AMD started banging the Accelerated Processing Unit (APU) drum shortly after purchasing ATI. But it wasn't until the company introduced its Jaguar architecture that we saw uptake really garner the big headlines. In rapid succession, AMD found its hybrid technology in the Xbox One, PlayStation 4, and E- and A-series Mobility APUs (code-named Kabini and Temash) for mobile devices running Windows.
If you're not familiar with the Jaguar design, it's an evolution of AMD's Bobcat architecture, which we first looked at back in 2010. The goal this time around was to improve Bobcat's modest performance by augmenting the amount of work it could do per clock cycle. AMD also gave Jaguar support for more modern instruction set extensions like SSE 4.1/4.2 and AVX. Of course, in the process, APUs equipped with Jaguar get an updated on-board graphics engine. Gone is the VLIW-based design prevalent in Radeon HD 6000-and-newer GPUs, replaced by Graphics Core Next.
The Kabini-based APUs leverage up to four Jaguar cores and as many as 128 GCN-based shaders. If you want more technical depth on that processor, have a look at AMD's Kabini: Jaguar And GCN Come Together In A 15 W APU.

Almost a year has passed since our first look at Kabini. Originally, AMD positioned the APU as mobile-focused, ideal for low-power devices wielding the flexibility of x86. More recently, the company revealed its intention to sell the chip into a desktop platform called AM1 as well, built around its FS1b processor interface.
Of course, AMD's marketing machine is aflutter, calling AM1 the world's first "system in a socket," since Intel's Bay Trail-D design isn't upgradeable. It's an embedded configuration with a soldered-down SoC. Indeed, most of the Bay Trail-equipped boards we've seen so far appear best suited to industrial applications. In contrast, AM1 motherboards seem more desktop-like, supporting up to 16 GB of DDR3 memory and accommodating active cooling.
| Athlon 5350 | Athlon 5150 | Sempron 3850 | Sempron 2650 | |
|---|---|---|---|---|
| CPU Cores | 4 | 4 | 4 | 2 |
| Clock Rate | 2.05 GHz | 1.6 GHz | 1.3 GHz | 1.45 GHz |
| GPU Frequency | 600 MHz | 600 MHz | 450 MHz | 400 MHz |
| Memory Data Rate | 1600 MT/s | 1600 MT/s | 1600 MT/s | 1333 MT/s |
| Total Cache | 2 MB | 2 MB | 2 MB | 1 MB |
| TDP | 25 W | 25 W | 25 W | 25 W |
| MSRP | $59 | $49 | $39 | $34 |
AMD is shipping AM1-capable APUs under the company's Athlon and Sempron brands. Currently, all of the available models are rated for 25 W and feature 128 shaders, although the graphics engine ranges from 400 to 600 MHz, depending on the specific implementation. Three of the four chips come with four cores and 1600 MT/s memory support, while the lowest-priced Sempron 2650 is a dual-core version with a 1333 MT/s transfer rate cap. Core frequencies range from 1.3 to 2.05 GHz, which does serve to hold back performance. Then again, Intel is shipping similar to slightly higher clock rates on its Silvermont-based CPUs (Intel Silvermont Architecture: Does This Atom Change It All?).
| Pentium J2900 | Celeron J1900 | Celeron J1800 | |
|---|---|---|---|
| CPU Cores | 4 | 4 | 2 |
| Base/Turbo Boost Clock Rate | 2.41/2.67 GHz | 2/2.42 GHz | 2.41/2.58 GHz |
| Dynamic GPU Clock Range | 688/896 MHz | 688/854 MHz | 688/792 MHz |
| Memory Data Rate | 1333 MT/s | 1333 MT/s | 1333 MT/s |
| Total Cache | 2 MB | 2 MB | 1 MB |
| TDP | 10 W | 10 W | 10 W |
At least in this segment, AMD and Intel are topping out with four cores (the Jaguar architecture can continue scaling, as it does in the Xbox and PlayStation, while Silvermont does as well in the Atom C2000 family). Intel's solution uses less power. However, the company's HD Graphics engine is also less capable. And memory support is limited to 1333 MT/s across the Bay Trail-D platform.
We'd expect AMD's socketed Kabini platform to be faster than the competition from Intel, based on all of those specifications, though Bay Trail-D may be more efficient. Let's take a closer look at the hardware for more detail.
We're testing two AM1 motherboards and a single Bay Trail-D platform.
ASRock AM1H-ITX
ASRock's AM1H-ITX is, as its name suggests, a mini-ITX board. Measuring just under 7" square, it doesn't occupy much space. Yet it offers an impressive number of interesting features. For instance, the platform comes equipped with an ATX power connector, but it can by driven by a 19 V DC adapter (and it has the corresponding input to prove it).
You get a lot of connectivity, given the motherboard's footprint. There are two USB 3.0 and two USB 2.0 ports on the I/O panel (plus a USB 3.0 header on-board), four display outputs (VGA, DVI, DisplayPort, and HDMI), GbE, and eight-channel audio with an optical output.

