as Intel's Core architecture. It was amplified by Sandy Bridge, which we saw in Intel’s Second-Gen Core CPUs: The Sandy Bridge Review, and further exacerbated by Ivy Bridge (Intel Core i7-3770K Review: A Small Step Up For Ivy Bridge). What does all of that mean for the end-user?

Mainstream vs. Enthusiast

It means that, depending on your workload, an Intel-powered machine could be a better choice. In other applications, as we'll see, AMD pulls into the lead. At any given entry-level price point, you're generally able to get a dual-core Intel chip or a quad-core model from AMD. In lightly-threaded apps, Intel's architecture shines. In parallelized tasks, AMD overtakes its competition with more processing resources.

But keep in mind that this sort of analysis is what you'd expect from enthusiasts, which we are. Not everyone breaks their hardware down by the workloads where it excels, or the detail settings it's able to achieve at playable frame rates. Regardless of whether you go with Intel or AMD, both entry-level processor line-ups provide more than enough performance for using office apps, browsing the Internet, and playing back the highest-definition video. Once you hit the Pentium/Celeron or A8/A6/A4 level, you're way ahead of what something like an Atom processor could muster.

The mainstream factor is what we're taking into consideration today. If you follow our Best Gaming CPUs For The Money or Best Graphics Cards For The Money columns, then you know that ~$110 is where we start getting excited about graphics and $100 is where CPUs catch our attention (though we still have recommendations all the way down in the $60 dollar range). Enthusiasts can't be bothered to care about anything under those prices. But a great many of our friends, family, and co-workers never notice if a given song takes 40 seconds to rip in iTunes or 55. They just want their computer to work so they can get on with their lives. You really can't impress those folks with neatly-organized benchmark charts. It takes something radically different to get their attention.

Software Is Most Apparent

Giving a mainstream user a PC with a CPU that's 25% faster just doesn't really register with them. But what about software able to utilize hardware-based features to totally change the way they use their favorite apps? Take Intel's Quick Sync technology as an example, slashing the time it takes to re-encode a video for use on your mobile devices. Or how about OpenCL, allowing you to apply effects to a video by harnessing graphics resources? Hardware-accelerated encryption and decryption, advanced vector extensions, and 64-bit computing are all capabilities that have taken some time for the software community to harness, but are now paying off in dramatic ways.

When we first looked at AMD's Llano architecture, we saw that it more elegantly incorporated the compute power of an Athlon II with an entry-level Radeon HD 6000-series graphics card. What we weren't able to test back then were any of the applications that'd spring up to support the Fusion initiative, utilizing x86 and graphics resources together. More than a year later, that software is starting to become widely available from big-name devs.

We decided to take two mainstream systems, one based on AMD's Llano architecture and the other on Intel's Sandy Bridge design, and draw a few comparisons. Has the performance profile of each changed in the last year using the applications in our usual benchmark suite? We look at how the apps have evolved, from thread optimization to OpenCL support to improvements attributable to better drivers.

2. Then And Now: Adobe Photoshop And WinZip

If you're interested in how some ISVs have exposed support for graphics-oriented APIs in their productivity software, check out Can OpenGL and OpenCL Overhaul Your Photo Editing Experience?, which digs into the details. Today's story is geared towards mainstream users and is primarily based on our standard benchmark suite. Over the course of the last year, a handful of our go-to tests were updated to exploit hardware able to leverage OpenGL and OpenCL support. Photoshop and WinZip are two of them.

Adobe Photoshop CS6

Adobe's Photoshop is one of the most highly-regarded image editing apps. The latest version, part of Creative Suite 6 (CS6), now supports GPU acceleration through its Mercury Graphics Engine (MGE). However, the implementation is limited to a few new filters and features. Adobe realized this through OpenGL and OpenCL, making a comparison to older versions of the software impossible. It also means that any OpenCL/OpenGL-capable hardware will give you access to acceleration.

