• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

More DLSS...

chameleoneel said:
DF does a really good job showing and explaining a few surprising details where AMD or Nvidia, technically look better than the other.

I think the situation with the puddle and the bamboo, looks to me like maybe AMD isn't tracing as many rays per frame. The reflection feels incomplete. Like the Bamboo itself isn't coded to trace as much detail, or something.
Click to expand...
maybe but it took away the "overly wet" look that i really dont like in most rt.
 
Hopefully they'll implement it in Cyberpunk and maybe Alan Wake 2. Both games have really good raytracing and both look quite a bit better with nVidia RR in my opinion. So they would be a good comparison point to get a feel for more of the technology.

pendragon1 said:
i like the amd look better, its wet concrete not a deep puddle so it shouldnt reflect as much as nv's side.
Click to expand...
Well the question is what do the designers want it to look like? Which one is getting closer to what they are actually trying to render? Games often make reflective surfaces overly shiny because it is eye catching, even though it isn't accurate.
 
Does FSR hold a candle to DLSS 4? A good friend is considering a 9070XT for 1440P. Trying to figure out if I should try and convince him to go for Nvidia instead. He may be able to swing a 5070ti or 5080 since he is getting most of the rest of his build for xmas. Thoughts?
 
Blackstone said:
Does FSR hold a candle to DLSS 4? A good friend is considering a 9070XT for 1440P. Trying to figure out if I should try and convince him to go for Nvidia instead. He may be able to swing a 5070ti or 5080 since he is getting most of the rest of his build for xmas. Thoughts?
Click to expand...
FSR 4 is really good, DLSS 4 is really good. In the Last Of Us Part 1, DLAA, DLSS any setting had severe motion streaking, not sure if issue with my 5090 setup or what, other games don't have this problem or is not noticeable. My 9070 XT with FSR 4 was outstanding, no streaking on motion, awesome motion clarity! On the 5090 I would get periodic pausing, subtle but there, on the 9070 was smooth throughout. Ended up playing most of the game on the 9050XT, it was the better experience. Rare occasion maybe but my experience.

Seen 9070XTs lately for $599, that is a steal and a better value in my opinion over either the 5070 or 5070Ti.
 
Blackstone said:
Does FSR hold a candle to DLSS 4? A good friend is considering a 9070XT for 1440P. Trying to figure out if I should try and convince him to go for Nvidia instead. He may be able to swing a 5070ti or 5080 since he is getting most of the rest of his build for xmas. Thoughts?
Click to expand...
yes enough that it is more about game support difference than quality difference (does the game he play actually has FSR 4 being an important question here), historically DLSS has larger support base
 
Last edited:
Dlss4 is a noticeable bit ahead of fsr4, but I agree it depends more on what games he plays and their support. Very few games have fsr4 and it's been slow in adoption. Also don't forget Nvidia's frame gen is very well refined at this point and in a lot of games.
 
noko and Luke are correct, the only thing I'll note is that there's way less games that support FSR4 than DLSS. It also isn't easy to override to the latest version in driver, as it is with DLSS, so net effect is you have more games that you are playing with FSR3 or 2 and those are not nearly as good.

In general the rule is that DLSS 4 aka "Transformer Model" is the best, but only slightly better than FSR 4, both are very good. Both are better than DLSS 2/3 aka "CNN Model" though it is still very good. It is quite a bit better than FSR 3, which is generally usable in quality mode but not less than that and not always. That is better than FSR 2, which is sometimes so garbage it really isn't usable and that is better than FSR 1 which is pretty much trash.

However, as they mentioned, implementation matters, some games are notably better or worse at one of them. Like Elder Scrolls Online got DLSS 2 and DLAA back in the day but it is a TRASH implementation and has massive ghosting everywhere, way worse than its native TAA (which is also not very good). In general though, you are going to find more games that support a newer version of DLSS, and those can usually be overridden to use DLSS 4 in the driver. FSR 4 support is more sparse at this point.
 
Blackstone said:
Does FSR hold a candle to DLSS 4? A good friend is considering a 9070XT for 1440P. Trying to figure out if I should try and convince him to go for Nvidia instead. He may be able to swing a 5070ti or 5080 since he is getting most of the rest of his build for xmas. Thoughts?
Click to expand...
Not really, but he should get a 9070 non-XT anyway because it’s a lot cheaper. In 2025, a midrange card is a midrange card and it’s going to give you a midrange experience whether you pay $520 for a 9070 non-XT or $930 for a 5080. The only real upgrade from any midrange card is an RTX 5090. That’s the only card that can enable things like path tracing or hit super high refresh rates, and small differences in performance don’t matter as long as you have 16 GB VRAM and enough raster horsepower to run max settings at 1440p, which the 9070 does. Anything more expensive than the 9070 but less powerful than the 5090 is just a waste of money, IMO, even though it’s nicer to have DLSS 4 than FSR 4.
 
