|Introduction and Specifications|
A couple of weeks back, we took a look at AMD's Llano mobile platform, with a mainstream notebook from Compal, and we came away impressed. If you're unfamiliar with Llano, be sure to check out our launch coverage (available here). But to quickly recap, Llano is the codename for AMD's latest Fusion-based product, that combines a CPU and GPU on a single piece of silicon. In the mobile space, Llano offers a capable quad-core CPU, with excellent DX11-class integrated graphics, in a very low power-envelope. Those things translate into good performance and solid battery life in a notebook, two highly desirable traits of mobile system.
AMD thinks they have struck the right balance between CPU and GPU performance though, and at some very budget-friendly price points. Ultimately, consumers will decide if AMD is on target. In the meantime, we've got the first AMD Llano-based desktop APU, the A8-3850, in the lab, complete with a quartet of A75-chipset based motherboards. Pairing an A-Series APU and A-Series chipset results in what AMD is calling their "Lynx" platform. Cool codenames aside, there's lots of tech to discuss here, so take a gander at the features and specifications below and join as as we find out if Llano is as desirable on the desktop as it is in a notebook...
|Llano and Lynx Explained|
We have already stepped through many of the technical details of Llano in our coverage of the mobile platform, but we’ll touch on some of the high level details again here.
Llano, in its current form, consists of four AMD “Stars” class CPU cores fused with a “Redwood” class, Radeon HD 6000-series GPU core, with up to 400 stream processors. Although based on similar CPU cores, the L3 cache used on current Phenom II processors has been eliminated, but the L2 cache per core has been increased from 512K to 1MB. Llano-based APUs are manufactured using Global Foundries’ 32nm HKMG process and consist of roughly 1 Billion transistors. In addition to the CPU and GPU cores, a northbridge and dual-channel DDR3 memory controller reside on-die, along with AMD’s UVD video engine, 24 lanes of PCI Express Gen 2 connectivity, and various digital display interfaces.
As we’ve mentioned, the graphics core used in Llano is derived from existing DirectX 11-class Radeon GPUs, so the above block diagram should look familiar to regular HotHardware readers. In the A8-3850 APU we’ll be looking at today, 400 shader or Radeon cores are present. They’re arranged in five SIMDs, with texture units and L1 caches attached. Other A-Series APUs will be offered with 320 Radeon cores as well, which means one of the integrated GPU’s SIMDs will be disabled. The command processor, graphics engine, and dispatch processor, as well as the UVD 3 video engine and display controller configurations found in discrete Radeon GPUs are all present here. In fact, the GPU core in Llano (which itself if codenamed “Sumo”) is very similar to the “Redwood” GPU at the heart of the Radeon HD 5670. The GPU is linked to the APU's northbridge and ultimately the CPU cores through the Radeon Memory Bus and a new Fusion Compute Link, but AMD is keeping details of said link quiet for now.
We’ve got a block diagram of the A75 chipset for you here, but the A55 is very similar. The A75 offers six SATA 6GB/s ports, with RAID 0,1,10 support and FIS based switching, HD Audio, 4x1 PCI Express Gen 2 lanes, 4 USB 3.0 ports, 10 USB 2.0 ports, and 2 USB 1.1 ports. In addition there are SD and IR controllers present, PCI support (for up to three slots), and mSATA support. The A55 is similar except that its SATA ports are of the 3GB/s variety and it doesn’t offer FIS based switching or native USB 3.0 support, so 14 of its ports are USB 2.0. The chipset connects to the APU using AMD’s Unified Media Interface (UMI), which offers 2GB/s of bandwidth.
|Vital Signs and APU Details|
Desktop Llano APUs, like the A8-3850 we’ll be looking at here, will use AMD’s FM1, 40mm x 40mm, lidded packaging as opposed to the FS1, 35mm x 35mm, lidless packaging of the mobile variants.
