davis.anthony
Veteran
Never played Days Gone on PS4 Pro did you?
Digital Foundry Article Technical Discussion [2024]
- Thread starter Shifty Geezer
- Start date
Never played Days Gone on PS4 Pro did you?
A very good video with lots of details. Unfortunately, we can’t say the same about the game. Those animations in particular don’t look mocaped. It looks really bad. 720p internal res for this poor looking game? This game just looks so bad.
Any game using FSR2 looks bad and worse than any good looking previous gen game. Alex on FSR2 on Skull & bones:
For me any game using FSR2 should be labelled as such. With FSR2 reflections / transparencies are using the lower native resolution and they are terrible to look at. It's incredibly distracting and immersion breaking. Worse than playing a PS3 or PS4 games which were at least consistent.
AC black flag runing on PS4 has superior reflections actually running at 1080p instead of... 720p here on PS5! The developers have completely lost their priorities. Final IQ should always be the target.
From now on, and after The Talos Principle 2, it's simple, if a game is using FSR2 I won't even consider buying it.
Agreed. The inability to actually board the enemy ship and fight the crew is mind boggling. That was easily one of the best bits of Black Flag.
We could also do that in AC Odyssey. Super fun indeed.
Flappy Pannus
Veteran
Yeah, it's great in DG. But it's also rendering starting from a native 1920x2160.
Checkerboarding is unfortunately likely not a saviour for FSR2's image woes today on console games though, because the inherent problem is the load modern engines are putting on this hardware (at least when going for 60fps) necessitates that the native resolution is going to be drastically lower than past gen games which employed checkerboarding (and often with a goal of 30fps to boot). Checkerboarding in these games would very likely look quite poor too, albeit I'd still be very curious to see the differences, particularly with stuff like occlusion issues and low-res buffer treatment.
The scaling of this game is also very weird. FSR2 has a cost, yes - but going from 720p native to 4k native "only" halves your frame rate?! I appreciate them aiming for a very stable 60fps but I'm wondering if it really needs to be 720p for that.
Hard disagree here. I think Guerrilla with Decima basically solved the reconstruction problem on consoles for sharp, clean and 60fps games with Horizon Forbidden West on PS5 (native res being only half CBR 1800p so about native 1270p which is pretty decent for alphas). Every others developers should look at what they have done here. With FSR2 they are reconstructing from too low, cost is super high cause they try to target too high final output. Here for Skull & Bones:
Something is bloody wrong here. if they can do native 4K 30fps with TAA (like Horizon FW) then they could do CBR 1800p 60fps. Cost of FSR2 must be higher than CBR. And native resolution is too low creating all those artefacts. You don't polish a 720p turd. Also CBR has other advantages compared p to FSR:
This means more stability in motion and I have seen this in almost all CBR solutions I have looked at. When motion is still slow, the final image actually passes for native stuff. The first time DF saw a CBRed game (A COD), they really thought it looked native when putting their eyes near the TV. Same thing with Dark Souls Remastered, they never thought it could be CBR 1800 (they still don't). My Gran Turismo S copy is still 1800p CBR and is also pretty good in motion. Not incredibly ugly artefacts there.
Again the only good thing about FSR2 is in very specific scenes: simple walls or rocks. Sure they do look sharp. But mostly everything else is a disaster. Vegetation in motion? pixel soup. Transparencies? Good god my poor eyes. Reflections? Hey we must use the native res here guys, sorry. Reflections behind transparencies? Someone help!
Why developers don't use more CBR? Well because it's hard to get right obviously! Plenty of things must be looked at, some effects must be done at a higher resolution and such. Even Guerrilla had trouble to get it right at first, and they are one of the most talented out there. But once the work has being done, the 30fps mode has being rendered obsolete for many.
I know checkerboarding from enough games to see that TSR and DLSS are vastly superior.
davis.anthony
Veteran
That's an over exaggerated claim as depending on the quality of the implementation will depend on what looks better.
In Death Stranding at least, DLSS easily beats Checkerboard Rendering. DLSS is simply a better upscaling method and really shouldn't lose under any circumstances.
davis.anthony
Veteran
Better on paper and in a perfect world, but we live in the real world.
I recently bought Battlefield 5 as it was on sale and DLSS in the game is atrocious.
DavidGraham
Veteran
I wouldn't say broken, since it's optimized for RDNA2 on the PS5, which should translate to nice performance on the RDNA2 PC side as well. A more reasonable explanation is the limited memory bandwidth on the RX 6700 compared to PS5.
Exactly, and in Ray Tracing as well.
Once you become GPU limited, using a higher CPU won't get you any more fps than possible. I used an i7 3770K for years, I upgraded to a 7800X3D recently, in heavy RT titles I didn't gain a single fps, because I was limited by my GPU at all times. Digital Foundry made sure to test only GPU limited scenes on the PS5.
No you can't, in the GPU limited scenes used by DigitalFoundry, you won't find a single difference between a 3600X and a 7800X3D.
I think Battlefield 5 still has DLSS 1, no?
I'd expect there are different implementations between console and PC with the PC implementation favouring Nvidia due the sonsorship deal. I don't know specifics for CP2077 but for example I know there are games that use in line RT on consoles but switch to DXR 1.0 on the PC side.
Avatar is another one that Richard mentions in the video uses a custom RT implementation for consoles.
I do agree the memory bandwidth, and perhaps VRAM capacity may also be hobbling the 6700 at those settings though.
Doubt it’s VRAM because the 2070S with its 8GB does much better.
They can only be sure for the PC version, not the console using underclocked laptop CPU. From 60 - 80fps I'd say as a general rule we can suppose most games will be CPU limited on consoles notably when many games use DRS so reducing the risk of being GPU limited.
This is why you need to use a modest CPU on PC as well to see which games start to be CPU limited with such a weaker CPU! This is not some unreasonnable request here! This has being done for decades on PC vs PC benchmarks, for good reasons.
DavidGraham
Veteran
DF picks the areas in which PS5 suddenly falls below 30/60 fps, while also being GPU limited on the RX 6700. Thus ensuring they are testing the areas with actual GPU limitations across all platforms. They don't randomly pick areas, they synchronize areas across both PC and PS5.
The code on PC could be more CPU limited, due to difference of APIs between platform. Generally speaking DX12 games are more CPU limited on PC than their PS5 counterparts. So using identical CPUs will actually harm the PC GPU experience. Never the less, this point is moot, considering what I said above.
Against DLSS1, sure CB wins. But against DLSS2? of course not. BF5 has the very first DLSS1 implementation.
davis.anthony
Veteran
I'm sorry but no, some games have horrendous ghosting with DLSS2 and to a level that I've not seen in any CB games I've played.
DavidGraham
Veteran
TAA has horrendous ghosting as well, often times both are tied (when TAA is ghosting hard in a game, DLSS ghosts as well, it's an issue of temporal solutions in general), but ghosting with DLSS is often significantly lower. However, we are debating the reconstruction of fine details and image components here. DLSS2 is without a question the superior solution here.
Similar threads
