Author: Carl Poirier
Editor: Howard Ha
Publish Date: Thursday, April 7th, 2011
Originally Published on Neoseeker (http://www.neoseeker.com)
Article Link: http://www.neoseeker.com/Articles/Hardware/Reviews/gigabyte_e350n_usb3/
Copyright Neo Era Media, Inc. - please do not redistribute or use for commercial purposes.
Its odd how long it took for AMD's Fusion APUs to be sent for review. They have been available to OEMs for a few months now, making Fusion products available since this past January. The motherboard giants made the DIY community wait; MSI, GIGABYTE, ASUS and AsRock products appeared didn't appear even on virtual shelves for weeks and even then, the products were and still are in very limited quantities.
To be fair, as in addition to being a new product, AMD's Bobcat platform is a completely new architecture. It will take a lot of work and time on the OEMs' part to learn to leverage Bobcat adequately compared to simply working with a new series of processors like AMD's AM2, AM2+, AM3 and more. In the latter case, the OEMs' development costs are greatly reduced since the motherboard architecture remains roughly the same.
GIGABYTE was one of the first to crank out a motherboard based on AMD Fusion, the E350N-USB3. Is it really well-built and optimized, or just rushed-out? Considering the trend of excellent build-quality found in Gigabyte's products, it would be a disappointment to find the E350N-USB3, our subject today, to be the opposite.
Integrated in the APU:
|Internal I/O Connectors||
|Back Panel Connectors||
Specifications are courtesy of GIGABYTE @ http://www.gigabyte.com/products/product-page.aspx?pid=3681#sp
There isn't much room to pack a lot of features onto a square with 17cm sides. Nevertheless, Gigabyte has put a full-size PCI-E onto the E350N-USB3, as well as two DDR3 slots. There doesn't seem to be any wireless communication built-in, unlike many other boards of this caliber. The cooling system for the two-chip solution consists of a single heatsink with a small fan in the middle. We'll find out soon enough it's decently silent.
The heatsink is affixed solidly with four screws, each aside a chip. The tracing density on the PCB is pretty impressive, too.
The motherboard is powered via a standard 24-pin ATX connector. In the same corner, there are the usual color coded front panel connectors, as on every other Gigabyte motherboard. A spare fan header is also present.
Despite its small size and the fact that it's not really targeted at overclockers, the E350N-USB3 features the great DualBios technology from Gigabyte. One can see the two BIOS chips in the following picture, labeled "M_BIOS" and "B_BIOS", for Main and Backup. The DB_PORT on the right, as intriguing as it is, is not documented. At the top, there is another 3-pin fan header, occupied by the heatsink fan. Wait, what? A 3-pin fan header? Unless Gigabyte has implemented voltage control for this fan, that means the fan's speed cannot be controlled dynamically. For a board targeting the HTPC market, it's rather disappointing, as the fan noise levels cannot be reduced.
On the other side, there is a 4-pin auxiliary ATX power connector. The CMOS battery as well as its reset header are right beside it, followed by the NEC USB 3.0 controller.
Further down are two internal USB 2.0 headers, as well as the HD Audio front panel connector. Drive connectivity is handled by four SATA III ports, provided by the Fusion Controller Hub. We also see that the PCI-E actually has only four lanes wired out of sixteen. Following the tracing at the back of the board, they come from the Zacate APU.
At the back, one will find 6 USB ports, of which two are USB 3.0. The integrated HD 6310 provides VGA, DVI-D and HDMI connectors. The Realtek ALC892 provides an optical S/PDIF and six analog jacks. Finally, there is also a legacy PS/2 connector, as well as the usual Gigabit LAN.
The E350N-USB3 is supplied a user's manual, a multilingual installation guidebook, a driver CD for Windows XP, Vista and 7, two SATA III cables, and the I/O shield.