While the APUs look very similar to current Phenom II and Athlon II processors from the top, thanks to the similar lid / integrated heat spreader, they use a completely different pin configuration and also a different socket. Desktop Llano APUs have 905 pins on their underside; socket AM3 Phenom II and Athlon II processors have 939 pins.
The AMD A8-3850 APU has a default CPU clock speed of 2.9GHz. As we’ve mentioned, it is four x86 cores each with 128 KB L1 Cache (64KB Instruction, 64KB Data) and 1MB of L2 per core—no L3 cache is present. Current Phenom II processors have similar L1 configurations, but only half the L2, plus a large 6MB L3. The larger L2 should help mitigate the loss of the L3, along with some other improvements to the cores. The 32mn “Stars” derived cores in Llano have improved schedulers and branch predictions units, along with some other low-level tweaks that result in an approximate 6% improvement in IPC performance over current-gen Phenom II processors.
|A75 Motherboards: MSI and ASUS|
For the purposes of this article, we acquired a handful of A75-based motherboards to give you all an idea as to what type of boards would be hitting the scene when AMD A-Series APUs are first made available.
First up, we have the MSI A75MA-G55. Like the other A75-based motherboards featured here, all of the chipset’s features are available on the A75MA-G55, but MSI works a bit of their own magic as well. The MSI A75MA-G55 features a mouse-friendly EFI BIOS that’s much easier to navigate than traditional text-only based BIOS menus, although it does look rather plain compared to some of the other boards (Asus in particular). The board is also in MSI’s “Military Class II” family and features super ferrite chokes, highly conductive polymer capacitors and / or solid capacitors throughout, which should offer increases stability and longevity.
Next, we have the Asus F1A75-V PRO. The Asus F1A75-V PRO is one of the more feature rich motherboards featured here. In addition to exploiting all of the features inherent to the A75 chipset, this board P8Z68-V PRO offers what Asus calls its “Dual Intelligent Processors 2”. The Dual Intelligent Processors consist of Asus’ EPU unit, which we’ve covered in the past, and the TurboV processing unit. The processors work together with the F1A75-V PRO’s digital VRM (DIGI+ VRM) and give users the ability to monitor and adjust power delivery. According to Asus, the combination of the programmable digital VRM and Dual Intelligent Processors 2 results in superior efficiency and longevity.
|A75 Motherboards: Gigabyte and ASRock|
Next up we have the Gigabyte A75-UD4H and ASRock A75 Pro 4 Socket FM1 motherboards for AMD’s A-Series desktop APUs.
Like the other boards we have shown you, the Gigabyte A75-UD4H makes use of all of the features available in AMD’s A75 chipset. In addition, this Gigabyte board offers a fairly extensive set of overclocking tools—as we’ve come to expect from the company. The board is a member of Gigabyte’s Ultra Durable class of products and sports the company’s traditional blue and white color scheme, with fairly heavy-duty heatsinks on the VRM (8+2 phase power) and chipset. Gigabyte boasts of the board’s 108db SNR integrated sound solutions as well, which pairs a Realtek 7.1 channel HD controller with a proprietary converter for increased sound quality, although we didn’t do any extensive testing to prove out that point in the limited time we had with the board.
The ASRock A75 Pro4 is a somewhat less-flashy board with a handful of nice features. Like the others, it exploits all of the features inherent to the A75 chipset (USB 3.0, SATA 6GB/s, CrossFire, Dual Graphics, etc.), but ASRock also included some useful additions like an on-board POST code reported, integrated Power and Reset switches, and an external Clean CMOS switch. For what it’s worth, this was also the board AMD shipped with the A8-3850 APU, which is a testament to its stability and performance. If AMD was willing to use the board as foundation of their review platform that has to say something about it. ASRock is also known for their value, so expect the A75 Pro4 to be very competitively priced when it hits the streets in the coming weeks.
|Test Systems and SiSoft SANDRA|
Test System Configuration Notes: When configuring our test systems for this article, we first entered their respective system BIOSes and set each board to its "Optimized" or "High performance Defaults". We then saved the settings, re-entered the BIOS and set the memory frequency to DDR3-1600 and IGP frame buffer size to 512MB (where applicable). The hard drives were then formatted, and Windows 7 Ultimate x64 was installed. When the Windows installation was complete, we updated the OS, and installed the drivers necessary for our components. Auto-Updating and Windows Defender were then disabled and we installed all of our benchmarking software, performed a disk clean-up, defragged the hard drives, and ran the tests.
