• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

NVIDIA DLSS 3.5 Ray Reconstruction

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
27,976 (3.71/day)
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
NVIDIA DLSS 3.5 launches today, supporting GeForce 20 and newer. The new algorithm improves the looks ray traced lighting, reflections, shadows and more. In our DLSS 3.5 Ray Reconstruction review we test image quality, but also discovered that VRAM usage actually goes down and performance goes up.

Show full review
 
Joined
Jun 6, 2021
Messages
711 (0.55/day)
System Name Red Devil
Processor AMD 5950x - Vermeer - B0
Motherboard Gigabyte X570 AORUS MASTER
Cooling NZXT Kraken Z73 360mm; 14 x Corsair QL 120mm RGB Case Fans
Memory G.SKill Trident Z Neo 32GB Kit DDR4-3600 CL14 (F4-3600C14Q-32GTZNB)
Video Card(s) PowerColor's Red Devil Radeon RX 6900 XT (Navi 21 XTX)
Storage 1 x Western Digital SN850 1GB; 1 x WD Black SN850X 4TB; 1 x Samsung SSD 870EVO 2TB
Display(s) 1 x MSI MPG 321URX QD-OLED 4K; 2 x Asus VG27AQL1A
Case Corsair Obsidian 1000D
Audio Device(s) Raz3r Nommo V2 Pro ; Steel Series Arctis Nova Pro X Wireless (XBox Version)
Power Supply AX1500i Digital ATX - 1500w - 80 Plus Titanium
Mouse Razer Basilisk V3
Keyboard Razer Huntsman V2 - Optical Gaming Keyboard
Software Windows 11
In a nutshell, nothing revolutionary, but a respectable step in the evolutionary process of efficiently increasing visual quality.
 
Joined
Oct 6, 2020
Messages
116 (0.08/day)
Processor AMD Ryzen 7 9800X3D
Motherboard Asrock X870E NOVA
Cooling Arctic Liquid Freezer III 360
Memory G.Skill TRIDENT Z 32GB 6000MHz CL30 DDR5
Video Card(s) MSI Suprim X RTX 4080 16GB
Storage Kingston KC3000 1TB
Display(s) Alienware AW3423DWF 34" 21:9 OLED
Case Antec C8
Power Supply NZXT C1200 Gold ATX 3.1
Mouse Roccat Kone Air
Keyboard OZONE StrikePro Spectra (CherryMX Red)
Software Windows 11 Pro x64
Which driver has been used with those benchmarks? I'm asking cause Nvidia released a new ones just today (537.42) and they support Phantom Liberty among the others.
 
Last edited:
Joined
Sep 10, 2018
Messages
7,015 (3.04/day)
Location
California
System Name His & Hers
Processor R7 5800X/ R7 7950X3D Stock
Motherboard X670E Aorus Pro X/ROG Crosshair VIII Hero
Cooling Corsair h150 elite/ Corsair h115i Platinum
Memory Trident Z5 Neo 6000/ 32 GB 3200 CL14 @3800 CL16 Team T Force Nighthawk
Video Card(s) Evga FTW 3 Ultra 3080ti/ Gigabyte Gaming OC 4090
Storage lots of SSD.
Display(s) A whole bunch OLED, VA, IPS.....
Case 011 Dynamic XL/ Phanteks Evolv X
Audio Device(s) Arctis Pro + gaming Dac/ Corsair sp 2500/ Logitech G560/Samsung Q990B
Power Supply Seasonic Ultra Prime Titanium 1000w/850w
Mouse Logitech G502 Lightspeed/ Logitech G Pro Hero.
Keyboard Logitech - G915 LIGHTSPEED / Logitech G Pro
Thanks for all the hard work you really deserve a break. I am excited to try this out today for myself.