Removing the heatsink reveals the new Fusion platform from AMD. This two-chip solution, a first for AMD, is composed of the Zacate APU, on the left, and the Fusion Controller Hub, on the right. The former contains the x86 cores, the GPU, the memory controller and many other things. The latter is all about I/O; it provides everything from USB ports to four extra PCI-E lanes made to attach extra controllers such as one for the NEC USB 3.0. This new update to the USB protocol isn't managed by default by the Hudson M-1, but rumors say that some later versions such as the D-1, for desktop, would be including it. As for the size, a penny was placed between the two dies for comparison. Unfortunately, no American penny could be found, so here goes one that's a little more valuable instead.
Here is a simple architectural view of the platform, including the different interfaces and the most important components.
The E-350 is fabricated in Taiwan, at TSMC, in 40nm. That's where the Radeon GPUs are made.
The floor view of this 75mm² die reveals that a lot of importance was given to DirectX 11 graphics thanks to the amount of real estate it occupies, to a point where the x86 processing elements are in a way overshadowed. Will this focus bear fruit? That remains to be seen. The video engine is quite imposing, too.
Speaking of the HD 6310, its architecture diagram shows that it has two Single-Instruction Multiple-Data engines, each composed of 40 stream processors for a total of 80. In comparison, the previous IGPs from ATI have 40 stream processors, while the HD 6870 has 1120. The graphics engine has everything the latter card has, except the rasterizer and hierarchical Z aren't doubled. The ultra-threaded dispatch processor is linked to the same cache system used in the Barts GPU.Also, whereas the latter contains 32 Z-Stencil and Color ROP units, today's subject has four, with their same respective caches. The big difference is that the L2 cache and the UVD3 are not connected to a few memory controllers, but instead to the unified northbridge, which also feeds the x86 CPU. Overall, the HD 6310 is really up-to-date with the latest architectural changes in the Radeon family.
This platform is made for small and efficient devices; the two chips do not occupy much space compared to the previous platforms composed of the processor, the northbridge and southbridge. In the following picture, here is a comparison with a full-sized ATX motherboard. In the mobile world, there are many AMD Fusion devices available already, ranging from netbooks to low-cost, full-sized notebooks. The same chips shown today are contained in many laptops in the 15" range that sell for as low as $400, as well as some all-in-one PCs. There is also a version more adapted for netbooks called the C-50, which sees its x86 cores clocked at 1GHz and the GPU at 280MHz, and has a TDP of just 9W instead of 18W. Hopefully, Neoseeker will be able to offer a review of this particular incarnation of AMD's Fusion soon.
The direct competitor to AMD's Fusion platform is NVIDIA's ION, naturally. In the following tests, ION peformance will be gauged with the Sapphire Edge Mini PC, which features an Atom D510 dual-core processor.
At the $150 mark, there are also a few other options available in the mini-ITX department. The ones featuring the latest technologies such as SATA III and USB 3.0 are much less common, though. Looking at the cheapest processors, it is clear that the AMD Sempron IIs are the better option. Paired with a Zotac 880G board, this combo runs for about the same price as today's subject. Therefore, it's also going to be participating in the competition.
Finally, just to see the difference in performance between the Athlon II and Bobcat, a X2 250u is going to be thrown in the ring. This CPU is also a dual-core clocked at 1.6GHz, much like the Zacate. Its TDP is much higher though, at 25W compared to the 18W with IGP.
Take note that the two AMD comparison systems are simulated. On an ASUS M4A89GTD PRO/USB3, the corresponding cores of an Athlon II X4 645 were disabled, and the core frequency was reduced. It gives only a rough idea of the performance, as some other parameters are not the same, for example the L2 cache which is twice the amount per core in the Semprons and Athlon X2s than in the X4s.
Intel Pineview & NVIDIA ION 2
The first overclocking test to be done is running the memory at 1333MHz, which is not officially supported by AMD. Gigabyte does allow the required multiplier to be selected, so that means all other components haven't changed frequency. This will allow determining what impact the memory frequency has on the CPU and GPU, which both share the single-channel of DDR3. For that purpose, the speed of the memory is going to be written in the graphs.
