Digital Foundry Article Technical Discussion [2024]

It's because game with PhysX isn't as demanding as that test makes it seem. You also likely weren't playing a 4k.





It's worthing keeping mind that PhysX likely doesn't scale that well after certain point. I think even once we got to Maxwell the gains from a having a dedicated card for PhysX started to become questionable (and I vaguely recall in some cases even a performance loss likely because the overhead outweighed the compute gain from the card).
 

Awesome stuff!

There was perhaps more frames dropped in my tests than I expected, but one aspect that I was impressed with was how responsive the dynamic res is - you often don't see so much as a ms blip on the frametime graph even with relatively large resolution swings.

Part of that may be due to Insomniac's approach with fixed up/downscale res targets rather than having a more fluid range, but regardless it works very well, even by console standards which are usually much better at this than PC implementations.
 
That's not what I got from it, seems they're just running on the code on the logic that's the fastest at running it.

Which consoles do anyway, because they're consoles.
maybe this will work even better. A new CPU 100X more efficient than current CPUs, the Monza CPU.

 

Awesome stuff!
This is fascinating. What's the downside to making tools like this available? I mean the performance analysis stuff, not the swinging around under the map. I don't see how this hurts the studio or anything.
 
This is fascinating. What's the downside to making tools like this available? I mean the performance analysis stuff, not the swinging around under the map. I don't see how this hurts the studio or anything.
You gotta ask yourself the opposite question: what are the upsides of making those tools available?

Devs aren't looking for reasons not to do things, they are looking for reasons to do them.
 

Pretty astonishing work.

Also, and I might sound totally crazy for this, but this version somehow manages to look.....more realistic to me....than the main console or PC versions. At least in a sense. There's something about the low resolution and reduced texture detail combined with the lower framerate character animations that gives it almost an FMV look. But even outside of characters, the sharpening and low res with very natural lighting model seems to give environments a boost to realism as well, where it's allowing my mind to make up the difference more than the 'clearer' and smoother versions of the game do.
 

Pretty astonishing work.

Also, and I might sound totally crazy for this, but this version somehow manages to look.....more realistic to me....than the main console or PC versions. At least in a sense. There's something about the low resolution and reduced texture detail combined with the lower framerate character animations that gives it almost an FMV look. But even outside of characters, the sharpening and low res with very natural lighting model seems to give environments a boost to realism as well, where it's allowing my mind to make up the difference more than the 'clearer' and smoother versions of the game do.
Agreed, that's an astonishing port for the hardware target, and I think they made all the right tradeoffs.

It's what I've been trying to get across about ray tracing and now more recently path tracing to a lot of folks. I get that everyone has different tastes, but for me at least, better pixels are always far preferable to more pixels.

People hyperfixate on resolution, especially internal rendering resolution now that upscaling is nearly universal.

The same can be said for UE5 and Nanite/Lumen and the argument that it's somehow a detriment to gaming in general that on current midrange PC hardware and console hardware, titles that use those features have a relatively low resolution target.

If I wanted to play a game with low geometric detail, unrealistic (or barely existing) lighting, shadows, and reflections, those games have already been made. Just load up a game from 10-15 years ago and crank it up to 8k120, assuming it supports it or can be modded to do so, and you're already there, especially if you can throw a high res texture pack at it.
 
Really impressive. The cuts to geo are a hard to see for the open outdoor environments but the town scenes look great.
 
wonder if quest 3 could run a VR version of this port.

IMO the game checks all the boxes for why it wouldn't be suitable for VR (on any system really). Everything from performance to mechanics to art. Sustained high FPS is out of the question. The gameplay mechanics involve a lot of uncontrollable animation, and it all drives your camera POV. The game's visual hallmark (forests) probably isn't going to hold up in VR -- it's either going to look like unresolvable pixel noise at a distance or a bunch of 2D cards and billboards close up.

The helmet view with the visor down could feel amazing though. It would mask the limited FOV of the HMD, you could bake-in the DOF effect to the visor edges given how close they are to the eyes, and you could add some kind of HRTF of what being inside the helmet sounds like. Even the weight and awkwardness of the HMD might be a benefit.
 

I don't believe Rich mentions the CPU he used, however.

So I was bored and decided to the test the ray tracing scene in CP2077 where PS5 seems to be doing much better than expected.

