591.74 New DLSS 4.5

killroy67

killroy67

2[H]4U
2FA
Joined
Oct 16, 2006
Messages
2,347
New 591.74 drivers supporting the new DLSS 4.5 and the new 2nd gen transformer model with Dynamic MFG 6X come Spring.
To get DLSS 4.5 you will need to opt into the beta Nvidia App, or wait until the final release on Jan 13th.
 
These seem ok haven't launched a game yet. Yeah these are ok I was getting hitches in BL4 but I just had to change my Rez back to normal either that or it's my ifi dac to speakers.

20260106120342_1.jpg

I'll drink to these CES 2026 drivers if you can still afford hardware that is.
 
Last edited:
6x FG seems wholly unnecessary to me. If your base framerate is low enough to want this, it'll be unusable with the input latency.
 
From discourse I'm finding online it seems to be pretty heavy performance penalty on the 3000 & 2000 series. Would have been insane fine wine for those cards if not for that
 
Comixbooks said:
These seem ok haven't launched a game yet.
Click to expand...
I launched Stalker 2, BF6 and BL4, seems fine but I'll actually play later this evening.
MavericK said:
6x FG seems wholly unnecessary to me. If your base framerate is low enough to want this, it'll be unusable with the input latency.
Click to expand...
It depends on if you have a good mouse that has low latency, I run a wired Razer Deathadder V3 and use MFG 3X, I don't notice any latency.
 
killroy67 said:
It depends on if you have a good mouse that has low latency, I run a wired Razer Deathadder V3 and use MFG 3X, I don't notice any latency.
Click to expand...
It has nothing to do with your mouse (unless you have a really shitty one), and everything to do with your base framerate. You have a 5080 so yeah, your base framerate is probably pretty good. I have a 5080 and also use FG, but if I turn up the settings too high in Stalker 2 I can definitely feel the input lag when base framerate drops.
 
MavericK said:
It has nothing to do with your mouse (unless you have a really shitty one), and everything to do with your base framerate. You have a 5080 so yeah, your base framerate is probably pretty good. I have a 5080 and also use FG, but if I turn up the settings too high in Stalker 2 I can definitely feel the input lag when base framerate drops.
Click to expand...
I run MFG 3X on Stalker and at that setting it seems fine, I haven't tried 4X, i use 3X on BL4 as well with no issue.
 
MavericK said:
It's not specifically about 3x, 4x etc, just FG in general even at 2x will add a latency hit. There is an additional latency hit for 3x and 4x as you go up, but less so than the initial loss from enabling FG in the first place.
Click to expand...
I may not notice any latency due to using a wired mouse, which I read was the best option due to it's direct connection, one of these days I'll have to try 4X and see how that is.
 
Installing these gave me a first. Changed my 3x monitors to surround(which i only enable when sim racing) and nvidia control panel crashes any time I try to undo Surround. Typically have zero issues.

Edit: unplugged one monitor, rebooted, all well again.
 
Slightly off topic, but I've never understood the need for FG outside compeititve play or whatever. Don't you already need to be pushing near 60 to use it? If so, what's the point? I've used DLSS upscaler, and uderstand that use case... but not FG.
 
toddw said:
Slightly off topic, but I've never understood the need for FG outside compeititve play or whatever. Don't you already need to be pushing near 60 to use it? If so, what's the point? I've used DLSS upscaler, and uderstand that use case... but not FG.
Click to expand...
Some people like the extra image smoothness that comes with going to higher fps. In some cases it does add a cool aesthetic, but it’s highly ymmv and the hate against it is because Nvidia has been touting it like a performance gain when it’s most definitely not.

Framegen is a disadvantage in competitive gaming as it introduces latency, artifacts (though seemingly less with 4.5) and the frames are estimations instead of actualities. You’re better off going 120fps non gen vs 220fps frame gen if competitive gaming.
 
toddw said:
Slightly off topic, but I've never understood the need for FG outside compeititve play or whatever. Don't you already need to be pushing near 60 to use it? If so, what's the point? I've used DLSS upscaler, and uderstand that use case... but not FG.
Click to expand...
In the few games I've used it, I def prefer to have it on if I'm around 60 FPS native. Go play Portal RTX, which will bring a 5090 to below 30 FPS native at 4K. So you basically need DLSS upscaling at balanced or so with FG to make it an enjoyable experience. That's what I had to do except at 1440 UW on my 5080 and it was much more enjoyable at 3-4x FG than it was hovering around 60 without FG for me.

Pretty much the same with CP2077 too. If it's not a fast-paced twitch shooter or something, then I really don't notice the latency difference enough to keep it off. I tested for artifacting a lot too and couldn't really find anything that was due to FG, it was all due to upscaling. So for me, FG is a legit feature to have so far.
 
toddw said:
Slightly off topic, but I've never understood the need for FG outside compeititve play or whatever. Don't you already need to be pushing near 60 to use it? If so, what's the point? I've used DLSS upscaler, and uderstand that use case... but not FG.
Click to expand...
It's really for when you're around 60-80 FPS and you have a high refresh rate monitor, so you want to be up near 120-200 FPS. It does make a noticeable difference in smoothness in that regard, but you have to have a decent base framerate in order to not get significant input latency.
 