The supported functions are indeed pretty useful, and utilizing them simply requires a driver update. This is a short list of the accelerated filters:

  • Adaptive wide angle
  • Liquify
  • Oil paint
  • Wart, puppet warp
  • Field blur, iris blur, tilt/shift
  • Lighting effects gallery
  • New 3D enhancements


Most of the features require Windows Vista or Windows 7, and at least 512 MB of graphics memory may be required, too. This isn't Adobe's first effort to accelerate certain operations within Photoshop. Check out this Adobe forum thread for more details. The thread also provides troubleshooting information.

WinZip 16.5

An overhaul of Corel's most recent version finally takes better advantage of threading, whereas old builds were fairly poorly threaded. For the first time in a long time, it makes a lot of sense to upgrade if you're a loyal WinZip user (at least from a performance perspective).

This latest version also employs OpenCL support on certain AMD-based graphics engines. This doesn't mean CPU performance gets deemphasized, though. In AMD Radeon HD 7970 GHz Edition Review: Give Me Back That Crown!, we dug deeper into WinZip's behavior, discovering that switching on OpenCL put more stress on the CPU, but barely touched the graphics resources. We asked Corel for an explanation, and received the following:

"Our algorithm dynamically uses OpenCL acceleration on files that will benefit from it, continuing to use the CPU cores for files that are processed faster on the CPU. GPU acceleration leverages the CPU to prepare files for compression. Operating in parallel, our GPU acceleration will enable the CPU to process files much faster (as the GPU completes compression and encryption tasks that the CPU starts), and so it is normal to see an increase in both CPU and GPU utilization when OpenCL is enabled. Depending on the file set you are using for your tests you’ll see varying degrees of GPU acceleration."

Our workload is actually a mix of large and small files, though it's possible that we'd see more speed-up, as a percentage, on lower-end CPUs, rather than the Sandy Bridge-E-based chips that are already overkill for such a task. The bottom line is that enabling OpenCL has a profound impact on speed.

Similarly, the Pentium enjoys a boost. Its two cores don't scale quite as well, though. And because Corel has yet to enable OpenCL support for Intel-based platforms in WinZip, its improvement isn't as pronounced.

3. Then And Now: Musemage And vReveal

Musemage 1.9.1

This software was introduced to us by AMD as a recently-optimized example of an application able to benefit from OpenCL. We used it in the aforementioned Radeon HD 7970 GHz Edition review a couple of months back and are revisiting it today.

Interestingly, Intel's architectural advantage means that its hardware is generally faster than AMD's when we test the conversion of a JPEG into an HDR image. Enabling OpenCL support on the A8 allows it to catch the Pentium. So, in this title at least, OpenCL is being used to attain parity with Sandy Bridge. Without the benefit of hardware acceleration, Llano clearly takes longer to complete the task.

We tried the app's internal benchmark, but didn't observe gains attributable to OpenCL. Our real-world test is more meaningful anyway, we think.

MotionDSP vReveal 3.2.0

vReveal is a video enhancement application designed to stabilize shaky clips, clean excessive noise, fix color levels to make video look more natural, and create panoramas out of panning shots.

The software jumps from 8 FPS to 17 FPS with OpenCL support turned on when we update from last year's version 2.0 to the 3.2.0 build available today. Obviously, that's a significant step. However, we discovered other differences as well that cause the latest version to run slower than the old one. This applies to both systems; AMD's A8 and Intel's Pentium both run slower.

4. Other Benchmarks

Most of the other benchmarks in our current CPU suite turn back similar scores, regardless of whether they were run last year or today. While the apps on the preceding pages are updated to the latest versions to exploit OpenCL and multi-core optimizations, we also ran the other tests as well in the interest of completeness. We updated all of those apps with the latest available versions, since we'll be migrating to a new benchmarking suite anyway.

It comes as no surprise that most apps perform similarly in 2012 as they did in 2012. There are some exceptions, though. 3ds Max, for example, runs slightly faster on AMD's A8-3870 after a more mature driver. Meanwhile, we updated 7-Zip from version 9.22 to 9.28 without seeing a difference. HandBrake 0.95 was upgraded to version 0.98 for this review, and it does in fact deliver a marginal boost on both test systems.

Be careful with Photoshop. If you want to save money by purchasing the older CS 5.1 version, think again! The latest build delivers better performance before even taking OpenCL into consideration. Our older benchmark script, which uses a handful of threaded filters, gets a nice speed-up.

