The COD dev and the leak document said 2ms.
As an aside, the reason PSSR is unused here it seems is in part the frame-time cost: it takes up 2.1ms of the frame-time
Source
Reading Thomas Morgan's words as is:
"here
it seems is in part the frame-time cost: it takes up 2.1ms of the frame-time, which in its current form is too great a chunk of the required 8.3ms frame-time required to run at 120fps. Later,"
and without a developer quote or Cerny spec to back that up I would argue GT7 completely contradicts that. Also simple maths suggests if 60fps with all the bells and pssr is internally 2560x1440, then 120 at the very best could scale to 1280x720, which devs don't do even on single player games, where sub 1080p native is an optics issue against PC master-race and DF opinion with 900p being on the apex. Removing the bells + PSSR was costing half the frame-time from maintaining 2560x1440 native @ 120fps
Doing a bit of maths (linear perform for resolution vs frame-rate which their results already show)
They are saying that if we compare native 120fps to 60fps pssr + bells. The, pssr + bells is equal to 1/120th of a second(8.3ms) in every 60fps frame given that native resolution is unchanged. Meaning that if we take their 2.1ms as true for pssr, then the bells cost 6.2ms @ 1440p, and 1/4 of the pixels ~= 1.6ms for the bells (@1280x720@120), then if say 1/4 of native for a 1280x720@120 + bells + PSSR, we need another ~=2.1ms to render without bells. meaning adding it all up 1.6ms + 2.1ms + 2.1ms is much less than 8.3ms. So clearly the bells and whistles and rendering cost are at least 2.5ms more, and PSSR then doesn't cost 2.1ms when refactoring that 2.5ms into the other numbers.
1440p is x 1.78 more pixels than 1080p, so we can factor those numbers (if true) 8.3/1.78(native render@1080p) + 6.2/1.78(bells@1080p) + 2.1ms(for PSSR) ~= 10.3ms, you can see that PSSR at any cost is beyond native 1080p + bells at 1080p120
Running the numbers one last time
1440p is 2.56 more pixels than 900p(1600x900), so 8.3/2.56 + 6.2/2.56 + 2.1ms ~ = 7.8ms
Which I think clearly shows if as DF thinks PSSR needs 2.1ms, then CoD would have a PSSR with bells and whistles mode a native 900p120fps.
The 2.1ms AntiAliasing metric in the graph images isn't just for PSSR, unless we think CoD doesn't have an AA solution, but that's what they are implying.
That measurement includes all the base AA techniques used for the 60fps mode, because render modes without PSSR still need a clean image, so it isn't like they aren't using those techniques, that would support DF's 0 - 2.1ms for PSSR claim.
The PS5 breakdown shows a AA budget on the base PS5 at over 1.2ms, so this actually supports the view that PSSR is closer to adding 1ms to the critical path, not adding 2.1ms like they are guessing at.