100% the settings menu in CP sucks though it always want's to enable other stuff for some reason.
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
27,976 (3.71/day)
Processor Ryzen 7 5700X
Memory 48 GB
Video Card(s) RTX 4080
Storage 2x HDD RAID 1, 3x M.2 NVMe
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
Which driver has been used with those benchmarks?
537.34 as recommended by NVIDIA for this specific article

Edit: added the test system table at the end of the performance page

Thanks for all the hard work you really deserve a break
Yeah it's been a busy week
 
Last edited:
Joined
Apr 30, 2008
Messages
4,902 (0.81/day)
Location
Multidimensional
System Name Boomer Master Race
Processor AMD Ryzen 7 8745H
Motherboard MinisForum 870 Slim Board
Cooling Mini PC Cooling
Memory Crucial 32GB 5600Mhz
Video Card(s) Radeon 780M
Storage Kingston 1TB SSD
Display(s) Sony 4K Bravia X85J 43Inch TV 120Hz
Case MinisForum 870 Slim Case
Audio Device(s) Built In Realtek Digital Audio HD
Power Supply 120w External Power Brick
Mouse Logitech G203 Lightsync
Keyboard Atrix RGB Slim Keyboard
VR HMD ( ◔ ʖ̯ ◔ )
Software Windows 11 Pro 64bit
Benchmark Scores Don't do them anymore.
TY for including the 4070 in this.
 
Joined
Oct 28, 2012
Messages
1,195 (0.27/day)
Processor AMD Ryzen 3700x
Motherboard asus ROG Strix B-350I Gaming
Cooling Deepcool LS520 SE
Memory crucial ballistix 32Gb DDR4
Video Card(s) RTX 3070 FE
Storage WD sn550 1To/WD ssd sata 1To /WD black sn750 1To/Seagate 2To/WD book 4 To back-up
Display(s) LG GL850
Case Dan A4 H2O
Audio Device(s) sennheiser HD58X
Power Supply Corsair SF600
Mouse MX master 3
Keyboard Master Key Mx
Software win 11 pro
"Better than native" -> Proceeds to show a plethora of comparisons where DLSS borks effects like smoke, shadows, DoF or bloom.

Why is the conclusion longer than the rest of the article anyway?
The shadows in some places are blurred, take a look at the "turbo" back there. In other places there is a superposition of contrast where it is very bright.
The shadows varying in sharpness might actually be a correction. In real life shadows are not always going to be sharp all the way through. Soft shadows are a feature in state-of-the-art pre calculated 3D renderers.
Soft Shadows in PRMan (pixar.com)
1695311753998.png
 
Joined
Jan 23, 2017
Messages
42 (0.01/day)
Its impressive tech but it obviously has some drawbacks right now, one being the absolute destruction of some textures, the final image comparrision shows that RR enabled just removes all detail from the floor marbling texture, almost like its not just running RR on the lighting but on the full scene and it's denoising textures?
 
Joined
Sep 10, 2018
Messages
7,015 (3.04/day)
Location
California
System Name His & Hers
Processor R7 5800X/ R7 7950X3D Stock
Motherboard X670E Aorus Pro X/ROG Crosshair VIII Hero
Cooling Corsair h150 elite/ Corsair h115i Platinum
Memory Trident Z5 Neo 6000/ 32 GB 3200 CL14 @3800 CL16 Team T Force Nighthawk
Video Card(s) Evga FTW 3 Ultra 3080ti/ Gigabyte Gaming OC 4090
Storage lots of SSD.
Display(s) A whole bunch OLED, VA, IPS.....
Case 011 Dynamic XL/ Phanteks Evolv X
Audio Device(s) Arctis Pro + gaming Dac/ Corsair sp 2500/ Logitech G560/Samsung Q990B
Power Supply Seasonic Ultra Prime Titanium 1000w/850w
Mouse Logitech G502 Lightspeed/ Logitech G Pro Hero.
Keyboard Logitech - G915 LIGHTSPEED / Logitech G Pro
The shadows varying in sharpness might actually be a correction. In real life shadows are not always going to be sharp all the way through. Soft shadows are a feature in state-of-the-art pre calculated 3D renderers.
Soft Shadows in PRMan (pixar.com)
View attachment 314558

People are so use to fake image techniques and non accurate lighting/shadows etc when they see something more accurate it throws them off. I get it though we've been conditiond by 30+ years of developers faking lightning/reflections etc somthing that is more accurate doesn't always look better.

Also this is a free extra toggle for anyone with an RTX gpu that in most scenarios seems to improve both image quality and performance I'll judge it in motion on a 65 inch oled later today but in videos I've seen it's impressive.

At least 50% of gpu owners don't even have an RTX gpu more if we include consoles to even use this so hopefully a more open solution is developed.
 
