Devastator Graphics Core
The most intriguing part about the design of the GPU integrated into Trinity processors, codenamed Devastator, is the fact that it is based on VLIW4 architecture. Since Llano graphics core was based on VLIW5 architecture, this AMD’s decision may strike as somewhat strange, and frankly we would expect to see Trinity use CGN architecture typical of the latest graphics accelerators. However, in fact, it is VLIW4 that makes it possible to improve the specific efficiency of the graphics core, which is artificially limited by the transistor count. AMD has already resorted to this “trick” with their Radeon HD 6900 graphics cards and it did work pretty well back then.
VLIW5 arranges ALU in groups of 5 per streaming VLIW-processor. However, it turns out not quite efficient and one ALU is idling in most cases. Therefore, VLIW4 structure of the Devastator, that implies the use of four ALU in the streaming VLIW processor, allows for more optimal utilization of the resources. Of course, the back side to this picture is smaller total number of execution units and lowering of theoretical peak bandwidth of the core, but the practical specific performance per square millimeter increases. And this is the best optimization approach for a hybrid processor die that accommodates not only the graphics core, but also the computing cores.
Overall, the Trinity graphics core has six SIMD engines, each of which consists of four texturing units and sixteen streaming VLIW processors. It adds up to 384 ALU, which is 16 ALU less than in the Sumo graphics core of the Llano processors. However, simple arithmetic doesn’t quite apply here, because Devastator’s ALU are usually heavier utilized than their predecessors. Moreover, relative simplicity of the streaming VLIW processors allows setting higher clock frequencies for the graphics core. For example, while the graphics core in the top Llano processor worked at 600 MHz, the graphics core in Trinity may reach up to 800 MHz speeds.
Since Devastator has twenty four texturing units (four TMU per each SIMD engine) and eight raster operation units (ROP), we can conclude that this graphics core equals to one fourth of the Radeon HD 6970 GPU. And this is really good, even taking into account that its operational frequency is somewhat lower and there is no dedicated memory bus with high bandwidth. In other words, AMD doesn’t mislead us by saying that their Trinity processors have integrated graphics core of a “discrete” quality. We can really expect new generation of hybrid processors to demonstrate very good 3D speed.
I am sure you won’t be surprised to hear that the Trinity graphics core is compatible with DirectX 11, OpenCL and DirectCompute software interfaces. Radeon HD 6900 using the same architecture as well as Trinity’s predecessors, Llano processors, also supported them. However, new integrated graphics inherited some features of the latest solutions with CGN architecture, too. Namely, Devastator has improved tessellation unit and supports all popular antialiasing modes, such as SSAA, EQAA and MLAA.
Trinity developers paid special attention to multi-media functionality of the graphics core in their new hybrid processors. The new core has the same AMD HD Media Accelerator unit as the latest GPUs. This unit includes hardware video decoding engine (UVD3) and H.264 hardware video encoding engine (VCE). The latter is particularly important for Trinity’s success in competition against Intel’s hybrid processors that have long been featuring Quick Sync for high-speed HD video transcoding. Now AMD processors also boast something similar, but unfortunately, we couldn’t yet test the VCE engine in action, because of driver support and software compatibility issues.
When AMD was working on their new hybrid processor for the desktop market, they wanted to make sure that the users wouldn’t feel deprived of the extensive monitor connectivity, which discrete graphics cards could provide. Namely, an integrated system with a Trinity processor allows connecting up to four independent monitors at the same time and supports all connection types including analogue VGA and digital DVI, HDMI and DisplayPort 1.2. It also supports four independent audio streams. However, there are only three actual outputs available and you will need to use the DisplayPort chain to connect the fourth monitor.
However, most impressively Trinity graphics supports Eyefinity. Of course, it will be a challenge to find a game capable of running at an acceptable fps rate on three-four monitors connected to the Devastator, but the mere fact that this feature is available indicates that AMD took the features and functionality of the second generation APU very seriously and made sure it was loaded to the maximum before rolling it out into the market.