DLSS 4.5 is out.
https://www.nvidia.com/en-us/geforc...i-frame-gen-6x-2nd-gen-transformer-super-res/


Nvidia App should auto prompt to update itself, and then serve you a new video driver.

Two presets for 4.5, which you can force onto most games. L and M

L is a heavier algorithm. M is lighter weight. But, still takes a bit more resources than DLSS4.


I have tried it in Expedition 33 and noticed a cleaner, but slightly too sharp image.
Color depth seems slightly off. Maybe because of the over sharpness. Color issue reminds me of DLSS 2.0

Less temporal stability. But, better clarity and scenes with particles floating around, those particles look a lot better. DLSS4 would tend to omit some of them from the image and overtabilize the ones you could see.

Disocclusion artifacts are noticeably cleaned up.

----------------------

1767685108408.png
 
I have only tried it in Silent Hill 2 and it looks better with DLSS 4.5 on Performance than it does using Quality on the DLSS version the game ships with.
 
These new algorithms are indeed pretty heavy.

On the Monster Hunter Wilds bechmark (which you can't download anymore) which gives you a final average FPS over the whole run:

I found a 13 frame difference between DLSS 4.5 L and 3.0 C. That's a huge dip, for this benchmark. I've never lost that much from any messing with settings or hardware**. There is one portion in particular, which is very GPU heavy. And it actually dipped below 60fps on that section.

4.5 M gave me back 2 frames, for a 10 frame difference between DLSS 3.0 preset C

DLSS 4.0 K is a 5 frame difference. Which I did know about, already.


- 4070 TI Super at 1440P, DLSS Quality.

Part of the trick I suppose, is that the new models can potentially look so much better, that a lower DLSS setting can still look better than an older algorithm, on a higher setting.


**Previously, the biggest FPS change I got in this benchmark, was activating 200S Boost on my 265K build. Gained 7fps on the final average. Which actually beats my 9800X3D build, by about 2 frames. (using same video card).
That's a one-off though. MH:Wilds likes Arrowlake. The X3D usually handily beats it in most games.
 
Last edited:
chameleoneel said:
These new algorithms are indeed pretty heavy.

On the Monster Hunter Wilds bechmark (which you can't download anymore) which gives you a final average FPS over the whole run:

I found a 13 frame difference between DLSS 4.5 L and 3.0 C. That's a huge dip, for this benchmark. I've never lost that much from any messing with settings or hardware**. There is one portion in particular, which is very GPU heavy. And it actually dipped below 60fps on that section.

4.5 M gave me back 2 frames, for a 10 frame difference between DLSS 3.0 preset C

DLSS 4.0 K is a 5 frame difference. Which I did know about, already.


- 4070 TI Super at 1440P, DLSS Quality.

Part of the trick I suppose, is that the new models can potentially look so much better, that a lower DLSS setting can still look better than an older algorithm, on a higher setting.


**Previously, the biggest FPS change I got in this benchmark, was activating 200S Boost on my 265K build. Gained 7fps on the final average. Which actually beats my 9800X3D build, by about 2 frames. (using same video card).
That's a one-off though. MH:Wilds likes Arrowlake. The X3D usually handily beats it in most games.
Click to expand...
10fps dip out of how many frames? What percentage?
 
So far reading various places online, if you use DLAA / Quality stick with preset K, if you use balanced and below, preset L & M's aggressiveness can be useful. I guess some people are showing that at L & M on Quality or higher setting, the sharpness is a bit overaggressive.

Honestly, in looking at comparisons so far, my eyes are having a tough time making out any differences, but it does sound like a tradeoff is L & M are much better with ghosting. Guess I will have to test and see.

Way back though, I was able to spot instantly the difference in K from the older models.
 
Auer said:
With everything including RTX maxed out, that's about where the 2070 taps out.
Click to expand...

Realistically speaking, it's where most cards tap out unless you can justify the cost of halo cards. When it comes to ray tracing, especially path based ray tracing, 1080p will get you the eye candy without spending a fortune while maintaining adequate image quality.
 
My 5070 gets ~60+FPS @ 4K in Hellblade 2 max settings with quality upscale and 2x frame gen turned on
 
kram182 said:
My 5070 gets ~60+FPS @ 4K in Hellblade 2 max settings with quality upscale and 2x frame gen turned on
Click to expand...

The problem is: If you're getting ~60fps with FG enabled, you can bet your bottom dollar that your real base fps is well below 50fps - Which is far from ideal.

