ConsoleWatch: PlayStation 4 "Orbis" CPU Specs Leak, But Likely Reflect Early Hardware
Let's start with the APU. The A8-3850 was one of AMD's earliest Llano chips; a 100W 2.9GHz quad-core with an integrated Radeon 6550D. It didn't ship in high quantity -- AMD chose to emphasize shipping out mobile Llano's rather than their desktop counterparts. Llano is a much stronger mobile chip than it is on the desktop; AMD has been unable to scale the chip to higher clock speeds (think 3GHz+) without dramatically increasing its TDP.
One fan's idea of what the PS4 "Orbis" might look like
Llano's Achilles heel is the interconnect between its CPU and GPU; the communication channel is very similar to an old-school motherboard northbridge implementation. It's easy for the CPU to transfer data to the GPU but much more difficult for the GPU to do the same -- which means it's also much harder for AMD to take advantage of Llano's array of GPU cores as a general-purpose compute array. Bandwidth is quite limited; Llano's GPU is critically dependent on main memory bandwidth to function well.
Now, let's talk about the Radeon 7670. It's a rebranded HD 6760, based on the budget Turks 40nm GPU. According to IGN, "When the APU is paired with the HD 7670, however, Sony will be able to utilize an asymmetrical CrossFire configuration to share the load of realtime graphics processing." That's technically true, but the benefits of Hybrid Crossfire are limited to DX10/11 games (it's unclear how AMD would scale the benefit to OpenGL and older DX9 titles are often slower than they'd be with just the dGPU.) Scaling benefits are also erratic and vary significantly from game to game.
The reason we don't believe reports that Sony would adopt the Llano+6760 is that AMD has <b>much</b> better hardware either already shipping or coming in the very near future -- certainly well before the PS4's launch date. Llano may have done a great job getting AMD's foot in the door, but the chip is based on AMD's four-year-old Shanghai CPU. AMD has no plans to continue building Llano at nodes below 32nm, and console manufacturers always plan to scale a CPU through multiple process nodes. The current PS3 is produced on a 45nm/40nm process, having begun life at 90nm.
If Llano's GPU was good, Trinity's is expected to be significantly better. More importantly, its GPU is based on Cayman, which means it's a substantial improvement over the HD 5000-era part baked into Llano. Similarly, AMD might well target a budget price point for whatever GPU the PS4 eventually uses, but it probably won't be derived from a graphics architecture that'll already be three years old by the time the system ships.
AMD's next-generation roadmap for 2013. Trinity is replaced by Kaveri / Kabini
If Sony is planning to use AMD hardware in a configuration similar to what IGN describes, we expect the shipping product will almost assuredly at least use Trinity and pair it with a 28nm HD 7000 chip. What's more likely is that the final product will be built around Kaveri; the architecture AMD expects to launch in 2013. Not only would a Kaveri-derived product put Sony on par with what's shipping in the mainstream market, the enhancements coming as part of AMD's Heterogeneous System Architecture would make it much easier for system programmers to exploit the CPU/GPU dynamic to extract maximum parallelism and performance.