Rarlab, the developers of WinRAR, must have also realized the importance of optimizing for threading. The latest version, 4.2, reduces the processing time by 50% in the case of the AMD system and roughly 35% on the Intel machine, which only has two processing cores.

5. Test Setup And Benchmarks
Test Setups And Software
Socket FM1 PlatformMSI A75MA-G55, Chipset: AMD A75, BIOS: 1.6b
Socket FM1 ProcessorAMD A8-3870K (32 nm, Llano, B0), 4C/4T, 3.0 GHz, 4 x 1 MB L2 Cache, w/ AMD Radeon HD 6550D, 100 W TDP
LGA 1155 PlatformGigabyte H67MA-UD2H, Chipset: H67, BIOS: F2
LGA 1155 ProcessorIntel Pentium G630 (32 nm, Sandy Bridge, D2), 2C/2T, 2.70 GHz, 2 x 256 KB L2 Cache, w/ Intel HD Graphics, 65 W TDP
DDR3 Memory2 x 4 GB DDR3-1600, Kingston KHX1600C9D3K2/8GX
Shared Platform Components
System DriveSamsung 470 Series SSD, 256 GB, SATA 3Gb/s
Power SupplySeasonic X-760, SS-760KM Active PFC
System Software & Drivers
Operating SystemWindows 7 Ultimate x64 SP1
Drivers and Settings
AMD Chipset DriversAMD Catalyst 12.6 Suite for Windows 7
AMD Catalyst Drivers 2011AMD Catalyst 11.7 Suite for Windows 7
AMD Catalyst Drivers 2012AMD Catalyst 12.6 Suite for Windows 7
Intel INF-Update-Utility 2011Version: 9.2.0.1030 (04/21/2011)
Intel INF-Update-Utility 2012Version: 9.3.0.1019 (11/27/2011)
Intel HD Graphics 2011Version: 15.22.50.64.2509 (09/03/2011)
Intel HD Graphics 2012Version: 15.22.54.64.2622 (01/21/2012)


Benchmarks

Video Benchmarks and Settings
BenchmarkDetails
HandBrake CLIVersion: 0.98, Video: Video from Canon EOS 7D (1920x1080, 25 frames) 1 Minutes 22 Seconds, Audio: PCM-S16, 48 000 Hz, Two-Channel, to Video: AVC1 Audio: AAC (High Profile)
MainConcept Reference v2Version: 2.0.0.1555, MPEG-2 to H.264, MainConcept H.264/AVC Codec, 28 sec HDTV 1920x1080 (MPEG-2), Audio: MPEG-2 (44.1 kHz, Two-Channel, 16-Bit, 224 Kb/s), Codec: H.264 Pro, Mode: PAL 50i (25 FPS), Profile: H.264 BD HDMV
Application Benchmarks and Settings
BenchmarkDetails
Autodesk 3ds Max 2010Version: 10 x64, Rendering Space Flyby Mentalray (SPECapc_3dsmax9), Frame: 248, Resolution: 1440 x 1080
Adobe Photoshop CS 5.1 and CS 6 (64-Bit) CPUVersion: 13, Filtering a 16 MB TIF (15 000x7266), Filters: Radial Blur (Amount: 10; Method: zoom; Quality: good), Shape Blur (Radius: 46 px; custom shape: Trademark symbol), Median (Radius: 1 px), Polar Coordinates (Rectangular to Polar)
Adobe Photoshop CS 6 (64-Bit) OpenCLVersion: 13, Filtering a 16 MB TIF (15 000x7266), Filters: Field Blur (Blur 15 px), Iris Blur (Blur 15 px), Tilt-Shift (Blur 15 px), Oil Paint (Stylization 4; Clealiness 2.3; Scale 0.8; Bristle 10; Angular Direction 300; Shine 1.3), Adaptive Wide Angle (Perspective, Focal Length 15 mm), Lighting Effect (Spotlight 1)
Adobe Acrobat X ProfessionalVersion: 10.0.0 Pro, == Printing Preferenced Menu ==, Default Settings: Standard, == Adobe PDF Security - Edit Menu ==, Encrypt all documents (128-bit RC4), Open Password: 123, Permissions Password: 321
Cinebench 11.5Version 11.5 Build CB25720DEMO, CPU Test: Single- and Multi-Threaded
Microsoft PowerPoint 2010Version: 14.0.4734.1000 (32-bit), PPT to PDF, PowerPoint Document (115 Pages), Adobe PDF-Printer
FritzFritz Chess Benchmark Version 4.3.2
MusemageVersion 1.9.1.4969, Convert 5 JPG Images to HDR
vRevealVersion 2.0 and 3.2.0.13029, Playback Rinderherde with "One-Click Fix"
PCMark 7V 1.04
7-ZipVersion 9.22 and 9.28, LZMA2, Syntax "a -t7z -r -m0=LZMA2 -mx=5", Benchmark: THG-Workload-2012
WinRARVersion 4.1 and 4.2, RAR, Syntax "winrar a -r -m3", Benchmark: THG-Workload-2012
WinZipVersion 15.5 and 16.5 Pro, Best Method, ZIPX, Benchmark: THG Workload
6. Components: AMD A8-3870 And Intel Pentium G630