PS5 quality settings and matched FSR2 dynamic resolution scaling settings obtained from this DF video.

A quick reminder of the frame rate that PS5 and other PC GPU's run at:

DF frame rates.png

4070ti at PS5 settings and with matched FSR DRS - 3.48x faster than PS5

PS5 settings.png

4070ti at PS5 settings with DLSS Quality mode - 4.18x faster than PS5 while offering better image quality

PS5 setings - DLSS Q.png

4070ti at PS5 settings with DLSS Quality mode plus frame generation - 6.59x faster than PS5 while offering better image quality

PS5 settings - DLSS Q + FG.png

4070ti at PS5 settings and with matched FSR DRS, but with MAX RT settings - 2.25x faster

PS5 settings - Max RT.png

4070ti with Path Tracing enabled with DLSS Quality mode - 1.92x faster

PT - DLSS Quality mode.png

A bonus comparison: PS5 RT mode vs Path tracing
 
Last edited:
So I was bored and decided to the test the ray tracing scene in CP2077 where PS5 seems to be doing much better than expected.

PS5 quality settings and matched FSR2 dynamic resolution scaling settings obtained from this DF video.

A quick reminder of the frame rate that PS5 and other PC GPU's run at:

View attachment 11011

4070ti at PS5 settings and with matched FSR DRS - 3.48x faster than PS5

View attachment 11012

4070ti at PS5 settings with DLSS Quality mode - 4.18x faster than PS5 while offering better image quality

View attachment 11013

4070ti at PS5 settings with DLSS Quality mode plus frame generation - 6.59x faster than PS5 while offering better image quality

View attachment 11014

4070ti at PS5 settings and with matched FSR DRS, but with MAX RT settings - 2.25x faster

View attachment 11015

4070ti with Path Tracing enabled with DLSS Quality mode - 1.92x faster

View attachment 11016

A bonus: PS5 Quality settings vs path tracing

Nice job, that's a really well done comparison, and really informative.
 
So I was bored and decided to the test the ray tracing scene in CP2077 where PS5 seems to be doing much better than expected.

PS5 quality settings and matched FSR2 dynamic resolution scaling settings obtained from this DF video.

A quick reminder of the frame rate that PS5 and other PC GPU's run at:



4070ti at PS5 settings and with matched FSR DRS - 3.48x faster than PS5



4070ti at PS5 settings with DLSS Quality mode - 4.18x faster than PS5 while offering better image quality



4070ti at PS5 settings with DLSS Quality mode plus frame generation - 6.59x faster than PS5 while offering better image quality



4070ti at PS5 settings and with matched FSR DRS, but with MAX RT settings - 2.25x faster



4070ti with Path Tracing enabled with DLSS Quality mode - 1.92x faster



A bonus: PS5 Quality settings vs path tracing
Are you sure you matched the settings and resolutions properly? Because DF has the 4070S beating the PS5 by 2.21x and it isn't much slower than the 4070 Ti. This would make your 4070 Ti 57% faster than the 4070S which is hard to believe.

 
So I was bored and decided to the test the ray tracing scene in CP2077 where PS5 seems to be doing much better than expected.

PS5 quality settings and matched FSR2 dynamic resolution scaling settings obtained from this DF video.

A quick reminder of the frame rate that PS5 and other PC GPU's run at:

*snip*

4070ti at PS5 settings and with matched FSR DRS - 3.48x faster than PS5

*snip*

4070ti at PS5 settings with DLSS Quality mode - 4.18x faster than PS5 while offering better image quality

*snip*

4070ti at PS5 settings with DLSS Quality mode plus frame generation - 6.59x faster than PS5 while offering better image quality

*snip*

4070ti at PS5 settings and with matched FSR DRS, but with MAX RT settings - 2.25x faster

*snip*

4070ti with Path Tracing enabled with DLSS Quality mode - 1.92x faster

*snip*

A bonus: PS5 Quality settings vs path tracing
Nice job man!
 
Are you sure you matched the settings and resolutions properly? Because DF has the 4070S beating the PS5 by 2.21x and it isn't much slower than the 4070 Ti. This would make your 4070 Ti 57% faster than the 4070S which is hard to believe.


Your video is of the 2160p 'resolution' quality mode and not the RT quality mode that I used for my comparison.

So yea, completely different to what I did as they have different resolution targets.
 
Last edited:
Back
Top