And what about GCN, what happened when Maxwell was launched? What happened to the magical HSA and the sheer number of partners in the foundation, where's the HSA hardware being deployed beyond AMD's camp? Where's the HSA software tools? Did Mantle change the competitive landscape on the GPU market?
While this may not be the best place for a defense of the company, such comments deserve a response:
The Maxwell launch was yet another leapfrog in a long list of leapfrogs. Is the older tech less competitive when newer tech comes out? Sure, less competitive, although beyond power usage, the difference is not as stark as nVidia marketing makes it. But the real story is the coming generational response, which is not really about GCN per se.
One of the innovations we expect from Carrizo is an HSA-enabled GCN GPU, which includes an ability to interrupt current tasks for a quick compute, then resume. That, along with hardware micro-task management, queueing and dispatch, enables a whole new style of high-performance coding.
HSA has various aspects which seem important, depending on individual orientation:
First there is the complex software system which compiles down to HSA intermediate language (HSA-IL), which then locally compiles on the target machine. This supports "write once, run anywhere" coding, also shipping code as HSAIL without developers having to consider the many different forms of PC hardware.
Then there is the on chip hardware interface for supporting interchangeable compute module layouts, e.g., CPU's and GPU's. This result of building the SkyBridge chips should be particularly important for the Semicustom business by using those modules.
Perhaps the most important HSA concept is "tight CPU/GPU compute," where relatively small sections of ideally parallel code can be handed from CPU to the SIMD GPU, and the GPU can hand off complex branching code to the CPU. That power is there to use or not, just like any other new instruction. One might expect forms of this programming style to appear first in the consoles, since they are under-powered and can use the help. But it also may be used in Embedded designs, which have their own custom code.
Mantle did of course change the competitive environment by pushing Microsoft to get out their DX12, which just happens to improve both CPU and GPU results in AMD machines more than others. This will be another performance increment for starter to mid-level Carrizo laptops.
But why leave out FreeSync, which further improves laptop gaming, and is currently only available on current AMD APU's like Carrizo, and on AMD mobile GPU's. While there is an nVidia software solution, one would not expect that to have the performance advantages of the AMD hardware interface.
A real company is not a TV show and does not succeed or fail in an hour, a day, or even a year. Semiconductor chip projects typically take 2 to 3 years, top to bottom, and then it is at least 2 quarters more before we see the business results. The SkyBridge advantage both for Standard Product and Semicustom aims to greatly reduce new chip development effort and time-to-market, an innovation which should be significant industry-wide.