It’s almost impossible to build two comparison systems that all readers will agree match up. Some folks prefer comparing an equal number of cores, others favor matching clock rates, and some like to pit specific features against each other. The vendors themselves typically focus on market segments. For our purposes, we put AMD's A8-3870 (four cores at 3.0 GHz) next to Intel's Pentium G630 (two cores at 2.7 GHz), yielding systems priced similarly.

Because we've already run a ton of numbers in previous stories, we know that AMD's APU is faster in applications able to exploit its four cores and in games running on its integrated graphics engine. Intel's architecture is far more efficient, though, so in tests that tax one or two cores, the Pentium enjoys a sizable advantage at much lower power.

Had we picked a Core i3 or faster Pentium G870 processor, the Intel machine would have been more expensive without much effect on the overall performance picture (since both are still dual-core parts). AMD would still likely win in well-threaded benchmarks, while Intel would extend its lead in apps like iTunes, in addition to games driven by discrete graphics, where AMD tends to be platform-limits. You'd need to spend even more for a Core i5 to beat the A8 in a more resounding way.

AMD’s A8-3870 includes four cores operating at 3 GHz, each with 1 MB of L2 cache, and a Radeon HD 6550D graphics core code-named Sumo. It requires a Socket FM1-equipped motherboard and dissipates up to 100 W of power.

We used MSI’s A75MA-G55 mini-ITX motherboard as the foundation for our system. Its four DIMM slots support up to 32 GB of memory. Two PCI Express x16 slots, one PCIe x1 slot, and one PCI slot handle add-in connectivity, while HDMI, DVI, and VGA ports facilitate flexible display options. USB 3.0 is available for high-speed peripherals.

Intel’s Pentium G630 centers on the company's Sandy Bridge architecture, manufactured at 32 nm. It features two 2.7 GHz cores with 3 MB of shared L3 cache (AMD's A8 doesn't include any L3). Per core, it's a lot faster than the APU, but its on-board graphics engine is also completely outclassed by the Radeon.

The Intel board is a Gigabyte H67MA-UD2H. It has a similar feature set as the MSI board we're using to host the AMD processor, but it adds DisplayPort and an eight-phase voltage regulator.

7. Benchmark Results: Audio And Video

Utilizing a single thread, the Sandy Bridge-based Pentium G630 clearly delivers more performance than the A8 platform at higher clock rates.

However, Cinebench can also be tested using all available cores. The threaded test runs significantly faster on the Llano-based system, its four cores trouncing Intel's two.

This test enjoys a quantifiable benefit from AMD's latest drivers compared to what was available one year ago, demonstrating an aspect that rarely gets explored in any site's reviews. Mainly, the results we generate at launch aren't necessarily static. Particularly when it comes to new graphics architectures, software improvements oftentimes nudge performance forward in a significant way.

We see a similar result at the end of our HandBrake testing, which is also optimized for threading. Our benchmark transcodes H.264 video recorded on a Canon EOS 7D. The quad-core chip wins, regardless of its lower overall IPC throughput.