On the board itself, you'll find four SATA connectors, a SATA power connector, two DDR3 memory slots, a PCIe x16 expansion slot, mini-PCIe for a wireless card, and a trio of fan headers, one of which is for the APU. That's a comprehensive suite of features, to be sure.
Gigabyte AM1M-S2H

Measuring roughly 6.5" x 9", Gigabyte's AM1M-S2H is a slightly larger microATX motherboard. The extra length creates room for two PCIe x1 slots, in addition to a PCI Express x16 slot. Gigabyte's I/O panel features VGA and HDMI video outputs, two USB 3.0 and two USB 2.0 connectors, GbE, and three analog audio outputs.

The board also hosts two SATA connectors, a pair of DDR3 memory slots, two fan headers, and three more USB 2.0 headers for additional peripheral connectivity.
ASRock Q1900B-ITX
Our Bay Trail-D representation also comes complements of ASRock in the Q1900B-ITX, armed with an integrated Celeron J1900 CPU. The rest of the board's attributes reflect a similar approach: it employs SO-DIMM slots and a single PCI Express x1 interface, with no support for full-sized graphics cards.
Moreover, the back panel hosts legacy parallel and serial ports, exposing the platform's embedded focus. You do get one USB 3.0 port, three USB 2.0 connectors, a VGA output, and HDMI connectivity. GbE and six-channel audio output are standard fare.

ASRock enables two fan headers, a USB 2.0 header, and a pair of SATA ports on the little board's surface. Passive cooling is made possible by the Celeron, which is another reason Intel's approach works so well in industrial environments.
We're comparing AMD's new Athlon 5350 to Intel's Celeron J1900. It's a great match-up because it gives us our first opportunity to test Intel's Bay Trail-D platform against AM1 using closely-matched clock rates (the Celeron is running at a 2 GHz base frequency that goes as high as 2.41 GHz, while the Athlon operates at 2.05 GHz).
While Intel's most powerful Bay Trail-D-compatible option is the Pentium J2900, the only difference between that chip and our Celeron J1900 is clock rate. The Pentium's base frequency is 410 MHz higher, while its peak Turbo Boost state allows for 250 MHz more. Additionally, the HD Graphics engine is up to 42 MHz faster on the Pentium.
| AM1 System | BGA 1170 System | |||
|---|---|---|---|---|
| Motherboard | Gigabyte AM1M-S2H, AM1 | ASRock Q1900B-ITX, BGA 1170 | ||
| Processor | AMD Athlon 5350, quad-core, 2.02 GHz | Intel Celeron J1900, quad-core, 2.0/2.42 GHz | ||
| Memory | DDR3-1666, 9-9-9-24 | DDR3-1333, 9-9-9-24 | ||
| Graphics | Radeon R3 Graphics, 600 MHz, AMD Catalyst 14.8 beta 8 | Intel HD Graphics, 688-854 MHz clock range, Graphics Driver 10.18.10.3366 | ||
| System Drive | Western Digital Black, 500 GB | |||
| Power Supply | XFX PRO850W, 850 W, 80 PLUS-certified | |||
And here are the benchmark details:
| Benchmark Configuration | |
|---|---|
| 3D Games | |
| Dota 2 | 60-Second Fraps, botmatch |
| Grid 2 | Direct X 11, Built-in Benchmark |
| Audio/Video Encoding | |
| HandBrake CLI | Version: 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) |
| iTunes | Version 10.4.1.10 x64: Audio CD (Terminator II SE), 53 minutes, default AAC format |
| LAME MP3 | Version 3.98.3: Audio CD "Terminator II SE", 53 min, convert WAV to MP3 audio format, Command: -b 160 --nores (160 Kb/s) |
| TotalCode Studio 2.5 | Version: 2.5.0.10677, 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 |
| Productivity | |
| ABBYY FineReader | Version 10.0.102.82: Read PDF save to Doc, Source: Political Economy (J. Broadhurst 1842) 111 Pages |
| Adobe Photoshop CC | Version 13 x64: Filter 15.7 MB TIF Image: Radial Blur, Shape Blur, Median, Polar Coordinates |
| 7-Zip | Version 9.28, LZMA2, Syntax "a -t7z -r -m0=LZMA2 -mx=5" Benchmark: THG-Workload-2012 |
| WinRAR | Version 5.1, RAR, Syntax "winrar a -r -m3" Benchmark: THG-Workload-2012 |
| WinZip | Version 17.0 Pro, Best Method, ZIPX Benchmark: THG-Workload-2012 |
| Synthetic Benchmarks and Settings | |
| 3DMark 2013 | Version: 1.0.1, Cloud Gate test |
| PCMark 8 | Version: 1.0.4, Home test, Conventional and OpenCL Accelerated |
| SiSoftware Sandra 2013 | Version: 2013 SP5c-1872, CPU Test = CPU Arithmetic, GPGPU Cryptography |
We begin with PCMark 8, which should help establish some expectations of the real-world testing.
The Home test is designed to reflect typical productivity workloads, including Web browsing, writing, gaming, photo editing, and video chat.

