AMD A8-3850 Llano Desktop Processor Review - Can AMD compete with Sandy Bridge?
Llano Graphics Engine and Product Lineup
AMD Llano Graphics Engine
We have looked at the x86 CPU core portion of the AMD A-series Llano APU now let's take a look at the GPU portion and see what architecture AMD has implemented. Based on the Redwood GPU design from the Evergreen generation (think Radeon HD 5600), the Llano APU possess a lot of GPU horsepower for an integrated solution.
Each line in the A-series will have its own iteration of GPU specifications with the A8 obviously taking the lead. With 400 stream processors (now being called Radeon Cores) built on 5 SIMD arrays of 80 each, 20 texture units, 2 ROPs and a 600 MHz clock speed, the top option is capable of 480 GFLOPS of raw computing power. (The A8 mobile Llano APU ran the GPU cores at 444 MHz, so we are seeing an increase here in the higher TDP options of the desktop market.) The A6 offering drops to 4 SIMD arrays (320 total Radeon Cores) and runs at slightly lower clock speed of 443 MHz.
If you have read any of our AMD GPU stories in the last 18 months the chances are good this diagram will be very familiar representing the "Sumo" architecture found in the Llano core. The update to the Redwood core includes the newer UVD3 engine, additional power gating support for better energy efficiency as well as being integrated to the new memory interface required for being slapped on an APU. Also interesting? This is the first GPU to be built on GlobalFoundries new 32nm process technology.
With the Radeon heritage users that pick up a system based on the AMD A-series of APUs will get many added benefits like this one: much better aniso filtering when compared to the CPU/GPU competition from Intel (on the left). If you don't know the significance of this diagram just know that the condensed, uniform appearance on the right will result in better image quality in gaming. I think this is a trend that cannot be easily overlooked - the graphics drivers from AMD have been and will continue to considerably outpace Intel's in terms of features and performance.
There was at least one interesting addition to the control panel suite this time around, a feature called AMD Steady Video. Using a heterogeneous computing model AMD's driver will have the ability to stabilize "bouncy" video that is usually associated with consumer cameras and unsteady hands. We were just getting a chance to play with this option before we had to finish up this review so expect more on this later.
AMD will definitely be pushing the benefits of the Llano A-series APU to users and OEMs both in standalone configurations as well as dual-graphics systems. More on that technology and its performance later in the review.
Obviously another big factor for AMD with Llano and the A-series of APUs is DirectX 11 support. Here you can see a very blunt demonstration of that fact as we see not only huge performance benefits with the A-series APU over the Intel Core i5 Sandy Bridge system but also that Intel's processor graphics simply can't run DX11 applications. Many of you will simply shrug this off saying that you can games with at least similar image qualities on the DX9 code path but DX11 offers much more than that including multi-threading support, tessellation support and Shader Model 5.0 support.
One area where the GPU combatants used to thrive is in the world of transcoding applications. First there were CUDA-accelerated transcoding programs, then NVIDIA and AMD supported ones and finally we are getting into OpenCL development down the same road. This slide uses software from Arcsoft to demonstrate the performance advantages seen on the AMD Fusion A-series of APUs where the red bar represents CPU-only transcoding and the green bar shows the advantages when enabling GPU acceleration. Using the DX11-class GPU on the A-series parts you get some very noticeable performance boosts as well as much lower CPU utilization during that transcoding time.
The problem of course with this argument now is that Intel has addressed the transcoding problem on Sandy Bridge with some dedicated logic on the chip, not using the shaders on the GPU. It turns out this results in a much faster transcoding application than even those accelerated by discrete or integrated GPUs from AMD or NVIDIA. The GPU-based designs used on AMD's APUs are definitely more flexible and extensible than the hardwired unit on Sandy Bridge, but we need to see some more software fall into our hands that really utilizes the GPU. Maybe the AMD Fusion Developer Summit will do just that...
Product Lineup and Positioning
With the mobile Llano launch there were seven different product SKUs announced at the time of launch to cover all kinds of different mobile markets - from the highest performing A8-3530MX running at 2.6 GHz to the lowest power option A4-3300 running dual-cores at 2.5 GHz at a 35 watt thermal envelope. For the desktop launch the lineup is a lot more simple.
Only four different processors are being released, two each into the 100 watt TDP category and two at the 65w level. Keep an eye on that specifications table though and note that the A8-3850 and the A6-3650 that fall in that 100 watt TDP do not actually include support for the AMD Turbo Core Technology and will instead run at the specified frequency the whole time. By the "whole time" though I do NOT mean that features like AMD Cool'n'Quiet are no longer functional - your processor will still idle down to 800 MHz and scale up to the 2.9 GHz level as the load increases. On the other hand, the A8-3800 and A6-3600 will run at speeds "as high as " 2.7 GHz / 2.4 GHz depending on the APU load but will at least run at 2.4 GHz / 2.1 GHz. That flexibility is what allows the processors to stay within that 65 watt thermal level for system builders.
The only other feature of note is that the A8-3800 and A8-3850 will include the faster integrated GPU configuration with 400 Radeon Cores while the A6 parts will utilize the 320 core variant. Those brandings, the Radeon HD 6550D and the HD 6530D will become more important later when we discuss how they integrated with discrete graphics to form the basis for AMD Dual Graphics technology.
What about pricing? This is where things will get very interesting for readers of PC Perspective. The A8-3850 (which is the processor we were sampled for testing) will come in at an MSRP of only $135. Yes, you read that correctly. The A6-3650 is going to run you $115. Pricing on the A8-3800 and A6-3600 weren't specified yet as they won't be available immediately but we expect them to put a very similar dent in your wallet.
From a competitive stand point that clearly defines what market AMD is going after here. The lowest priced Core i5 processor is the Core i5-2300 priced at $185 - well beyond the price of the most expensive Llano APU. That really just leaves the $125 Core i3-2100 to take the brunt of the attack from AMD on the Sandy Bridge front line. Because of that matchup you will see a WHOLE LOT of Core i3-2100 benchmarks thrown into the mix in this review.
But right now, let's take a look at the chipset (or Fusion Controller Hub) that all Llano-based systems are going to be built around and see how features match up with current Intel offerings.