@tormentos said:
@ronvalencia said:
Example with Red Dead Redemption 2
X1X GPU's 6 TFLOPS and rasterization improvements yields native 4K's 8,294,400 pixels performance.
PS4 Pro GPU's 4.2 TFLOPS yields 4,147,200 pixels performance. Based from PS4 Pro's experience, Sony claimed 8 TFLOPS is needed for 4K which is correct for scaled up PS4 Pro's design.
For 30 hz 4K, rendering process needs to render within 33 ms frame completion time and there's many paths to reach this goal.
From Sony's POV, 8 TFLOPS would reduce their frame completion time within 33 ms target for 8,294,400 pixel render performance. This is near brain dead improvement path since AMD was just slapping on additional CU with the same quad rasterization units based GPU design. Fury X has 8.6 TFLOPS with worst TFLOPS per frame rate yields when compared R9-390X.
From MS's POV, 6 TFLOPS increase and bottleneck reduction with rasterization path would reduce their frame completion time within 33 ms target for 8,294,400 pixel render performance. X1X GPU's ROPS has high speed 2MB render cache almost like Vega ROPS's unified 4MB L2 cache design. RTX Turing ROPS has doubled L2 cache storage relative to Pascal counterparts.
The next console hardware should be combining both points of view which assumes Sony learns from NVIDIA and MS's X1X's design approach.
Another interesting observations
1. RX-590's 7.1 TFLOPS at 1545 Mhz clock speed rivals Fury X's 8.6 TFLOPS at 1050 Mhz. At 1545 Mhz clock speed, RX-590's quad rasterization units would be 47 percent faster than Fury X's quad rasterization units at 1050 Mhz. PS; RX-590 needs 64 ROPS and faster memory modules.
2. Vega 56 at 1710 Mhz with 12.2 TFLOPS beating Strix Vega 64 at 1590 Mhz with 13.02 TFLOPS. Vega 56 at 1710Mhz's classic GPU hardware (quad rasterization, quad geometry, 64 ROPS, L2 cache, command processor) is 7.5 percent faster over Strix Vega 64 at 1590 Mhz.
TFLOPS doesn't include classic GPU hardware's read/write paths.
Understand the reasons for NVIDIA Pascal and Turing GPUs has high clock speed approach.
1. lol how many games are not true 4k on xbox one X?
2. So now because RDR2 is 4k that mean the xbox one is a 4k machine..Hahahhaaa
3. Lets ignore Rockstart Engines runs at 1080p on xbox one like GTA5 when most game don't do that on xbox one.
4. The PS4 Pro perform lower because it has a weaker GPU,the rest is history all the 203GB/s bandwidth crap that MS spew for months helped the xbox one in nothing when its GPU was a piece of crap.
5. And there are recorded gaps of more than 100% between the PS4 and xbox one,but i remember how you chose to ignore those.
1. That's not the argument. Higher spec hardware has higher chances for native 4K when compared to lesser hardware.
There are many games on X1X that are native 4K that are not native 4K on PS4 Pro.
2. That's not the argument. My point is the performance gap between the two boxes and shows Sony's 8 TFLOPS comment for 4K is true for their PS4 Pro's GPU configuration while not true for X1X''s GPU configuration.
Open world Far Cry 5 is native 4K on X1X not just RDR2.
3. Your argument with GTA5 is a red herring when RDR2 on XBO is NOT running 1920x1080p. .
4. The facts
Both PS4 Pro and X1X has similar size chip area budget i.e. ~321 mm2 APU with 40 CU GCN (36 CU being active) vs 359 mm2 APU with 44 CU GCN (40 CU being active) + 2MB render cache for ROPS.
MS focus on improving rasterziation power while Sony focused on TFLOPS via RPM. Sony stayed with pure FLOPS arguments since PS2 and later PS3.
TFLOPS is meaningless without rasterziation power being factored in.
5. 32 MB SRAM can impose hard resolution limits despite relative CU power difference between XBO and PS4. Both XBO and PS4 has dual rasterization units setup i.e. convert floating point geometry into integer based pixel grid GPU hardware core design.
MS's 208 GB/s memory bandwidth argument can NOT have 203 GB/s frame buffer write in burst mode since it's hard split between read and write I/O links. Furthermore, it can't overcome CU bound issues as shown in my W5000 vs 7850 example. XBO's 32MB SRAM design was abandoned with X1X design.
When compared to L2 cache level bandwidth (e.g. 1 TB/s on R9-290X at 1Ghz), XBO's 32 SRAM is not significantly better over 256 bit GDDR5-5500. XBO's 32 SRAM design has flaws. Too bad AMD's ROPS are not connected to high speed L2 cache until Vega ROPS while X1X GPU's ROPS are connected super fast 2MB render cache.
Multi-MB high speed cache is useful when there's excess CU power and memory bandwidth not evolving with TFLOPS increase pace.
For RDR2,
PS4 Pro's 1920x2160 delivered 2X over PS4's 1920x1080 results as per TFLOPS increase scaling. PS4 to PS4 Pro shows old GCN TFLOPS improvements i.e. effectively two PS4 GPUs glued together as stated by Mark Cerny. Two Pitcairn GCNs with 20 CU = 40 CU. Baseline Polaris IP applied with selected Vega IP e.g. RPM.
X1X's 3840x2160 delivered 4X over PS4's 1920x1080 results which shows other improvements besides 3.26X TFLOPS increase over PS4. X1X's GPU is based on Hawaii GCN with 44CU. Hawaii GCN's ROPS has tiny few KB render cache which seems to be expanded into 2MB with X1X GPU version. Baseline Polaris IP applied with selected Vega IP e.g. ROPS being connected to multi-MB cache.
https://www.kitguru.net/gaming/matthew-wilson/red-dead-redemption-2-console-resolutions-and-performance-confirmed/
The Xbox One S is the weakest console and that shows, with busy areas of the game running at around 27 to 28 frames per second at 872p, or 66 percent of Full HD. Meanwhile, the PS4 manages a full 1080p, although it can struggle to maintain 30 frames per second in busy towns.
The PS4 Pro and Xbox One X are really where things start to get interesting. The PS4 Pro runs the game at 50 percent of full 4K, or 1920×2160 with checkerboard rendering in place. However, performance is more stable compared to the base Xbox One and PS4. The Xbox One X on the other hand is described as the most stable way to play the game, with just one cutscene frame drop. Microsoft’s top-end console also runs the game at a native 4K resolution without dynamic scaling, making it the best looking version to boot.
XBO S's 66 percent of Full HD is about the same TFLOPS gap percentage with PS4.
PS4 to PS4 Pro shows 2X results as per TFLOPS gap.
Console GCNs prior to X1X roughly follows TFLOPS scaling.
Log in to comment