Our MainConcept workload involves transcoding MPEG-2 into H.264, and the result is similar once again. The A8's four cores dominate the Pentium's two. Unfortunately, we see no speed-up from our more modern installation on either company's hardware.

Converting music to AAC in iTunes is not a threaded task, so the application cannot use more than one core. Naturally, then, Intel takes the win.

The job takes 1:42 on AMD's A8 and 1:25 on the Pentium. Is the difference notable? Sure. But we think it's improbable that mainstream users will make a huge deal about it, just as they might not care that Intel's Pentium takes almost a minute longer to finish the MainConcept test above (the pendulum swings both ways).

We found similar results using the Lame MP3 encoder.

8. Benchmark Results: Fritz, 7-Zip, WinRAR, And Adobe Photoshop

We found no differences using 7-Zip a year ago compared to today. It was a heavily-threaded app back then and remains so now, favoring AMD's A8 over the dual-core Pentium.

Finally, truly exciting results! Back in 2011, WinRAR showed Intel with a huge advantage in our workload. Changes made to the program more recently see our two comparison machines much more evenly-matched. Regardless of the architecture under your PC's hood, though, performance is significantly better, so be sure to upgrade!

It is interesting, though, that while WinRAR is more highly-optimized than it was, Intel's two cores win out over the quad-core A8. Perhaps there's work left to do still to fully tax AMD's architecture.

Our Photoshop benchmark runs faster on today's updated systems.

The filters that we apply exploit as many cores as we throw at them, which is why AMD enjoys an advantage. The Pentium isn't far off, though, using half as many cores, a lower clock, and notably less power.

9. Benchmark Results: 3ds Max, Adobe Acrobat, Fritz, And PCMark

3ds Max runs a bit slower on AMD's updated system, but the difference is within the margin of error. Nevertheless, the A8 wins thanks to its four cores operating in parallel.

Creating a complex document in Adobe’s Acrobat X has always been faster on Intel's systems thanks to its single-threaded nature. Consistently, the most efficient architecture wins.

Fritz is a chess-based metric able to tax all of the x86 cores in your desktop. Naturally, then, AMD's quad-core A8-3870 has little trouble outperforming Intel's dual-core Pentium G630.

We also ran Futuremark’s PCMark 7, which measures performance in applications built into Windows 7, returning a synthetic score. Although Intel's advantage is quantifiable, both platforms finish this one fairly close together.

10. OpenCL Results: Photoshop CS6 And WinZip

With all of our baseline testing out of the way, we're now able to gauge the effect of OpenCL acceleration in the applications that support it.

Switching on the feature in Photoshop CS6 has a profound effect on AMD's Llano-based chip, so long as you're using one of the six filters currently accelerated.

It'll be interesting to see whether Core i3s and Pentiums based on Ivy Bridge will support OpenCL. The API is supported by HD Graphics 4000 and 2500. However, on all Sandy Bridge-based parts, it's simply emulated by the processing cores. This could remain a massive advantage for AMD as OpenCL-optimized applications continue surfacing, particularly in the mainstream space where Intel has a bad habit of cutting important features in the name of differentiation.

We found something similar with WinZip 16.5. Upgrading from version 15.5 to 16.5 already had a massive impact on AMD's performance. To be clear, Intel got a great speed-up as well.

But because 16.5 supports OpenCL on compatible AMD platforms, we can turn the feature on in Corel's settings screen and watch the A8 catapult into a commanding lead. Really, this is GPU acceleration at its best.

11. OpenCL Results: Musemage And vReveal

AMD's A8 starts by trailing Intel's Pentium. However, activating OpenCL puts the Llano-based chip on par with the G630.

We've run vReveal in some of our graphics card stories. The problem there is that we're using very high-end hardware to minimize platform bottlenecks as we benchmark high-end GPUs. As a result, vReveal never has any trouble achieving a smooth 30 FPS.

That's not the case on more mainstream platforms, though. Applying the app's effects drops frame rates into single-digit ranges as each CPU struggles under the workload. Turning on OpenCL gets AMD's SIMDs involved, more than doubling the frame rate. Seventeen frames per second still isn't considered smooth, but you're at least looking at the video output in a more real-time format before applying a final render.