G.Skill supplied is with a dual-channel, 8GB (4GB x 2) DDR3-1866 memory kit for the purposes of this review. The kit was tested for compatibility with Llano / Lynx and Sandy Bridge, so it was a perfect match for our test systems here. Please note, while this kit is capable of 1866MHz, it ran at 1600MHz by default, in dual-channel mode, which is how we ran it on all of the test systems.
We began our testing with SiSoftware's SANDRA 2011, the System ANalyzer, Diagnostic and Reporting Assistant. We ran three of the built-in subsystem tests that partially comprise the SANDRA 2011 suite with AMD's Llano APU (CPU Arithmetic, Multimedia, and Memory Bandwidth). All of the scores reported below were taken with the APU running at its default clock speeds of 2.9GHz with 8GB of DDR3-1600 RAM running in dual-channel mode on the ASRock A75 Pro4 motherboard.
The SANDRA CPU test shows the AMD A8-3850 offering up just shy of 33GFLOPS, which puts it just behind an Intel Core i3. In the Multimedia tests, the APU finished right about on-par with similarly clocked Phenoms, which is to be expected considering they use similar CPU execution cores. And in the memory bandwidth tests, the A8-3850 offered about 14.8GB/s of peak bandwidth.
|Futuremark PCMark Vantage|
|Next up, we ran our test systems through Futuremark’s total-system performance evaluation tool, PCMark Vantage. PCMark Vantage runs through a host of different usage scenarios to simulate different types of workloads including High Definition TV and movie playback and manipulation, gaming, image editing and manipulation, music compression, communications, and productivity.
Most of the sub-tests used to come up with the final scores in each category are multi-threaded as well, so the tests can exploit the additional resources offered by a multi-core CPU.
PCMark Vantage shows the AMD A8-3850 APU powered configurations mostly trailing even the Core i3 dual-core Sandy Bridge based systems in the majority of tests. Intel clearly has stronger x86 performance at this time, but in tests that are more GPU dependent, the story is significantly different.
|Futuremark PCMark 7|
|Futuremark's PCMark 7 is the latest version of the PCMark suite, recently released this spring. It has updated application performance measurements targeted for a Windows 7 environment. Here's what Futuremark says is incorporated in the base PCMark suite and the Entertainment suite, the two modules we have benchmark scores for you here.
The PCMark test is a collection of workloads that measure system performance during typical desktop usage. This is the most important test since it returns the official PCMark score for the system
The two PCMark7 modules we ran illustrate the difference between AMD's and Intel's design philosophies. As you can see, the AMD-powered rigs offer better performance in the Entertainment module--which relies more on the GPU--than they do in the overall PCMark test. The trend is reversed on the Intel-based systems due to their relatively powerful x86 cores, but lower-performing graphics cores. Ultimately though, the A8-3850 trails even the dual-core Core i3-2120 here in both tests.
|HD Media Playback and Encoding|
|On the video decode side of the equation, we viewed an assortment of HD movie trailer clips and monitored CPU utilization. With a relatively powerful chip like the AMD A8-3850, we didn't expect HD video playback to be an issue but regardless, we looked at CPU utilization playing back the 1080p Flash video clip pictured below. We also fired up an I am Legend movie trailer .mov file with Windows Media Player, an used PowerDVD to play other various formats, while taking note of thread activity in Windows Task Manager Performance Monitor.
