AMD’s 2012 A-Series platform has arrived, and with it, some exciting new technologies. The code name ‘Trinity’ has been assigned to this latest AMD APU offering, which features the introduction of a new x86 core architecture, called ‘Piledriver’. So, let’s have a look at this Trinity APU and see what the new technologies bring to the table.
Specifications and Features
When compared to last year’s Llano processors, AMD has implemented some noteworthy changes listed below.
New x86 architecture featuring “Piledriver” cores
- Supports up to 4 cores and support for the latest ISA instructions including FMA4/3, AVX, AES, XOP
- Branch Prediction and Cache enhancements over the previous “Bulldozer” cores
- 2MB L2 cache per dual-core module (up to a total of 4 MB)
- Max Turbo Frequencies up to 4.2 GHz
- Configurable via AMD OverDrive
New GPU Cores
- Featuring VLIW 4 architecture
- Up to 384 shaders
- Up to 800 MHz
- Up to 8xAA and 16AF support
- Controllable via AMD OverDrive
- DirectX®11 Support
AMD Turbo Core 3.0
- Adds frequency to GPU and CPU cores (bi-directional)
- Controllable via AMD OverDrive
UVD and VCE
- Video Encode and Decode Hardware to offload CPU
- AMD Picture Perfect support with HD Post Processing technologies
Support for new display technologies
- AMD Eyefinity technology for 3+1 monitor support
- Display Port 1.2 support
Along with the new 2012 A-series APU comes a platform change to FM2. Unfortunately, Trinity APUs will not be backward compatible with the FM1 platform; you will need to upgrade the motherboard along with the Trinity APU. The FM2 platform will be available in three chipset flavors: A55, A75, and A85X. The different chipset offerings target a specific market segment, ranging from entry level to performance users.
|PCI Express 2.0 Support||1×16||1×16||1×16 or 2×8 AMD Crossfire|
|SATA||6×3 Gb/s||6×6 Gb/s||8×6 Gb/s|
|USB 3.0, 2.0, 1.1 Ports||0+14+2||4+10+2||4+10+2|
Here are the AMD A-Series APU specifications as provided by AMD.
Here are a few snapshots of the A10-5800K taken by hokiealumnus before he forwarded the APU on to me. Don’t get too excited, you can only photograph a CPU/APU so many ways. Externally, most of them look the same anyway. If you feel like counting, that would be 904 pins versus Llano’s 905 pins.
The Trinity Architecture
The Trinity A10-5800K is classified as a quad core on the CPU side. There are two Piledriver modules, each with two Piledriver x86 cores and one shared FPU. On the GPU side, there are 384 VLIW4 Northern Islands Radeon cores. All of this is built on a 32 nm SOI die containing 1.303 B transistor chips (up from Llano’s 1.178 B). Because die size is of major importance, AMD was able to implement these improvements with only a small increase in size from Llano’s 228 mm² to 246 mm² for Trinity.
The Piledriver modules are strikingly similar to what we had with Bulldozer, as both have the same dual integer and shared single floating point core structure per module. Each integer core has its own scheduler and L1 Cache. The FP core shares cache with the dual cores within the same module. On the L2 Cache side of things, there is a total of 4 MB split between the two modules. One thing worth mentioning here is support for DDR3-1866 memory straight out of the box. My guess is this was done to provide higher bandwidth for the integrated graphics, especially on the A10 APUs.
Below are labeled diagrams of the Trinity APU, which will give you a better idea of how things are structured.
Turbo Core 3.0
Since the early days of Intel’s Turbo Boost, AMD has been playing catch up so to speak. Don’t get me wrong here as AMD was very quick in responding with a solution of their own called ‘Turbo Core’. While AMD’s initial Turbo Core release was completely functional, the end result did not give the same level of performance/speed gains as the Intel iteration. Turbo Core 3.0 is AMD’s latest effort to close in tighter to Intel’s Turbo Boost technology. The Turbo Core 3.0 technology used in Trinity APUs can actually move power between the CPU and GPU as a unit. There aren’t many programs or applications that put a large load on both the CPU and GPU simultaneously (except for a few game titles and perhaps video/photo editing software), so the ability to channel power where it’s most needed can be a big advantage.