The Athlon 5350 beats Intel's Celeron J1900 by about 10% in the CPU-oriented version of the benchmark. But when OpenCL acceleration is enabled, AMD's new APU leads by much more. This shouldn't come as a surprise, given the Athlon's super graphics component and its compute potential.

The Athlon establishes an advantage in the Dhrystone module. Both processors are much more evenly matched in the floating-point-oriented Whetstone metric.

The Athlon 5350 fares well in the GPU-accelerated version of Sandra's cryptography benchmark, where plenty of memory bandwidth and AES-NI support are particularly useful. The Radeon shaders are also quite helpful for hashing.

3DMark's Cloud Gate benchmark gives us a synthetic peek into GPU performance. The Athlon carves out a fairly small advantage in the physics sub-test. However, its graphics score almost doubles the Celeron, which in turn affects the overall outcome.

Futuremark's Peacekeeper Web browsing benchmark affords AMD another win, again in the 10-15% range.

Fair warning here: this synthetic benchmark was provided by AMD to demonstrate the company's new GPU-powered JPEG decoding acceleration, enabled in the Catalyst 14.8 beta 8 driver used for this review. Of course, we're most interested in how this feature works in the real-world.
Our suite of media transcoding tests spans from light, single-threaded workloads to taxing threaded benchmarks. More than likely, if you build a system based on one of these low-power processors, you'll touch the less demanding applications and leave the big video projects to your main workstation.




Regardless of metric, AMD's Athlon 5350 is faster by a fairly consistent margin. Given that we're dealing with two quad-core processors, we have to look to other variables like architecture, clock rate, and memory bandwidth to explain the disparity.
Our productivity-oriented tasks also present single- and multi-threaded workloads, all of which are fairly common on an average desktop system. The benchmarks we run on this page include printing a PowerPoint presentation to PDF, optical character recognition on a scanned document, and photo editing.


Again, the Athlon 5350 bests Intel's Celeron J1900 by somewhere between 5-15%. It does appear that Silvermont trails by less in more heavily-threaded jobs, while the single-threaded measurements yield a larger win for AMD.

There are two components to this test. The red line reflects a series of threaded filters that fully utilize the host processor. Intel's Celeron takes the win there. The other leverages completely different OpenCL-accelerated filters, which AMD's on-board graphics engine tears through.
All three of these applications are immensely popular, and manipulating archives of files is particularly relevant to lightweight platforms like the ones we're testing.



AMD's AM1 platform does really well in 7-Zip and WinRAR. However, it trails the Celeron in WinZip's EZ benchmark, which applies maximum compression. With that said, AMD can claim a win in the OpenCL-accelerated component of the test.
Although you probably wouldn't consider these gaming-centric platforms, we did want to assess their ability to handle mainstream titles (especially since AMD makes such a big deal about its GCN-powered graphics component).
Keep in mind that the Athlon 5350 represents the very best you'll see out of AM1, while there is one faster offering from Intel. Of course, the AM1-based platform will take a discrete upgrade if you need more performance, while Bay Trail-D does not.


