AMD Kaveri GPU Has True Shared Memory, Follows Trinity

It was just yesterday that we reported on the AMD Kaveri appearing in the company's beta driver device ID list, and now we get to write about that accelerated processing unit again.

Kaveri is the chip that will follow in the footsteps of Trinity, the APU that currently represents AMD's heterogeneous computing technology. It will be launched next year (2013) and, naturally, is expected to be superior to Trinity in every way, shape and form. Granted, superior in shape and form would translate into a smaller processor package. But we digress. The AMD Fusion Developer Summit is currently on in Bellevue, Washington. There, the Sunnyvale, California-based company is sharing info on past, present and future endeavors. Kaveri happens to be one of them. The company did not say if the APU would have its own FPU and instruction scheduler, or if a more common solution, like L3 cache, would be implemented.

The corporation did clearly say one thing though: Kaveri will be the first APU with true shared, unified address space between the CPU and GPU. The x86 cores and the GCN (Graphics Core Next) graphics processing unit (GPU) will share data between them, eliminating all data copying and wasted bandwidth. All tasks, be they computations or 3D rendering, will run better. This is the next logical step towards a true unified processor architecture, where the x86 and GPU are no longer separate entities. Speaking of which, the GPUs will be part of the Radeon HD 8000 series, which showed up in the same device ID list  we mentioned above.

Looking at the whole situation, we have to give AMD credit for managing to compete fine with Intel's 22nm Ivy Bridge, in power and performance, and even beat it in graphics, while lagging one manufacturing process behind (Trinity is 32nm-based). Now APUs just need quad-channel memory support and a similar, if not better, showing on the 28nm technology.

Posted in: Hardware, Preview