When it comes to overclocking the whole system, there are many missing options. One can adjust the reference clock from 100MHz to a mere 120MHz, as well as fiddle with some voltages. The CPU and CPU-NB multiplier aren't available as on an AM3 system. Their respective voltages are available though, as well as those for the DDR3 DIMMs and Fusion Controller Hub (FCH).
Finally, for the integrated HD6310, the BIOS allows adjusting its frequency between 500MHz and 2GHz. The frame buffer size, which is the quantity of memory that's reserved for the GPU, can also be set.
Shooting for the maximum overclock on the CPU, the first step to do was to lower the memory multiplier. Then, the reference clock was increased notch by notch, until reaching the maximum 120MHz. To be stable at that frequency, the core voltage had to be increased by 0.15V. At these settings, there was no way to use the highest memory multiplier, which would have clocked the Mushkin DIMMs at 1600MHz. It looks like the memory controller of Zacate just isn't strong enough, as the maximum stable frequency found was 1493MHz, using a reference clock of 112MHz. Increasing its voltage above 1.1V didn't allow us to push it any further.
In other words, the maximum frequency of the x86 cores could not be set at the same time as the memory one, since the range in which the reference clock can be set is pretty limited, and the CPU multiplier isn't available for tweaking. So which one to choose? As illustrated in the following pages with the 1066MHz and 1333MHz memory runs, in all cases the GPU benefits from the higher bandwidth, and in some cases the CPU as well. For that reason, the 213MHz memory frequency increase setting was chosen over the 128MHz increase in core clock setting.
The final settings used were:
Finally, the GPU frequency was also ramped up in the search for the highest stable setting. As the testing drew on, it seemed more and more impressive, up to a point where it became hard to believe. At the maximum setting allowed, the GPU would still pass through benchmarks without a fuss. There are two things that seemed to confirm it really was running at an increased speed. First, it wouldn't run the benchmarks without an increase in PCIe PLL voltage, and secondly, the benchmarks indicated a higher score. The only two settings modified were:
This program includes benchmarks for most hardware. The CPU arithmetic and multi-core efficiency benchmark will be run as well as memory bandwidth and latency.
The CPU arithmetic test puts the Bobcat in front of the Atom dual-core, and right below the Sempron 145, which gets beaten after an overclock. Communication between the two cores is a bit slower than the AM3 architecture, which was already pretty slow compared to Intel offerings, including Pineview. The memory test shows that the E350N-USB3's single-channel memory controller is pretty weak, as the Atom one, even with DDR2 at 800MHz, shows a significantly higher bandwidth. The situation is not reversed while being overclocked. The latency however is greatly improved when the memory multiplier is upped, scoring much better than the Sapphire Edge. When the reference clock is at 112MHz, the latency matches that of the Athlon II.
HandBrake is an application that converts sound and video files to other formats. It makes use of all available threads so it can exploit the processor to its full potential.
POV-Ray, for Persistence of Vision Raytracer, is a 3D rendering software that has impressive photo-realistic capabilities.
In HandBrake and POV-Ray, the E350N-USB3 didn't score really well; it finished last in both, but not far from the Atom dual-core in the video encoding test. The Athlon II and Sempron did not allow anyone to get close, though. In POV-Ray, the in-order architecture didn't seem to be a disadvantage, as the Sapphire Edge scored nearly on par with the Athlon II. Finally, in both cases, overclocking the memory increased scores a bit, and a bump to the CPU frequency also allowed to the E350N-USB3 to overtake the competition.
7-Zip is a compression program, much like WinRAR. It features a built-in test, which gives a score for compression and decompression.
Cinebench 11.5 is another rendering program supporting an insane amount of threads. The image is processed by chunks, each running on a particular thread.
7-Zip seems to benefit a lot from the increased memory speed. At 1333MHz, the Bobcat beats the Atom dual-core in compression as well as the 2.7GHz Sempron in decompression. At 1.79GHz, it almost matches the Athlon II in decompression.