Joined
Feb 1, 2013
Messages
1,270 (0.29/day)
System Name Gentoo64 /w Cold Coffee
Processor 9900K 5.2GHz @1.312v
Motherboard MXI APEX
Cooling Raystorm Pro + 1260mm Super Nova
Memory 2x16GB TridentZ 4000-14-14-28-2T @1.6v
Video Card(s) RTX 4090 LiquidX Barrow 3015MHz @1.1v
Storage 660P 1TB, 860 QVO 2TB
Display(s) LG C1 + Predator XB1 QHD
Case Open Benchtable V2
Audio Device(s) SB X-Fi
Power Supply MSI A1000G
Mouse G502
Keyboard G815
Software Gentoo/Windows 10
Benchmark Scores Always only ever very fast
People are so use to fake image techniques and non accurate lighting/shadows etc when they see something more accurate it throws them off. I get it though we've been conditiond by 30+ years of developers faking lightning/reflections etc somthing that is more accurate doesn't always look better.

Also this is a free extra toggle for anyone with an RTX gpu that in most scenarios seems to improve both image quality and performance I'll judge it in motion on a 65 inch oled later today but in videos I've seen it's impressive.

At least 50% of gpu owners don't even have an RTX gpu more if we include consoles to even use this so hopefully a more open solution is developed.
Reality is hard, so hard that...
buprestid-beer-bottle-56a51eea3df78cf77286559e.jpg
 
Joined
Jul 13, 2016
Messages
3,370 (1.09/day)
Processor Ryzen 7800X3D
Motherboard ASRock X670E Taichi
Cooling Noctua NH-D15 Chromax
Memory 32GB DDR5 6000 CL30
Video Card(s) MSI RTX 4090 Trio
Storage P5800X 1.6TB 4x 15.36TB Micron 9300 Pro 4x WD Black 8TB M.2
Display(s) Acer Predator XB3 27" 240 Hz
Case Thermaltake Core X9
Audio Device(s) JDS Element IV, DCA Aeon II
Power Supply Seasonic Prime Titanium 850w
Mouse PMM P-305
Keyboard Wooting HE60
VR HMD Valve Index
Software Win 10
"Better than native" -> Proceeds to show a plethora of comparisons where DLSS borks effects like smoke, shadows, DoF or bloom.

Why is the conclusion longer than the rest of the article anyway?

There's loss of fine detail as well. In image 2, the grey pipes on the right hand size and the barrier on the left hand side both loose detail. You can notice the loss of detail in the pipes from a distance without zooming in.

The floor tiles in image 5 looks like plastic with RR enabled and loose a lot of detail compared to the native image. The trash cans too loose a lot of the rust detail.

The foliage highlighted in image 4 has sharpening artifacts and unnaturally stands out. If you look closely there are a lot of sharp black pixels added that makes them stand out where native the pixels transition normally.

You can also see that it appears that some light sources have been disabled. Again in image 2 you can see the light mid left is off with RR enabled and on in native.

RT in general in these screenshots isn't doing a very accurate job, in the 2nd image again it's adding shading under many objects even though in a brightly lit outdoor scene the sun is bouncing off the ground should be hitting those spots. I have to wonder if they are calculating any ray's reflecting at all. I would describe this RT style as Dollhouse, everything looks more fake and unsettling.
 
Last edited:
Joined
Aug 7, 2019
Messages
367 (0.19/day)
The shadows varying in sharpness might actually be a correction. In real life shadows are not always going to be sharp all the way through. Soft shadows are a feature in state-of-the-art pre calculated 3D renderers.
What the hell are you actually talking about? In the very first comparison the wall has a hard shadow on the DLSS side and a progressive soft shadow on the native one.

People are so use to fake image techniques and non accurate lighting/shadows etc when they see something more accurate it throws them off. I get it though we've been conditiond by 30+ years of developers faking lightning/reflections etc somthing that is more accurate doesn't always look better.

Also this is a free extra toggle for anyone with an RTX gpu that in most scenarios seems to improve both image quality and performance I'll judge it in motion on a 65 inch oled later today but in videos I've seen it's impressive.