As far as Turbo Core 3.0 and the APU we are reviewing today, you can expect a jump from the base clock of 3.8 GHz to 4.2 GHz on the CPU side; and the GPU side will top out at 800 MHz. The image below was taken from the Trinity Press Deck that was handed out when the Trinity Mobile APUs were released back in May of this year, but it should still give you a good idea of the performance gains Tubo Core 3.0 provides. Unfortunately, none of the documentation I received for the A10-5800K had a similar image to draw from. Make note of how the different benchmarks affect the CPU and GPU speeds independently, based on which of the two are utilized the most during the test. Pretty slick, eh?
In addition to what we have already discussed, the Trinity architecture brings a host of other features to the table. A couple of items that got my attention are the Radeon RAMDisk and Memory Profile features. The Radeon RAMDisk feature is the result of an agreement between AMD and Dataram corporation, with the end result being an AMD branded version of their RAMDisk software. Coupled with compatible memory and a motherboard that supports it, the AMD Memory Profile (AMP) feature will be a useful tool when overclocking the system memory beyond the JDEC specifications.
We could literally go on for days trying to cover every little aspect. So, before I dive in to the GPU portion of the APU, please peruse the slide show below for some of the other features Trinity has to offer.
The GPU Side/Radeon Graphics
Without question, AMD is banking heavily on the performance of the embedded graphics solutions Trinity offers. You probably noticed by the SKU information above that Trinity APUs will incorporate Radeon 7000 series (Northern Islands) graphics processors as the weapon of choice. Ranging from the HD 7480D found on the lower end A4-5300 APU, all the way up to the HD 7660D solution found on the A10-5800K, AMD claims the A10-5800K’s HD 7660D will perform better than the Intel i3 iGPU, even when combined with a low end GT630.
The HD 7660D solution found in our sample APU features 384 Radeon Cores (shaders) divided between six SIMDs and is clocked at 800 MHz. In the literature we were provided, AMD claims overclocking the GPU to speeds over 1 GHz is possible. We’ll definitely check that out later!
When compared to Llano, the GPU has had a major facelift. Llano’s GPU architecture was based on Radeon 5000 series technology, and now Trinity is based on the Northern Islands Radeon 6000 series architecture. AMD has also infused a new video encoder from the latest Radeon 7000 series GPUs (more on this below). Along with the Radeon 6000 architecture came Graphics Core Next (VLIW4). This means better per core efficiency, but perhaps more important is the tessellation performance increase over Llano’s GPU. After all these improvements were made, AMD wrapped them up in a tidy bow and gave them 7000 series names for Trinity.
If you find that you need more graphics horsepower than the onboard solution provides, the Dual Graphics feature makes it possible to add a discrete graphics card to work in conjunction with Trinity’s GPU. Fortunately, you don’t have to spend a lot of money either. AMD recommends anywhere from a HD 6450 to HD 6670 discrete video card, depending on the the APU being used.
Eyefinity has made its way to the Trinity as well. This gives you the ability to connect up to four monitors to the integrated GPU. You’ll probably have to daisy chain off of the Display Port to get the fourth monitor.
As far as the multimedia experience goes, the UVD engine has made it’s way over to Trinity as well. Actually, it’s UVD3, the latest and greatest version that is used on Radeon 6000 and 7000 series GPUs. With UVD3 you can experience support for full HD Content, and it even has 3D Blu-Ray capabilities. The UVD3 engine also includes a H.264 video encoder that is similar to Intel’s Quick Sync.
The following group of AMD marketing slides highlight some other features of the GPU side of Trinity APUs. Most of the slides are actually very informative, so please take a minute to view them. Also, take note that some of the slides are from the release of the Trinity mobile APUs back in May of this year, but the technologies are the same. Those reviewers sure got a lot more pretty pictures than we did!
Overclocking and Stability
Admittedly, it has been quite a while since I have personally toyed with any AMD based system. I do, however, remember a slightly cumbersome mathematical formula that had to be adhered to if there were to be any hopes of a successful overclock. Fortunately, that has all changed for the better. How well an AMD processor will overclock is, of course, dependent on many factors; but the process is much like what you may be used to if you’re coming from the Intel side of things. Much like overclocking an Intel Sandy Bridge or Ivy Bridge processor, you simply manipulate the core ratio (requires a K series APU), memory ratio, and set your voltages as needed. Obviously, for extreme overclocking there is much more to it than that, but the ease at which you can begin the adventure is a welcome sight.
