jeff_rigby
Banned
While a GPGPU can do physics calculations, the setup time is higher than for a FPU or a PPU+SPU combination. For parallel array work like Voice and gesture recognition where on-going (GPU compute units already setup) comparisons on streams (thus the name stream processor) of audio or video a GPU with hundreds of processors is best.if its a 'normal' AMD APU the GPU isn't that amazing, but I could see Sony going with something like that and using the embedded GPU for ancillary tasks. eg physics acceleration, encoding video for remote play to vita, post processing etc. Then have a separate GPU for the main grunt work.
They have a roadmap and some of the roadmap requires 28nm and 20nm which is just coming on-line at Fabs. AMD has been planning this for 5 years as has IBM. They had to wait for the technology to make it practical, some for Sony. Second GPU may move to the APU @20nm in 2 years.Otherwise for a proper next gen console you'd need a completely custom solution - there isn't anything on the normal AMD APU roadmap with the kind of power we're talking about here is there? and if AMD could squeeze a quad core CPU alongside a 7970m level GPU on one chip, why wouldn't they have done that for the PC market?
Another point, I expect Sony and IBM have been talking and planning the PS4 since before the PS3 was released. AMD should have joined the conversation just after 2008 when IBM turned to AMD for GPUs and they formed the Consortium (IBM, AMD/Global Foundries and others). IBM in 2010 dropped support for Cell but Sony patented a 1PPS4SPU HSA module Dec 2010 that had crossbar-coherent cache and more that are very similar to AMDs HSA plugin building block design.... coincidence? Dec 2011 patent and the 1PPU+4SPU element is interchangeable with the term APU which is talked about in 2009 for AMD FUSION APUs.
This can be Sony totally ignorant of AMD plans and the benefits of using GPUs for stream processing (Voice and gesture recognition 2008 Toshiba and Sony talking about it for TVs and more using Cell) which IBM was researching and publishing whitepapers using Nvidia GPUs and Cuda in 2008 or more likely Sony developing 1PPU4SPU modules for use in other CE equipment and the PS4. It could be either.....
If Sony wasn't (as some believe) forced to change plans to use AMD because the 1PPU4SPU wafer patented in Dec 2010 was not ready Dec 2011 and Sony switched to AMD because they had a solution READY (latest Sony comment says they are not worried about being later if that makes the PS4 better) then the 1PPU4SPU does still have a use. Any other logic has to assume Sony is stupid and suddenly decided to switch from Cell to AMD APU (X86+GPU).
The PS3 could be redesigned with 1 or 2 (1PPU4SPU) modules in a SOC with stacked memory to be released at the end of this year. OS would have to be redone but that same OS could be used in the PS4 if it had 1 or 2 (1PPU4SPU) modules to provide backward compatibility. Toshiba is using what is essentially 1 CPU plus 4 SPUs to provide head tracking, Gesture recognition and video codec support in their 4K TVs (No dual use for GPU so SPUs are more practical in a TV).
This is all wild speculation with little to support it. Current developer hardware has FPUs, next (3rd) generation developer platforms may have Jaguar + Cell as FPU. Developers will use the same calls and may not know the Cell is in the SOC. Backward compatibility would not need to be known by developers and added functionality using Cell may wait till after PS4s release. IF a slimmer slim is announced at TGS 9/20/2012 then my wild speculation has more weight. Sept is when Khronos has targeted AR and is the next Gnome 6 month cycle which is the same as the GTKwebkit and Gstreamer cycle.