At least 50% of gpu owners don't even have an RTX gpu more if we include consoles to even use this so hopefully a more open solution is developed.
It's not more accurate. DLSS is making details pop up when it shouldn't, washing surfaces for no reason and sharpening stuff that makes it uncanny.

There's loss of fine detail as well. In image 2, the grey pipes on the right hand size and the barrier on the left hand side both loose detail. You can notice the loss of detail in the pipes from a distance without zooming in.

The floor tiles in image 5 looks like plastic with RR enabled and loose a lot of detail compared to the native image. The trash cans too loose a lot of the rust detail.

The foliage highlighted in image 4 has sharpening artifacts and unnaturally stands out. If you look closely there are a lot of sharp black pixels added that makes them stand out where native the pixels transition normally.

You can also see that it appears that some light sources have been disabled. Again in image 2 you can see the light mid left is off with RR enabled and on in native.

RT in general in these screenshots isn't doing a very accurate job, in the 2nd image again it's adding shading under many objects even though in a brightly lit outdoor scene the sun is bouncing off the ground and hitting those spots. I have to wonder if they are calculating any ray's reflecting at all. I would describe this RT style as Dollhouse, everything looks more fake and unsettling.
The whole DLSS thing is a mess and I honestly don't know how reviewers are praising it and forums swarmed by people with seriously damaged vision.
 
Joined
Oct 28, 2012
Messages
1,195 (0.27/day)
Processor AMD Ryzen 3700x
Motherboard asus ROG Strix B-350I Gaming
Cooling Deepcool LS520 SE
Memory crucial ballistix 32Gb DDR4
Video Card(s) RTX 3070 FE
Storage WD sn550 1To/WD ssd sata 1To /WD black sn750 1To/Seagate 2To/WD book 4 To back-up
Display(s) LG GL850
Case Dan A4 H2O
Audio Device(s) sennheiser HD58X
Power Supply Corsair SF600
Mouse MX master 3
Keyboard Master Key Mx
Software win 11 pro
What the hell are you actually talking about? In the very first comparison the wall has a hard shadow on the DLSS side and a progressive soft shadow on the native one.
You are talking about that one? It's still soft, (compare it to the fence shadow) just not as soft it's also from a moving object since the position is also different.
1695314890248.png
1695314912412.png


Shadows is not the thing that I would criticize with DLSS 3.5, but rather how it's altering some texture when it's fixing the Path tracing. Better trees reflections, but the marble is not marble anymore. (That's DLSS VS DLSS 3.5 btw)
1695315037252.png
 
Joined
Sep 10, 2018
Messages
7,015 (3.04/day)
Location
California
System Name His & Hers
Processor R7 5800X/ R7 7950X3D Stock
Motherboard X670E Aorus Pro X/ROG Crosshair VIII Hero
Cooling Corsair h150 elite/ Corsair h115i Platinum
Memory Trident Z5 Neo 6000/ 32 GB 3200 CL14 @3800 CL16 Team T Force Nighthawk
Video Card(s) Evga FTW 3 Ultra 3080ti/ Gigabyte Gaming OC 4090
Storage lots of SSD.
Display(s) A whole bunch OLED, VA, IPS.....
Case 011 Dynamic XL/ Phanteks Evolv X
Audio Device(s) Arctis Pro + gaming Dac/ Corsair sp 2500/ Logitech G560/Samsung Q990B
Power Supply Seasonic Ultra Prime Titanium 1000w/850w
Mouse Logitech G502 Lightspeed/ Logitech G Pro Hero.
Keyboard Logitech - G915 LIGHTSPEED / Logitech G Pro
You are talking about that one? It's still soft, (compare it to the fence shadow) just not as soft it's also from a moving object since the position is also different.
View attachment 314567View attachment 314568

Shadows is not the thing that I would criticize with DLSS 3.5, but rather how it's altering some texture when it's fixing the Path tracing. Better trees reflections, but the marble is not marble anymore
View attachment 314569

Hub did a pretty good comparison of what it does much better and what it does worse.


Also pretty interesting.

 

DY69SX

New Member
Joined
Jan 29, 2023
Messages
23 (0.03/day)
Well looks like Ray Reconstruction doesn't work for me on my 2060 Super!! Just update CP2077 to Patch 2.0 then I cleanly installed new Nvidia drivers and RR are blanked!! And in Performance way I have 5700x 32gb ram and at 1080p Ultra+Psych RT+DLSS Quality i get +-30 FPS Hue Hue Hue!!
 
