In conversation with Sasa Marinkovic we’ve got plethora of information about Trinity. AMD was right when they said that "the future is fusion"; even if this phrase sounded like just another exemplar of promotional copy, the "fusion" in question really brought something big to the table that few expected. This was the reason why we took advantage of Sasa Marinkovic visiting our hometown, so we met him in the picturesque setting of a river bank in order to chat with him about what we can expect from the Trinity platform. Sasa is currently maintaining the position of Industry Trends Marketing in AMD, after recently switching from his previous post at Product Marketing - Desktop and Fusion Software, making him one of AMD's leading experts in the field.
One could say that the APU (Accelerated Processing Unit) is arguably AMD's best product ever, for the simple reason that they still have virtually no competition in this segment. The idea is painfully simple - put the CPU and GPU cores together with the chipset onto a single piece of silicon. This gives the buyer two key system components at an unbeatable price, so that a motherboard and some RAM complete the backbone. True enough, Intel has been shipping out CPUs with integrated GPUs on them for a while now too, but their performance in the graphics department just isn't nearly up to par. AMD's integrated GPU is basically a simplified version of the chips we're seeing on discreet graphics cards, which makes it able to offer very tangible (and usable) performance. Having seen and presented Llano about a year ago, we were more than eager to come across Trinity, the next-generation APU that was first sampled at this year's CeBIT, in AMD's showcase in which the integrated GPU was neatly running DiRT 3 in Eyefinity mode (i.e. on three monitors).
The new APU generation redesigns both the CPU and the GPU, chiefly to improve speed, but also bring new capabilities. As we've already mentioned, the APU is comprised of the CPU, GPU and Northbridge. The CPU section has been thoroughly worked over, with AMD's K10.5 architecture and its Stars cores finally going to history, the new CPU being based on the Bulldozer architecture. This is the second generation of the Bulldozer technology, which is the best AMD currently has to offer. As the Piledriver cores kept AMD in the 32 nm production process, their position was hardly enviable, as they were being forced to increase chip performance without significant increases in size. The engineering team pulled this off nevertheless, increasing the physical size of the chip by only 7%, while gaining much more in terms of performance compared to Llano.
The APU units can have one or two Piledriver modules, each of which contains two cores, which gives the typical dual- and quad-core approach of the Bulldozer architecture. In practice, this means that the two integer units have their own L1 cache, with executive units as well as schedulers, but while sharing a single FPU. While all this is well-known already, other segments of the chain have been vastly improved. According to Sasa Marinkovic, execution units have been rendered significantly better, and the same goes for prefetching and branch prediction.
The new generation of Turbo Core technology also can't be left out, now available in its 3.0 version. This is a very important selling point of the new APUs, since Turbo Core is now finally able to use the APU's potential to the maximum. In Llano APUs, Turbo Core was only able to overclock the CPU section of the unit, and that only within the preset TDP values. However, the technology now has free access to the GPU as well as the CPU, and a very flexible one at that. If the CPU cores are deemed under little load, while the GPU is "sweating", the internal controller logic will boost only the GPU to the maximum TDP threshold. The same goes vice versa, and we have to admit that this is indeed a significant change.
The GPU has suffered its share of changes too, as it's now based on the VLIW4 architecture. If the name doesn't tell you much, we'll remind you that this was the architecture featured on the Cayman-based graphics cards, namely HD 6990, HD 6970 and HD 6950. Of course, the versions found in the latest APUs are simplified, without the ludicrous number of stream processors the chips are famous for, but still more than sufficient for decent gaming, even by "discreet" standards; and don't forget the potential of GPU acceleration for other uses either. Interestingly enough, Cayman has up to now been the only chip to contain the long-developed VLIW4 architecture, with Trinity coming in only second. The strongest chips installed into the latter will have six SIMD engines, each of which has 16 groups of four stream processors.
Clearly enough, this chip will only be available in the most powerful APUs, with further segmentation duly done according to price and intent. Besides the maximum 384 stream processors, then, the new chips also have up to 16 texture and eight ROP units. Specs like these are most akin to Radeon HD 6650 in the discreet gamma, although the latter is based on the VLIW5 architecture; performance remains about the same between the two. This is only a theoretical insight into what you can expect from the strongest APUs, but we should have some cold hard test-yielded numbers very soon. As for technologies implemented, since we're basically talking about a Cayman chip, the list of supported features is about the same. DirectX 11, ShaderModel 5.0 and Open GL 4.1 are all standard-fare and in accordance with the hardware specs of the chip, so it's easy to see why the GPU implemented inside the Trinity APU is so much stronger than Intel's HD Graphics.
The third-gen UVD also makes an appearance, causing offload in x86 cores when playing SD or HD video, which is fantastic for everyone who's consumption- or temperature-oriented, especially on mobile platforms, where battery life is key. There's also the inevitable Eyefinity support, enabling both mobile and desktop versions of the Trinity APU to connect to multiple monitors and create an SLS system.
Mobile versions first
As was the case a year ago, the new APU generation is presented on mobile platforms first. As far as desktop PCs are concerned, the Virgo chips, as their codename goes, will be available to us in the third quarter, roughly to coincide with the "back to school" season. Mobile APUs were rolled out first, with a rather complete line-up. The chips are codenamed Comal and cover three TDP values. The energy-efficient modules will have a TDP of 17 and 25 W, while the strongest ones will be declared to 35 W. This is far less than the versions slated for desktop use, coming in at 65 and 100 W, just like with Llano models. The branding system has remained unchanged, with the products carrying the markings A10, A8 and A6. The actual model will be stated by the four-digit number starting with 4, clearly positioning the model in its group (for instance, A8-4500M), while the "M" suffix implies the mobile version clearly enough. The first wave of mobile APUs relies on a total of five models. The number of x86 cores can be two or four, depending on the model. The clocks range between 2.1 and 2.7 GHz, in accordance with the requirements posed by the mobile segment these days. Integrated graphics cards, regardless of their similarity with the HD 6000 series, carry names taken from the current HD 7000 series. The number of stream processors and actual clocks will distinguish them clearly enough, with the former ranging between 192 and 384.
A bright future
As an endnote, we can only say that we're very hopeful of both mobile and desktop versions of the latest APUs. AMD can be given all credit for improving the APU philosophy and make sizeable steps ahead over time. We believe that this will be a major "pro" for this company in the months to come and that due attention will be given to it in the upcoming period, both in regular improvement and market positioning. Simply enough, integration is the future and the future is here. AMD is the first one to step forth, it seems, and we can but hope that others follow in their wake. We give our due thanks to Sasa Marinkovic for the effort invested to present us with the features of AMD's Trinity platform.