12. Harnessing The Benefits Of Feature-Rich Hardware

There are a number of ways you could compare Intel's and AMD's hardware, labeling one elegant and the other brutish.

On one hand, AMD tackles x86 workloads using a quad-core APU rated at 100 W, while Intel holds its own with a dual-core 65 W part. In this context, AMD is brute-forcing performance as Intel operates efficiently. On the other, software developers are increasingly harnessing AMD's on-die graphics resources to accelerate optimized applications, while Intel is left to plod through everything in software. Increasingly, the mantle of elegance is being shifted toward AMD's approach.

Now, Intel does enable OpenCL support on its HD Graphics 4000- and 2500-equipped Ivy Bridge-based chips. However, the entry-level is where performance is needed most, and the company's lower-end parts don't include support for this yet. Moreover, certain applications still only support OpenCL on AMD's graphics hardware.

Today's story involved an APU that was available one year ago and a CPU that came around this time last year. Back then, we had older versions of many of these apps and very few examples of OpenCL-enabled software. Now, we can go back, maintain the same hardware platforms, and compare how the software side has evolved. In many cases, both AMD and Intel are enjoying big speed-ups as developers write more threaded code. But as we incorporate Open CL-based testing, AMD is clearly enjoying the most drastic improvements.

Intel: Like Clockwork

Don't misunderstand: Intel is downright surgical when it comes to refining, improving, and evolving its x86 architectures. Performance continues along on an upward slope, even as the company cuts into power consumption. For this reason, unbiased enthusiasts gravitate toward Core i5 and Core i7 processors.

Not everyone is an enthusiast, though, and Intel's story gets less compelling as you slide down its heavily differentiated product stack.

AMD: The Value Company

AMD has to battle more fiercely in that mainstream space, where 10 or 15% separating benchmark results doesn't mean much at the end of the day to someone using a PC casually. Really, any modern machine with a couple of cores is fast enough for office-oriented applications. By giving you four cores for the same price as Intel's two, AMD virtually assures victory in threaded applications, even as it's forced to concede to Intel in other less-optimized titles.

Again, though, those x86-based victories almost seem inconsequential. The real action is happening in GPU-based acceleration. That's where we just saw the potential to double performance (or halve a task's time to complete).

2011 Versus 2012

Part of today's story also revolved around updating drivers and upgrading the software we use in our benchmark suite. Typically, you expect successive versions of the same application to (ideally) improve performance, particularly if a developer puts time into utilizing processing resources or some other hardware-based feature. We saw several examples of performance boosts that affect AMD and Intel alike in our new suite, WinRAR being the best example.

What To Choose?

Intel’s price structure forces you to spend a lot more money if you want a quad-core CPU. So, our comparison comes down to AMD's four-core A8 against Intel's Pentium or Core i3 with two cores. We know from experience (and just saw again) that AMD wins almost every time if the Llano architecture's cores can all be utilized. Intel's advantage kicks in when you start testing more legacy titles, or a lot of the unoptimized benchmarks we were using back in 2011. The Pentium also has a big advantage when it comes to power consumption. Up until this point, the platform right for you (or the one you recommend to family and friends) depends mostly on what you're using it for.

But then there's the graphics variable to consider. AMD's Radeon HD 6550D engine is much faster than Intel's HD Graphics logic. In fact, the Radeon is good enough for occasional lightweight gaming. Of course, if you're any sort of enthusiast, you already know that discrete graphics are still a requisite. Neither Intel nor AMD yet offer a compelling-enough experience to give up our graphics cards in games. So, it's more probable that an APU-equipped machine is most useful for accelerating applications like Adobe Photoshop CS6, Musemage, vReveal, and the latest version of WinZip. One year ago, AMD didn't have that value proposition. In 2012, a more mature software ecosystem makes this a point to consider before making a purchase. Consequently, the APU is more attractive now than it was back when we published AMD A8-3850 Review: Llano Rocks Entry-Level Desktops.

More than anything, we're excited to see that comparing two systems after the passage of one year can yield substantial improvements just by upgrading software versions and drivers. Moving forward, as developers figure out whether OpenCL has any application to their software, we expect performance to jump up, in some cases, more than any evolution of x86 would be able to achieve on its own.