AMD was kind enough to include a Gigabyte GA-F2A85X-UP4 FM2 motherboard in the reviewers’ kit. I hope to do a review on it over the coming weeks, so stay tuned for that! From first impressions, the motherboard seems adequate for our test setup. In fact, while we’re at it, let’s go ahead and provide a list of the system specs used for testing the A10-5800K APU.
Gigabyte GA-F2A85X-UP4 Motherboard (Review to follow in the coming weeks!)
AMD A10-5800K Trinity APU
G.Skill 2X8 DDR3-2400 MHz Gb F3-2400C10D-16GTX
OCZ Vertex4 128 GB SSD
EVGA Superclock CPU Cooler
Corsair HX1050 Power Supply
Window 7 x64 (Fresh Install)
As I normally do, I like to first test the system at stock speeds. The screen shot below shows the memory set to 2133 MHz, which is above the official 1867 MHz AMD officially lists. Even though the memory kit I am using is a 2400 MHz kit, I was unable to get the system to post at that speed. I’m not prepared to blame that on the APU however, because I only have this one motherboard to test with. With that said, it’s still nice to see the system stable using memory settings beyond advertised specifications.
At stock APU speeds, the system had no problem completing a ten minute run of AIDA64’s stress test.
In an attempt to find a stable 24/7 overclock, I dropped the memory back down to 1867 MHz, so it complied with AMD’s specifications. Unfortunately, I was only able to get up to 4.4 GHz before the CPU voltage had to be raised beyond 1.5 V. The other problem I ran into was the inability of any temperature monitoring software to correctly monitor the APU’s temperatures. Even the EasyTune6 software from Gigabyte would not give correct temperatures. I ran a stress test at 1.5 V to the CPU and it showed 40 °C as a maximum temperature; I think not! AIDA64 showed 27 °C; no way. This pattern held true for the many monitoring utilities I tried. Even AMD’s own Overdrive software was whacked out when it comes to temperature monitoring; have a look at the picture below. the screen shot was taken while sitting idle at the desktop with the CPU clocked to 4.4 GHz.
Anyway, here is another AIDA64 stability test at 4.4 GHz, which it passed.
While the overclocking was a bit disappointing, it’s hard to know where to place the blame. If I were to wager a guess, I’d say that AMD didn’t do themselves any favors by bundling this particular motherboard in my reviewers’ kit. I find it odd that every monitoring solution out there couldn’t read temps correctly. To be honest, I’m not sure the BIOS is either, or at least not in a reliable manner. You may have noticed in the above picture that CPU-Z is showing the voltage at 1.428; it was actually set to 1.4875 in BIOS.
CPU Side Benchmark Results
The first set of benchmarks are the AIDA64 CPU and FPU tests. There are stock and overclocked results below using the comparisons built into AIDA64. You should be be most interested in how this A10-5800K stacks up against the a8-3850, Phenom II 1055T, and the i5/i7 comparisons. In the stock speed CPU runs, the A10-5800K did pretty well when compared to it’s AMD brethren and held it’s own against the Intel competition. In its overclocked state, naturally, the results got even better with consistent performance gains over the Llano A8-3850, no doubt due the increased clock speed.
The Floating Point test results were not so kind to the A10-5800K, it lost out to the Llano A8-3850 in all the tests, except for the VP8 run. This held true across both the stock and overclocked runs. We aren’t talking a major difference, but a difference none the less. The shared FPU architecture will adversely effect FPU testing every time as the results below show.
|Stock Speed AIDA64 CPU Tests|
|Stock Speed AIDA64 FPU Tests|
|4.4 GHz AIDA64 CPU Tests|
|4.4 GHz AIDA64 FPU Tests|
SuperPI uses an older set of x86 floating point instructions as its basis. We already know from the above AIDA FPU testing that slower times here will be no surprise. In fact, AMD gave up a long time ago trying to optimize their processors for what they call “old architecture”. What is surprising is the 32M results, which show the A8-3850 actually beating the Trinity APU when both are at their stock speed. Keep in mind several of the comparisons are much higher up the food chain than the A10-5800K, not to mention, in some cases cost hundreds of dollars more too!
