Xbox One S wouldn't be able to do native 4K games any better than the normal PS4, actually less so because:
Xbox One GPU: 1.3 TFLOPs and 13.6 Billion pixels/sec fillrate (16 ROPs * 853 MHz)
XBox One S: 1.4 TFLOPs and 14.6 billion pixels/sec fillrate (16 ROPs * 914 MHz)
PS4 GPU: 1.8 TFLOPs and 25.6 billion pixels/sec fillrate (32 ROPs * 800 MHz
People almost never mention pixel fillrate anymore, just TFLOPs. Even though fillrate is not as important as it used to be (shader ops took over), you still need much higher pixels/sec to render at native 4K (or any res) with a decent framerate while at least maintaining the same level of graphical fidelity. The lower pixel fillrate on original Xbox One is one of the main reasons many of its games were sub-1080p (900p, 720p) while the PS4 versions were 1080p
Microsoft never said Xbox One S would do games at 4K, only HDR in games, and 4K Ultra HD Blu-ray movies and streaming video.
They have a console specifically for native 4K games, it's called Project Scorpio, for Holiday 2017.
1920x1080p vs 1600x900p points to ESRAM storage limitation (without tiling) or shader power difference.
XBO's 16 ROPS limitation with 8 to 16 bit datatypes are mitigated by compute mode TMU workarounds.
With 32 bit datatype, both 16 ROPS and 32 ROPS are memory bandwidth limited.
XBO's 16 ROPS limitation can be workaround, but shader power difference can not be workaround.
---------------
For Scorpio... Let say it's 1310 Mhz with 36 CU for 6 TFLOPS
For 32bit datatype, 1300 Mhz x 16 ROPS x 16 bytes = 332 GB/s. Another 16 ROPS cluster wasted for 32bit datatype, but fine for 16 bit datatype.
For 16bit datatype, 1300 Mhz x 32 ROPS x 8 bytes = 332 GB/s.
Based on Polaris' effective memory bandwidth which includes memory compression, Scorpio has 330 GB/s effective memory bandwidth.
For 32bit datatype, 1300 Mhz x 32 ROPS x 16 bytes = 664 GB/s.
Where's that 1 TB/s HBMv2...
---------------
For NEO... Let say it's 911 Mhz with 36 CU for 4.2 TFLOPS
For 32bit datatype, 911 Mhz x 16 ROPS x 16 bytes = 233 GB/s Another 16 ROPS cluster wasted for 32bit datatype, but fine for 16 bit datatype.
For 16bit datatype, 911 Mhz x 32 ROPS x 8 bytes = 233 GB/s.
Based on Polaris' effective memory bandwidth which includes memory compression, NEO has 225 GB/s effective memory bandwidth.
For 32bit datatype, 911 Mhz x 32 ROPS x 16 bytes = 466 GB/s.
Higher 32bit datatype usage can be associated with PC's max detail settings, hence effective memory bandwidth play a higher role with frame rate results.
AMD needs another memory compression revision to reduce thier GPU's higher FLOPS being gimped by memory bandwidth issues.
Log in to comment