Some news are worth mention, yet not worth entire thread, much less article covering… so that is the purpose of this thread.
And let's start with todays NDA end on new AMD APU's AMD Beema (GPU 800MHz, CPU 2.4GHz) and Mullins (GPU 500Mhz, CPU 2.2GHz) we can tell some informations. Previous APU (that is AMD call a CPU with GPU in one package) Kabini (GPU 600MHz, CPU 2.0GHz) and Temash (GPU 400MHz, CPU 1.4GHz) are quite outperformed by these new offerings.
The GPU part is basically same as Kabini - GCN architecture with 128 stream processors, 8 texture units and 4 ROPs. However with better design (38% less leakage = less heat and even less power required) and higher clocks is the performance higher and not only by the mere clocks bump. The memory interface (on witch GPU depend heavily) was reworked for the DDR3L standard. It go up to DDR3L-1866 clocks, witch should give some boost at well. (AMD says 5% on 3DMark CloudGate over DDR3-1600)
DDR3L is less power hungry, because it use 1.35V, as oposed to 1.5V to 1.6V used for most DDR3 rams (and hence the ram controller itself).
All in all, 15W APU Beema is in graphic about ~10% faster that 25W APU Kabini and also ~50% faster that 15W Haswell Pentium 3556U Intel is not going to be happy about this development.
As the CPU part go, the clocks also recieved a healty bump.(mainly the slower "tablet" CPU get raise from 1.4GHz to 2.2GHz, witch is quite a lot). The core architecture is now codenamed Puma+. 19% less leakage. The new "turbo" mode is more aggressive, so now the STAPM technology allow the CPU run at turbo (when need) before the 40°C limit of outside case temperature is reached, and only then the throttling come in action.
In short it means more power when need and less waste of power for powering apps, that did not benefit from turbo.
And the probably most interesting change is that these new APUs containing a whole ARM Cortex-A5 core (Trusted Platform Module, TPM) with own Cryptographic coprocessor. AMD call this together as PSP (Platform Security Processor) and NSA must be really "happy" about this development.
(time to push for backdoors, you $#%!)
Now only if developers support this, since that is the main question, if this move will prove as valid in long run.
There is also slightly more logical (for customer) labeling of the GPU cores, atc. But that it out of the quick news scope