WPrime shows a similar scenario to SuperPI, as far as the pecking order goes.
Moving on to some real world performance testing, we come to Cinebench R10 and R11.5. We show some good numbers when compared to the other AMD comparisons.
For some additional real world testing, we’ll move on to PoV Ray and x264, which will test rendering and video encoding. The comparisons changed a bit here, but you will get an idea of the performance level when compared to higher up CPUs. When overclocked, the A10-5800K actually was nipping at the heals of the Intel i5 2500K in the PoV ray test. The x264 test show some pretty impressive results as well.
The GPU Side – Overclocking and Benchmarks
AMD mentioned in the provided literature that they have seen the A10-5800K’s GPU overclocked over 1 GHz (from 800 MHz stock) with stock cooling. I set out to test this claim and I’m happy to report that with a small bump of the Northbridge voltage, I was able to get to 1050 MHz. Without the ability to monitor temperatures correctly, I thought it best to stop at the point where AMD’s claims were verified.
The gaming benchmarks below are using the same settings we use for all video card reviews here at Overclockers. We will be comparing the onboard HD 7660D GPU to Intel’s HD Graphics 4000. To show you how the AMD Trinity GPU scales when overclocked, those scores will also be in the graphs below. For those of you who are not yet familiar with our video card testing, please visit our recently updated video card testing procedure page. Even though the Intel and AMD comparison GPUs are integrated solutions, they still deserve the same torture we put a discrete video card through, right? I think so.
We’ll start things off with the synthetic benchmarks, which were a clean sweep for AMD. The A10-5800K beat the Intel HD Graphics 4000 iGPU handily, and in some cases, more than doubled what the Intel could produce. You can also see the integrated HD 7660D scales nicely when overclocked.
The game benchmarks showed the same domination over Intel, except for Dirt 3, which was pretty much a dead heat. When AMD showed its dominance, in most cases it was quite substantial. Good scaling was again noticed while the GPU was overclocked. A great showing by AMD all around here.
I think it’s safe to say AMD has succeeded in surpassing anything Intel has to offer on the integrated GPU front, and by quite a bit. If you took a minute to review our updated video card testing procedure, then you know that all the gaming benchmarks are run at their maximum settings. Undoubtedly, if the settings are turned down some, these games would be able to reach a playable FPS. Call me impressed.
Overclocking – The Trifecta
So, now we know the A10-5800K is capable of overclocking the CPU, GPU, and memory. Let’s take a look at a few benchmark results with the CPU at 4.4 GHz, the GPU at 1050 MHz, and the memory bumped up to 2133 MHz. As you can see by the screen shots below, we got a pretty good bump in the scores when compared to the above graphs. The A10-5800K scales quite nicely again here.
The pricing structure for the Trinity APUs can best be described as aggressive. In fact, there isn’t a Trinity processor in the lot that retails for more than $122.00. Below is a chart displaying the prices for the Trinity offerings.
To be honest, based on the graphics performance alone, I thought the pricing would be much higher. The other thing that impressed me with the pricing is there is very little difference between locked and unlocked skews. For those that only want a CPU with no integrated graphics, AMD has you covered with two Athlon X4 offerings. It’s pretty easy to see you have the ability to build a wide range of systems, and do so at an affordable price.
The message AMD wants to get across to the user is “The whole is greater than the sum of its parts”, also known as the emergent theory. What they are trying to say here is that the combined CPU and GPU technologies will bring forth a balanced computing experience, which encompasses an elevated gaming, multimedia, and per watt performance increase over previous AMD CPUs/APUs. To this end, I think they have succeeded.
For the overclockers among us, you will enjoy the ease at which this can be accomplished. While we were not able to get super high overclocks, I attribute most of that to the infancy of the platform. I’m sure as the motherboard BIOS versions mature, overclocking will improve.
The AMD A10-5800K is tweakable, has an awesome integrated GPU, is definitely priced right, and performs beyond my initial expectations. All this adds up to it being Overclockers Approved!
-Dino DeCesari (Lvcoyote)