Core i5-2500: 1080p Flash Video
Core i3-2120: 1080p Flash Video
AMD A8-3850 APU: 1080p Flash Video
All of the video clips we played back worked flawlessly and exhibited very low CPU utilization in the single-digits or low double-digits. Full-screen 1080p Flash video is notorious for high CPU utilization when using older drivers or Flash Players, but the situation looks good today. All of the platforms represented here performed very well in our HD Media playback tests, although AMD does offer some video enhancements that Intel does not.
Cyberlink's Media Show Espresso is a video conversion tool that imports various video media files types and converts them to other standard video formats for publication, distribution and / or streaming. In this test, we take a 184MB high definition 1080p AVCHD video clip and compress and convert it to a iPhone 4 H.264-encoded .MP4 file. Times are measured in minutes:seconds with lower times representing faster throughput in the video conversion process.
We ran this test both with an without hardware acceleration on all of the platforms and the differences in performance were huge. To put it simply, Intel's Quick-Sync technology rocks. With it enabled, all of the second-gen Core processors ripped though the video in 10 seconds. But even without Quick-Sync though, the Intel chips put up good numbers. Intel's dual-cores only slightly trailed AMD's quads. What's interesting to note is that the A8-3850's GPU is actually slower to encode video than the CPU cores--at least with the latest version of MediaShow Espresso.
|LAME MT Audio Encoding and Cinebench R11.5|
In our custom LAME MT MP3 encoding test, we convert a large WAV file to the MP3 format, which is a popular scenario that many end users work with on a day-to-day basis to provide portability and storage of their digital audio content. LAME is an open-source mid to high bit-rate and VBR (variable bit rate) MP3 audio encoder that is used widely around the world in a multitude of third party applications.
In this test, we created our own 223MB WAV file (a hallucinogenic-induced Grateful Dead jam) and converted it to the MP3 format using the multi-thread capable LAME MT application in single and multi-thread modes. Processing times are recorded below, listed in seconds. Shorter times equate to better performance.
Audio encoding is most certainly not a strong point for the AMD A8-3850. This test only exercises two cores, so half of the APU's resources remain idle, but even against Intel's dual-core processors, it's not pretty for AMD.
Cinebench R11.5 is an OpenGL 3D rendering performance test based on Cinema 4D from Maxon. Cinema 4D is a 3D rendering and animation tool suite used by 3D animation houses and producers like Sony Animation and many others. It's very demanding of system processor resources and is an excellent gauge of pure computational throughput. This is a multi-threaded, multi-processor aware benchmark that renders and animates 3D scenes and tracks the length of the entire process. The rate at which each test system was able to render the entire scene is represented in the graph below.
Cinebench R11.5 is somewhat analogous of AMD's design vision with Llano. Here, the A8-3850 is able to pull ahead of Intel's dual-core processors in the CPU test, but trails the Core i5 quad-core by a considerable margin. In the OpenGL test though, which taxes the integrated GPU, the A8-3850 simply smokes the Intel-based competition.
|Low-Res Gaming: Crysis and ET:QW|
For our next set of tests, we moved on to some in-game benchmarking with Crysis and Enemy Territory: Quake Wars. When testing processors with Crysis or ET:QW, we drop the resolution to 1024x768, and reduce all of the in-game graphical options to their minimum values to isolate CPU and memory performance as much as possible. However, the in-game effects, which control the level of detail for the games' physics engines and particle systems, are left at their maximum values, since these actually do place some load on the CPU rather than GPU.
The much more capable GPU integrated into the AMD A8-3850 gives is a huge advantage in even these low-res, low-quality games tests. Due to the lackluster performance of the Intel HD graphics cores integrated into the Core i3 and i5 processors tested here, they are held back considerably and the A8-3850 ultimately offers more than double the performance.
|Gaming: APU and Dual Graphics - 3DMark 06 and FarCry 2|
|FutureMark states that, "3DMark's score is an overall measure of your system’s 3D gaming capabilities, based on comprehensive real-time 3D graphics and processor tests. By comparing your score with those submitted by millions of other gamers you can see how your gaming rig performs, making it easier to choose the most effective upgrades or finding other ways to optimize your system. 3DMark06 has been downloaded more than any other 3D benchmark and the ORB database now contains over 8.5 million 3DMark06 benchmark scores from around the world."