The Cinebench CPU test showed the Zacate APU to be almost twice as fast as the Atom in single-thread performance, and about 17% faster when competing against its four threads. Pumping the core speed yielded a significant increase, but not enough to reach the Regor and Sargas dies. In the OpenGL test, the story is quite different, as even with the memory at 1066MHz, Brazos trumps all the competition. At 1333MHz, it's almost twice as fast as the 880G chipset! For an increase from one generation to the next, this is quite satisfying. With an overclock it's even faster, and according to the numbers seen on the Internet, it even gets dangerously close to the desktop versions of Sandy-Bridge.
3DMark Vantage is the stunning sequel to 3DMark 06, which is also going to be run. Futuremark's benchmarking programs have always been at the center of every bragging match; the best way to show that one has got the greatest gaming rig is to show the highest 3DMark score. These benchmarks put systems through a series of strenuous tests, and provides the user with a score to brag about!
PCMark is similar to the 3DMark suite, except that it includes many other tests like hard drive speed, memory and processor power, so it is considered a system benchmark and not just a gaming benchmark.
The Fusion architecture gets an edge over ION 2 in PCMark Vantage, but even overclocked, the full-size desktop platform is out of reach.
In 3DMark06, the E-350 and HD6310 are ahead of the Athlon II X2 and HD4290 by a great margin when memory is clocked at 1333MHz, though it's pretty much a tie if the frequency is kept at officially supported speeds. A further overclock allows it to overtake the Sapphire Edge which otherwise would have been way above in performance. It's not the same game in 3DMark Vantage though, as even with the lower memory setting, Fusion has a large lead. What is interesting to see here is that the memory speed has a big influence on the performance of the IGP, which sees its score more than double compared to the previous top IGP from AMD. The NVIDIA ION 2, with its dedicated memory, isn't far behind. For the CPU score, Zacate is more than one third faster than the Atom dual-core.
Street Fighter IV from Capcom has a free benchmarking tool that will be used to test the HD 4200 capabilities. World in Conflict is developed by Massive Entertainment and it also has a benchmarking tool, accessible in the game settings. Both games will be tested at 1280x1024. For World in Conflict, the graphics details will be set to "Very Low", and for Street Fighter IV they will be at the lowest settings.
In these two gaming benchmarks, the HD 6310 did not seem much faster than the HD 4250 until the memory was bumped to 1333MHz. A further increase to all the system components allowed the HD 6310 to be even faster than the HD4290 with 128MB sideport memory in one of the benchmarks, and beating the Atom-based system when it was not overclocked.
The popular, free media streaming website now hosts 720p as well as 1080p content, which happens to be quite processor intensive. Fortunately, Flash offers GPU acceleration since version 10.1 to offload work on the GPU, so that's why the CPU usage is measured to gauge the hardware acceleration and processor performance.
The Fusion platform can clearly play 1080P content without any sweat, as the CPU usage is very reasonable and the video stutter-free, and in this test it's really not that big of a deal if the competition does better.
To measure power usage, a Kill A Watt P4400 power meter was used. The following numbers represent the power drain for the entire benchmarking system, not just the video cards themselves. For the 'idle' readings, the power drain from the desktop, with no applications running was measured; for the 'load' situation, the average power consumption was taken while running the OCCT power supply test, stressing both the video card and processor, for a couple of minutes.
Unfortunately, the great performance seen in the previous pages come with higher power consumption. It should be noted however that the E350N-USB3 was tested with a 3.5" desktop hard drive, which is much less eco-friendly than the 2.5" one found in the Sapphire Edge. Also, there is most probably some power optimizations done in Sapphire's all-in-one solution due to the fact that its hardware is set in stone, whereas the Fusion board is designed for expandability and compatibility. Therefore, actual power usage in a device of the same type as the Sapphire Edge or a laptop should be better.