I ditched my 4k display (it's on the floor behind me) for 2 x 1200p displays - I run an RTX 4070S, which basically performs on par with a 5070, and I'm loving path based RT with DLSS and FG 2x enabled at 1200p at ~90+ fps.
 
Mazzspeed said:
The problem is: If you're getting ~60fps with FG enabled, you can bet your bottom dollar that your real base fps is well below 50fps - Which is far from ideal.

I ditched my 4k display (it's on the floor behind me) for 2 x 1200p displays - I run an RTX 4070S, which basically performs on par with a 5070, and I'm loving path based RT with DLSS and FG 2x enabled at 1200p at ~90+ fps.
Click to expand...

Base/true/raster frame rate runs between ~30+-45FPS now.

I already played through the game before it got a recent big update including for graphics, was about the same then but more closer to staying around ~45+FPS than closer to ~30+FPS now IIRC (but I still played with framegen to keep graphics maxed and stay above monitor GSYNC/FreeSYNC lower limit of 48FPS). I didn't have unplayable problems with latency when I did play it, I could notice it especially when dipping towards 30s but wasn't unplayable just took a little to adjust after I did notice it and then you forget it exists. Usually at native 30FPS the disconnect between framerate and input latency is too much for me it feels like punching in a dream or something, around native 45FPS and I can't notice anymore myself really vs 60FPS+
 
Even if you find the latency to not be a problem with a base frame rate in the 30s before turning on frame gen, you will still massively increase the chances of artifacts. I've seen so many people have base frame rates in the 30s and claim their game looks great with frame gen when in reality it's just loaded with visual issues.
 
jobert said:
Even if you find the latency to not be a problem with a base frame rate in the 30s before turning on frame gen, you will still massively increase the chances of artifacts. I've seen so many people have base frame rates in the 30s and claim their game looks great with frame gen when in reality it's just loaded with visual issues and anomalies.
Click to expand...

Depends on the game in my experience. Hellblade I didn't have any that I noticed when playing so left it on. Last of US remake for example I turn frame gen off (played @ 1440p so was able to hit above monitor frame limit fine with frame gen off, just got a new 4K monitor so haven't ran that one yet @ 4K) because I notice with frame gen on, specifically in my case any patterned carpet when you're inside, it would artifact/image trail on the cross hatch pattern on the carpet and was too distracting. I wouldn't be surprised if there's always artifacts no matter what for sake of argument, but like latency I think it might get subjective per game on what is artifacting, can you notice it when playing vs just pixel peeping, etc.

Granted LoU remake is AMD frame gen too IIRC?

Also my Hellblade 2 FPS when I initially played through it with closer to ~45FPS+ was also when I was still @ 1440p vs now @4K
 
Last edited:
Yes, it definitely depends highly on the game. In avatar frontiers of Pandora I see artifacting with frame gen on if my base frame rate is in the low 50s even. Adding 3x or 4x frame generation with base frame rate in the low 50s in that game produces absolutely laughable amounts of artifacting. I can't even imagine having a base frame rate in the 30s for that particular game.
 
So tried this DLSS 4.5 out on SOTTR and Diablo IV tonight. Much more testing to come obviously, but I normally use DLAA, 4K, max settings. I did lose some performance with Preset M.

In SOTTR is didn't matter, the 5090 slays that game. Numbers were lower, but still well above 100 fps.

In Diablo IV (Which has a ton of RT), DLAA was pushed hard, even with FG. What i did instead was goto Quality DLSS with FG on and my frames were locked solid a 144. Latency was only like 12ms as well, and it felt super smooth. 5090 only used 320W instead of my normal 475W+ in that game, and temps were like 47C.

What i found (so far, more testing to come) was DLSS Quality M looked indistinguishable from DLAA Preset K for my eyes.

The key next will be for me to try BF6 and Forza 5 and see if I notice better clarity in motion even if that means I drop to Quality from DLAA as IQ looks the same.
 
chameleoneel said:
The point of DLSS 4.5, is to try Performance mode. And probably have better image quality and a bit more frames, than 4.0 at Quality

Frane gen also got a quality and consistency boost.
Click to expand...

Yeah if it gets to the point of lower DLSS quality for me now I might as well lower actual game settings IMO (up to now, if I can't notice a difference myself I'd rather just lower DLSS settings assuming I can't notice so we'll see - I haven't tried the new upscaler yet)
 
Frame-gen is objectively much smoother now, especially at 4x.

When I initially bought my 5090, there were a few games that weren’t running well maxed out at 4K. When I enabled 4x frame-gen, my frames definitely spiked, but the games didn’t really feel much smoother. Basically, it wasn’t worth it.

That’s just not the case any longer. Now, if a game is running sub 60fps (say, 45 fps or so) I can enable 4x frame-gen and most of the games feel relatively smooth.

