It has been more than a decade since AMD launched its first APUs based on integrating low-end gaming GPUs within the processor, a path that Intel has also followed today. But the gaming market is not only based on enthusiasts who spend a month’s salary in the gaming market, but also on consumers who would like to have an integrated PC powerful enough to play the latest games without problems.
For that type of audience, high-performance APUs, with a power similar to that of new generation consoles such as PlayStation 5 and Xbox Series X would be ideal and it is an untapped market. So how come Intel and especially AMD don’t exploit it?
It would take a very complex intercom
If we talk about a more or less simple APU then we find that communicating its components is not difficult. But if we want to develop a much more complex APU with an integrated GPU equivalent to a mid-range in PC, a larger number of CPU cores and a higher bandwidth, then the internal communication becomes something much more complex, enough for a simple change delays a design several months from its launch.
A high-performance APU would be similar to what we see on consolesBut the consoles are only competing against each other and they are not in the middle of the PC APU product cycles where releases are almost yearly. A console is launched when the manufacturer of the same is ready to do so, instead andThe lag of a high-performance APU on PC can become on a product that does not meet expectations and accumulates as unsold stock.
When designing the organization of a future processor, the distribution of the different components must be taken into account. This means that the intercom has to be designed from scratch on the chip. The problem? Just having to reorganize the components, eliminate some and add others causes you to have to return to the design table due to the change in the organization of the wiring.
NoCs as the foundation for high-performance APUs
Right now, we are in the middle of the transition to NoCs, Intel’s recently introduced IPU is a clear case of where things are going to go in future designs. But we also have Xilinx technology recently acquired by Intel and NVIDIA DPUs based on Mellanox technology.
But what do the NoCs paint in all this? Basically NoCs by their nature completely eliminate the need for direct communication between components. Since these communicate directly with the SmartNIC that is in the central part of the chip through a central infrastructure. Because the communication is done through the SmartNIC, the organization of the different components can be totally free and the changes of this can be executed more quickly, even designers can add and remove elements without problems.
The NoC will give the possibility of seeing integrated mid-range graphics as APUs and therefore more complex configurations in that regard. The entry-level range is already being taken over by integrated GPUs, so the next step is to integrate low-end gaming GPUs and even mid-range ones if possible.