Slightly OCed CPU, slightly more memory BW, double the amount of CUs? I call BS!
I agree.
No comeback on the design differences between having the UVD and UVD as VCE in the APU vs Southbridge. It you also look at the XB1 APU which does have what amounts to a UVD in the APU, it has the same issues seen below and it pre-processes the vision input in Kinect and pushes HEVC for video chat to reduce the overhead in the PCIe between the APU and Southbridge.
The PS4 takes the ARM IP out of the APU and moves it to Southbridge
UVD in Southbridge.
Advantages:
1) Video chat, and game recording then take no CPU/GPU/Memory cycles from the APU or GDDR5 memory.
2) Southbridge using DDR3 can support Network standby, Root of Trust boot and Media TEE with fewer power islands in the APU.
3) You get a more secure TEE following all ARM recommendations.
4) Post processing for Video out and pre-processing video in for Gesture/head tracking take no GPU or GDDR5 memory cycles. The same Xtensa accelerators used for codecs are also used for vision processing.
5) HDMI 2's HDCP occurs in Southbridge as does CEC and all HDMI 2 features.
Disadvantage if this is not the case UVD in the APU:
1) Both video buffer and codec compression occur in the APU and a uncompressed HDCP 2 encrypted video stream to the custom HDMI that supports HDMI 1.4 and 2 and a codec compressed stream travels down the PCIe to Southbridge for video out to the HDMI chip and Hard Disk Game Recording
2) Game Data up from Southbridge via the same PCIe and video input for vision processing up through the same PCIe.
3) One of the VR modes has separate video plus VR distorted 120 FPS going down the PCIe to Southbridge at the same time.
4) HDMI HDCP negotiation is passed from the HDMI chip up through the PCIe to the ARM TEE in the APU.
5) CEC commands are passed up the PCIe to the ARM IP and acted upon then also forwarded back down the PCIe to the HDMI chip. HDMI 2's multi-view support (encrypted Video) over the LAN is also passed down the PCIe
6) HDMI 2 supports two views on the same monitor or with a breakout box seen in AMD's EyeInfinity or the Sony VR Goggles two TVs. These two views have to travel down the PCIe
That's 4 video streams down and 2 (stereo remember) video streams up the PCIe bus. Now add video chat while in game and you add another 2 streams up for 3D, it's now 4 video streams down and 4 up the PCIe.
With case one where Southbridge contains the ARM video processing only one video stream travels down the PCIe. Vision pre-processing in Southbridge would reduce the two video streams to a depth map up the PCIe which takes 50 % of the bandwidth that 1 video stream needs.
I think this explains the ooVoo delay till all the vision processing and Video is provided by Sony APIs calling Southbridge ARM IP. It makes sense that this occurs with the launch of VR this October. Sony may have been using Southbridge for some of the OS video but I think they have a NDA with AMD and Microsoft not allowing openVX and Playready from Southbridge till October.