The one game that has truly benefited from this tech is Avatar Frontiers of Pandora at 4K resolution with the graphics set to Unobtanium: enabling 4x frame-gen, in this case, is essential for smooth performance. The 5090’s AI tech, in this case, is staggering.
 
Hershy said:
Frame-gen is objectively much smoother now, especially at 4x.

When I initially bought my 5090, there were a few games that weren’t running well maxed out at 4K. When I enabled 4x frame-gen, my frames definitely spiked, but the games didn’t really feel much smoother. Basically, it wasn’t worth it.

That’s just not the case any longer. Now, if a game is running sub 60fps (say, 45 fps or so) I can enable 4x frame-gen and most of the games feel relatively smooth.

The one game that has truly benefited from this tech is Avatar Frontiers of Pandora at 4K resolution with the graphics set to Unobtanium: enabling 4x frame-gen, in this case, is essential for smooth performance. The 5090’s AI tech, in this case, is staggering.
Click to expand...
If anything Avatar Frontiers of Pandora has proven to be the absolute worst frame gen experience I have ever seen if running below 60 fps. Even at 2x I could see quite a bit of artifacting and at 4x it was laughably bad.
 
Last edited:
polonyc2 said:
DLSS 4.5 Ray Reconstruction | Updated with 2nd Gen Transformer
Click to expand...
wantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwant.

:D

This has be excited since I'm such a RT fanboy and Ray Reconstruction is amazing, but it could use improvements for sure. It'll be interesting to see how much heavier it is going to be. Guess I'll have to leave Cyberpunk installed just to see how it looks.
 
Sycraft said:
It'll be interesting to see how much heavier it is going to be.
Click to expand...
If you have modern (lovelace and up) Tensor core it could be very small performance impact, at least that Nvidia claim:
  • Efficient Denoiser: The new model delivers 35% more compute capability, and processes 20% more parameters, while maintaining similar performance to the previous model.
Which I would assume only true if you have good 8 bits support, a bit like at launch dlss 4.5 performance cost was quite small on a 4070 super or 5070 gpu while very large on a 3090.
 
Sycraft said:
wantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwantwant.

:D

This has be excited since I'm such a RT fanboy and Ray Reconstruction is amazing, but it could use improvements for sure. It'll be interesting to see how much heavier it is going to be. Guess I'll have to leave Cyberpunk installed just to see how it looks.
Click to expand...

Guess I'll be doing another playthrough of RE9. I forced the Gen 1 Transformer model for RR through Nvidia Inspector but it still had some of that typical path traced noise. Excited to see how much cleaner it's going to look now.
 
LukeTbk said:
If you have modern (lovelace and up) Tensor core it could be very small performance impact, at least that Nvidia claim:
  • Efficient Denoiser: The new model delivers 35% more compute capability, and processes 20% more parameters, while maintaining similar performance to the previous model.
Which I would assume only true if you have good 8 bits support, a bit like at launch dlss 4.5 performance cost was quite small on a 4070 super or 5070 gpu while very large on a 3090.
Click to expand...
I do, I have a 5090 so it has good 8-bit and 4-bit performance. Of course it is notable that even on that, DLSS 4.5 hits harder than 4. It's perfectly acceptable and usable, in fact I like to us the L preset which hits even harder, but you do take a bigger FPS hit than using M, even on the chonk that is the 5090.

So it wouldn't surprise me if this new RR hits harder too, even on hardware with FP 8 acceleration.
 
III_Slyflyer_III said:
So tried this DLSS 4.5 out on SOTTR and Diablo IV tonight. Much more testing to come obviously, but I normally use DLAA, 4K, max settings. I did lose some performance with Preset M.

In SOTTR is didn't matter, the 5090 slays that game. Numbers were lower, but still well above 100 fps.

In Diablo IV (Which has a ton of RT), DLAA was pushed hard, even with FG. What i did instead was goto Quality DLSS with FG on and my frames were locked solid a 144. Latency was only like 12ms as well, and it felt super smooth. 5090 only used 320W instead of my normal 475W+ in that game, and temps were like 47C.

What i found (so far, more testing to come) was DLSS Quality M looked indistinguishable from DLAA Preset K for my eyes.

The key next will be for me to try BF6 and Forza 5 and see if I notice better clarity in motion even if that means I drop to Quality from DLAA as IQ looks the same.
Click to expand...
320 watts seems to be sweet spot for power to heat output on the 5090. Haven't had much time to tinker for a couple weeks but I like that power to performance ratio and I'll be keeping it somewhere near that for most games. i hate the summer heat and adding more power just doesn't appeal to me at all. Will be testing more when time permits.
 
You must log in or register to reply here.
Back
Top