Joined
Aug 7, 2019
Messages
367 (0.19/day)
You are talking about that one? It's still soft, (compare it to the fence shadow) just not as soft it's also from a moving object since the position is also different.
I meant fence, sorry. The fence loses all the softness progression on its shadow.
 
Joined
Sep 21, 2023
Messages
30 (0.06/day)
"Better than native" -> Proceeds to show a plethora of comparisons where DLSS borks effects like smoke, shadows, DoF or bloom.

Why is the conclusion longer than the rest of the article anyway?
Unfortunately, scenes are not static. Sun, lights, people, smoke and other things move and change between screenshots. It makes it hard to gauge what is borked vs what has changed in the scene.
 
Joined
Aug 7, 2019
Messages
367 (0.19/day)
Unfortunately, scenes are not static. Sun, lights, people, smoke and other things move and change between screenshots. It makes it hard to gauge what is borked vs what has changed in the scene.
The smoke that covers the building in the second comparison isn't gone, the DLSS just killed it. In the same fashion it added cell-shading to the cocrete pipes or over-sharpened the grid shadow when it's supposed to be soft.
 
Joined
Apr 23, 2020
Messages
12 (0.01/day)
Can someone please clarify something for me:
DLSS 2.0 = Upscale
DLSS 3.0 = Upscale + Frame generation?

I haven't been following this subject for a while. Thanks in advance
 
Joined
Apr 14, 2018
Messages
703 (0.29/day)
Why does better than Native keep getting pedaled in any DLSS review/marketing materials? By definition you cannot create more from less information; textures will never be better than native, not to mention a slew of other artifacts introduced from upscaling techniques.
 
Joined
Jun 14, 2020
Messages
3,559 (2.14/day)
System Name Mean machine
Processor 12900k
Motherboard MSI Unify X
Cooling Noctua U12A
Memory 7600c34
Video Card(s) 4090 Gamerock oc
Storage 980 pro 2tb
Display(s) Samsung crg90
Case Fractal Torent
Audio Device(s) Hifiman Arya / a30 - d30 pro stack
Power Supply Be quiet dark power pro 1200
Mouse Viper ultimate
Keyboard Blackwidow 65%
Well looks like Ray Reconstruction doesn't work for me on my 2060 Super!! Just update CP2077 to Patch 2.0 then I cleanly installed new Nvidia drivers and RR are blanked!! And in Performance way I have 5700x 32gb ram and at 1080p Ultra+Psych RT+DLSS Quality i get +-30 FPS Hue Hue Hue!!
It only works with PT enabled for now.

Why does better than Native keep getting pedaled in any DLSS review/marketing materials? By definition you cannot create more from less information; textures will never be better than native, not to mention a slew of other artifacts introduced from upscaling techniques.
Youl learn to ignore them in time. It's people who bought the wrong GPU and now they pretend they can tell the difference. If you ask them for a blind test they will disappear into the ether.
 
Joined
Sep 10, 2018
Messages
7,015 (3.04/day)
Location
California
System Name His & Hers
Processor R7 5800X/ R7 7950X3D Stock
Motherboard X670E Aorus Pro X/ROG Crosshair VIII Hero
Cooling Corsair h150 elite/ Corsair h115i Platinum
Memory Trident Z5 Neo 6000/ 32 GB 3200 CL14 @3800 CL16 Team T Force Nighthawk
Video Card(s) Evga FTW 3 Ultra 3080ti/ Gigabyte Gaming OC 4090
Storage lots of SSD.
Display(s) A whole bunch OLED, VA, IPS.....
Case 011 Dynamic XL/ Phanteks Evolv X
Audio Device(s) Arctis Pro + gaming Dac/ Corsair sp 2500/ Logitech G560/Samsung Q990B
Power Supply Seasonic Ultra Prime Titanium 1000w/850w
Mouse Logitech G502 Lightspeed/ Logitech G Pro Hero.
Keyboard Logitech - G915 LIGHTSPEED / Logitech G Pro

another in depth comparison of the pros and cons of using it although the person doing the testing says he will be using it at home.
 
Top