The strongest APU Kaveri ever appeared

On the first day of the annual developer conference, AMD will openly disclose the technical details of the next-generation mainstream APU Kaveri, which is unusually powerful. AMD claims that Kaveri will be the most powerful APU in history and the most developer-friendly APU. It will be officially released on January 14, 2014 (after the end of CES 2014), and will successively land on the desktop, notebook, embedded, and server sectors. The Socket FM2+ interface will be used on the desktop for the first time – the previous generations will be notebooks first, After DIY, this was finally reversed.

The strongest APU... Is the previous generation stronger than the next generation? At CES 2014, AMD will also announce more details about Kaveri APU.

12 Heterogeneous Computing Units: Kaveri will be equipped with up to 12 heterogeneous computing units, including up to 4 roller core CPU cores (two modules) and up to 8 GCN architecture GPU computing units (512 stream processors) The former has 4MB of L2 cache, and the latter is comparable to the Radeon HD 7750.

Compared to bulldozers and pile drivers, the new roller core will continue to increase computational efficiency, improve single-core execution efficiency, and enable faster “feeding” of the core, that is, front-end evolution. The GPU architecture is consistent with the latest R9/R7 series, and it can also support DX11.2. It also supports TrueAudio and PCI-E 3.0. More precisely, it is the same evolutionary GCN as R9 290X/290 and R7 260X.

856 floating-point performance: Floating-point performance of 856GFlops, up to 10% over Richland 779GFlops.

Specific specifications revealed: In a footnote explaining the source of floating-point performance, AMD even unambiguously disclosed the specific number and specification of the Kaveri APU flagship model: A10-7850K, quad-core, with a frequency of 3.7 GHz (may be the base frequency) , graphics core R7 series, 512 stream processors, frequency 720MHz.

Analysis of the specification principle: In addition to the evolution of specifications, Kaveri's greatest mission is of course to achieve a preliminary and complete HSA (heterogeneous system architecture): The first time a unified addressing is achieved, CPU/GPU can access the entire memory space together (maximum capacity 32GB ), and the new "heterogeneous queuing" (hQ) ensures equal CPU/GPU interaction, which is equally flexible in creating and distributing workloads.

The computing potential of the APU will therefore be completely released, and this is called "the most developer-friendly".

As for whether to support DDR4, AMD has not confirmed it. Although AMD said that heterogeneous computing is not related to the type of memory, DDR3, DDR4, and GDDR5 do not matter, but according to the situation, Intel will have to introduce DDR4 to the desktop after the Skylake platform, and the new memory is not yet fully implemented. Market preparation, the possibility of Kaveri is very small.

Finally, the graphics architecture has been significantly synchronized with the desktop level: TrueAudio synchronization was introduced more unexpectedly, of course, thanks to the deployment of improved GCN architecture. Prior to this, the APU had to lag one or two generations of desktop graphics on the graphics architecture, and this time it was finally synchronized.

Kaveri, APU's four-step strategy is the third step, which is the key stop for the company.