Dota 2 is probably one of the most played titles, but it's not exactly demanding. Neither is Grid 2 at its lowest detail settings. Yet, AMD's Athlon barely manages to post performance numbers close to 30 FPS. Sure, you can call those numbers playable. But they're not particularly good-looking. Intel's Celeron J1900, on the other hand, is simply too slow for gaming.
Based on our results, I'd guess that AMD's AM1 platform should be able to handle less-demanding MMOs like World of Warcraft, lightweight shooters like Left 4 Dead, and a wide range of even more casual games (such as Angry Birds).
I did fire up Battlefield 4 and Neverwinter to see if the Athlon could cope. But even at 1280x720 using the lowest possible settings, it couldn't.

At first glance, it looks like the Athlon 5350 uses significantly more power than Intel's Celeron J1900, especially when it comes to the combined CPU and GPU load. But even in that extreme case, only 13 W separate the two platforms.
To put our measurements into perspective, many high-end desktop PCs use more power at idle than these systems under the most taxing load we can contrive.
When they drop down to idle, these systems are separated by a negligible 3.5 W.

As for thermals, Intel appears to fare quite a bit worse. But bear in mind that the Celeron J1900 is using a passive heat sink, whereas the Athlon benefits from active cooling.
Intel comes away with the advantage then, as its highly-integrated motherboard/processor combo generates no noise and won't accumulate dust.
When I was asked to write about AMD's new AM1 platform, I wanted to do more than just collect benchmark numbers. I wanted to get a real-world sense of Kabini's capabilities beyond the mobile environments it was originally designed to serve. I surfed the Web, did my social networking, worked on documents, played YouTube videos, and tried to use the hardware as I would my own workstation. To my surprise, I didn't notice any difference between the Athlon 5350, Celeron J1900, or my own Core i7 in those common tasks.
Start hitting the low-end processors hard with a taxing workload, though, and the true desktop-oriented hardware pulls right away. Sure, AMD's Athlon might be a little snappier than the Celeron (an observation backed up by my data), but by a much smaller margin. In a game like Dota 2 or Grid 2, the Athlon can manage smooth frame rates at low details, while the Celeron is wholly incapable of usable performance. Get ambition and fire up Battlefield 4, though, and both low-power platforms choke.
Having said that, after using the hardware, I can comfortably say that the AM1 platform paired with an Athlon 5350 can deliver a satisfying experience in common computing and entertainment tasks. I can also say that, given a choice between Intel and AMD in this particular segment, the AM1 platform clearly wins. Intel is a bit more miserly with power, but a sub-20 W difference is largely irrelevant in the desktop space. So, congratulations AMD.
But I run into a problem when I try to imagine recommending an AM1 platform over, for instance, AMD's FM2+. Sure, a Sempron 2650 and motherboard might only cost about $60 together. But an A4-4000 and entry-level Socket FM2+-equipped motherboard combo starts in the $90 range. If you simply consider your options down the road, that extra $30 opens much larger world of options that AM1 cannot match. And frankly, the 3.2 GHz A4-4000 should clean house in a majority of our tests compared to the Athlon 5350, which is $10 more expensive than the A4.
|
|
|
|
|
|
|

Admittedly, AMD isn't targeting the traditional desktop computing segment with AM1. It's going after a new pseudo-desktop arena referred to as "PC-like devices". This is the battleground where Android-equipped set-top boxes and media players are taking pieces of the traditional desktop machine's pie. Perhaps AM1 will help builders offer a low-cost alternative to the more powerful desktops, and claw back some market share.
A PC is so much more than just a CPU and motherboard, though. The rest of the components, such as memory, hard disks, and an operating system, already make up much of a budget machine's price tag. So I'm skeptical of this platform's ability to reclaim ground for the PC.
From an enthusiast's perspective, it's hard to imagine an environment where AMD's new AM1 platform is ideal, except in cases where very low-power and diminutive enclosures are desirable. Otherwise, this could be the foundation for a cheap computer a more mainstream user with simpler needs uses to check email and browse the Web. In ultra-low-cost developing markets, it probably also makes a lot of sense. But if you have higher aspirations for an upgradable platform, look elsewhere: Socket FM2+ is a vastly superior vehicle from a performance perspective, and scales many orders of magnitude higher than AM1.