Gigabyte's hardware implementation of AMD's first Fusion chipset in the form of the E350N-USB3 is somewhat disappointing, as there are many critical features missing, and there is not much done to add more value to the platform. Of course, there are some features such as Gigabyte's DualBios, the ON/OFF Charge, its 333 Onboard Acceleration consisting of USB 3.0, SATA III and 3x USB Power, as well as its Ultra Durable 3 technology with most importantly twice the amount of copper in the PCB. These are very neat, but most of these features plus competing equivalents from other manufacturers can be considered the norm in the enthusiast market nowadays.
These features also cannot not change the fact that the fan isn't even thermally controlled, which would allowing throttling in order to reduce noise. In most cases, the small fan on the heatsink does not make much noise, but at this size, it can become whiney easily with dust and wear. Having a 4-pin fan header or some kind of voltage control (like MSI uses) for the 3-pin system fan headers would have been great.
Furthermore, many boards targeted at the same market feature integrated WLAN. Having a mini-PCIe slot for such a card at least would have been great, but as it is if one wants to connect to his wireless home network using this board, he has to either use the full-size PCI-E for an expansion card, or use up one of the six USB ports for a WiFi adapter.
The E350N-USB3 performed greatly against its competitors, and it is one of the first AMD Fusion products so its performance might very well get a slight increase with future BIOSes and implementations, not to mention drivers. In fact, the Intel Atom dual-core clocked at 1.66GHz was beaten ferociously in most of the CPU tests by the out-of-order Bobcat at 1.6GHz. Clearly, the latter's x86-64 cores are much faster than Intel's, but in some other areas, the Atom stayed above; as reported by SiSoft Sandra, this includes inter-core communication as well as memory bandwidth. However, the memory latency was substantially better when today's subject was clocked at 1333MHz, a speed not officially supported by AMD. However, looking at current products in both the mobile and desktop segment, many of them feature such a memory speed by default, which also improved scores of number-crunching benchmarks and applications. Meanwhile it could not compete with the Sempron II, let alone the dual-core Athlon II clocked at the same speed. The latter, in the form of the 250u, has a higher TDP though, at 25W. However, there is another version that was released recently, clocked at 1.8GHz and featuring a 20W TDP. So for raw x86 performance in a low power envelope, dropping this one in an AM3 mini-ITX motherboard would be a better solution.
Clearly, the strength of the E-350 is its integrated Radeon graphics core. With the memory at 1066MHz, the HD 6310 is already a winner against the ION 2 in most of the gaming benchmarks, and that's before taking into account its better set of features such as DirectX 11 support, OpenCL 1.1, and the Unified Video Decoder 3. The faster memory greatly increased the performance of the integrated HD6310 across all tests, increasing the lead significantly. The most impressive scores were in 3DMark Vantage, where the GPU score was 25% higher than NVIDIA's solution, and in Cinebench OpenGL test where it was almost one third faster. Once overclocked, the integrated Radeon enjoyed complete domination. As for AMD's own previous best IGP, the 890GX paired with 128MB of sideport memory, it got beaten as well. The slower 880G was put to shame in all tests. The Flash acceleration on the E-350 wasn't as good as the competition though, but still very great nonetheless.
All this performance greatness did come with a drawback though, as the power consumption is higher than the Atom D510 and the NVIDIA ION 2.
The E-350 seemed to have good overclocking capabilities, thanks to the Gigabyte board which has the reference clock and voltage available for adjustment. Unfortunately, the former maxes out at 120MHz, which could be reached easily. The IGP could also attain the maximum clock allowed, with an increase in PCI-E PLL voltage. It's too bad the BIOS has such limits, as the E-350 would have gone even further.
Overall, the E350N-USB3 is a good motherboard having the usual features and great build quality from Gigabyte. Unfortunately, its selling price is a bit high at $150, as comparable offers from MSI, ASUS Sapphire are available for cheaper.
Please do not redistribute or use this article in whole, or in part, for commercial purposes.