3DMark06 tells the whole story of AMD's design philosophy with Llano. The AMD A8-3850-based systems offered strong graphics performance with relatively low CPU scores that trailed Intel's Core i3-2120 dual-core, while the Intel-based rigs had relatively good CPU scores, with much lower GPU-related scores. Also note the Core i3-2100T couldn't even finish the HDR/SM3.0 test.
FarCry 2 shows the stark contrast in GPU performance between AMD's and Intel's platforms. The A8-3850 offered playable framerates at 1600x900, while the second-gen Core processors did not. Also note that running the A8-3850 in dual-graphics mode with the Radeon HD 6670 installed in the system almost doubled the framerate.
|Gaming: APU and Dual Graphics - Metro 2033 and ET:QW|
|Next, we turned up the graphics workload a notch or two, with Metro 2033, a 3D graphics stress test if we ever saw one, albeit using more relaxed settings.
AMD's A8-3850 APU offered about 75% better performance than the Intel processors here, due to the APU's much more powerful integrated GPU. Once again, we also see excellent scaling with the A8-3850 running in dual-graphics CrossFire mode with a companion Radeon HD 6670 installed in the test system.
Once again, the AMD A8-3850 APU with its integrated Radeon HD 6650D graphics core decimates the Intel HD graphics integrated into the second gen Core processors. The A8-3850 offered roughly double the performance of Intel's HD graphics. Interestingly, though, dual-graphics mode didn't work in this game, hence the lack of performance scaling when a companion GPU was added to the system.
|Total System Power Consumption|
Before bringing this article to a close, we'll take a look at power consumption of the AMD A8-3850 and some competing platforms. Throughout all of our benchmarking and testing, we monitored how much power this new APU was consuming with a power meter, versus other test systems we used for benchmarks in the previous pages. Our goal was to give you an idea as to how much power each configuration used while idling on the desktop and while under a heavy workload Keep in mind, this is total system power consumption being measured at the outlet and not the the individual CPUs or GPUs alone.
The A8-3850's power consumption characteristics when paired with an A75-based motherboard are fairly impressive. Idle power consumption is simply minuscule--we're talking about a quad-core processor here with 400 Radeon cores (plus RAM, drives, a PSU, fans, etc.) sucking down only 51-55 watts at idle. And under load (where we taxed both the CPU and GPU with concurrent runs of Cinebench and LAME) total system power consumption hovered right around 150W, which is competitive with the Core i5-2500.
|Performance Summary and Conclusion|
Performance Summary: We have a few performance-related metrics to summarize here, namely the AMD A8-3850 APU's performance versus competing offerings form Intel and the performance of all of the A75-based motherboards we tested compared to each other. Let's get the motherboards out of the way first. To put it bluntly, all of the motherboards we tested performed similarly. Because the northbridge, GPU, and some PCIe connectivity all reside on the APU, there is very little that can be tweaked at the motherboard / chipset level to affect overall performance, hence the similar benchmarks scores throughout.
There will initially be two Llano-based desktop APUs hitting the market, the A8-3850 we showed you here and a lower-end A6-3650. The AMD A8-3850 (2.9GHz CPU, 600MHz GPU with 400 Radeon Cores, 4MB of L2 cache, 100W TDP) will be priced at $135. The AMD A6-3650 APU (2.6GHz CPU, 443MHz GPU with 320 Radeon Cores, 4MB of L2 cache, 100W TDP) will be priced at $115. Two other APUs will follow shortly thereafter...
The A8-3800 and A6-3600 are lower-clocked, lower-power versions of their counterparts, that support AMD's Turbo Core technology, so their cores can dynamically clock up or down on the fly. Pricing for these two chips wasn't available just yet, but it's not like there's very much guess work to do since only $20 separates the A8-3850 from the A6-3650. Also note that even more affordable dual-core variants are planned, but AMD's not making any official announcements just yet.