Seems to work really well in Stalker 2 where before there were a lot of artifacts with foliage and swirling leaves and stuff.
 
As someone who jumped from a 9070 XT to a 5090, I used to use AFMF/FG in 90% of my games with no issues, now I turn off FG practically everywhere because of artifacts.
I'm going to test the new DLSS 4.5 and hopefully it will improve the FG problem...
 
https://www.nvidia.com/en-us/geforc...-cards/5/580689/dlss-45-super-resolution-faq/

1767827475139.png


I've tested in quite a few games now. I'm very happy with the performance and image quality. Decided I'll be using DLSS Performance with Model M and MFG now at 4X in all my games. This model really improved multiframe gen and the ghosting around UI elements. Locked 240fps is beautiful. The improvement to particles in Model M is astounding, really pops with HDR.
 
DLSS is old enough that I want to try using something like DLSS Swapper for games with ancient .dll files of DLSS, but im totally confused on what model M, L, K, J, whatever are. I almost always use Quality if not DLAA/DLDSR. Which model would you use for that type of quality @ 4k rez? (RTX 5090).
 
DF-1 said:
DLSS is old enough that I want to try using something like DLSS Swapper for games with ancient .dll files of DLSS, but im totally confused on what model M, L, K, J, whatever are. I almost always use Quality if not DLAA/DLDSR. Which model would you use for that type of quality @ 4k rez? (RTX 5090).
Click to expand...
For dlaa/quality use model K.
 
pandora's box said:
I've tested in quite a few games now. I'm very happy with the performance and image quality. Decided I'll be using DLSS Performance with Model M and MFG now at 4X in all my games. This model really improved multiframe gen and the ghosting around UI elements. Locked 240fps is beautiful. The improvement to particles in Model M is astounding, really pops with HDR.
Click to expand...
You think it's worth using Performance over Quality?
 
toddw said:
Slightly off topic, but I've never understood the need for FG outside compeititve play or whatever. Don't you already need to be pushing near 60 to use it? If so, what's the point? I've used DLSS upscaler, and uderstand that use case... but not FG.
Click to expand...
depends on the use case, people that game at 4k 120hz appreciate being able to lock in better fps. My PC is also a HTPC hooked up to my oled which supports 120hz and vrr so I'll take FG to help smooth it out and keep it closer to 120hz. Like you said I tweak settings to try and get my 60fps min and let frame gen take it from there, been extremly happy with how it works already so I'm actually very excited for this new version.

I think one of the big selling points of DLSS 4.5 is adaptive frame gen that can be forced on for any dlss game making it so you can target a certain fps (say 120 in my case) and it will scale the FG to keep it there.
 
Last edited:
pandora's box said:
I do. Unless you’re pixel peeping instead of playing the game ;)
Click to expand...
Are you on 1440p or 4k? Don’t know if you answered it before but just wanted to confirm.
Reason to ask is with dlss4 found that at 1440p dlss balanced was the min setting I’d use for image quality not distractingly worse than native. At 4k with the added pixels found that performance (or one setting lower than balanced) was the min. This was more noticeable in high image quality games at lower fps, so below 120fps.

With the new dlss 4.5 if the settings could be bumped one step lower while maintaining or even slightly improving image quality then the compute penalty will be worth it on 40 and 50 series cards. Should get a nice fps or power efficiency bonus.
 
In CP 2077, my choices are essentially DLSS performance with PT, DLSS quality with RT or native without RT on my 5070 ti, with quite a few tweaks, that are barely noticable in image quality. With the TF model I generally prefer PT with DLSS performance as the detail loss isn't that much compared to DLSS quality and removing all RT kind of ruins the mood of the game when you are used to RT in it so 4k native is out of the question. The PT just makes the lighting look more natural in a lot of scenes and it is kind of hard to unsee it once you have used it for a few hours.

Did try the new 4.5 with M model in CP 2077 and results are quite impressive. Performance seems to be close to the same as the K model, but the image is a lot cleaner. The difference between DLSS performance(1080p) and quality(1440p) @4k doesn't seem to be as large as before, mainly a little bit more sharpness and fine detail. Of course a native 4k DLAA image is noticeable sharper than using any DLSS, but good luck with running that with PT. FG seemed to be a bit more stable and didn't pick up on any artifacts with a bit of running around. Rain was the main issue with FG in CP 2077 though, so will have to test it more, but initial impressions is that the Nvidia software stack is getting even more impressive.

I went with the Nvidia 5070 ti over AMD 9070 XT due to the software stack (DLSS 4 TF etc.) and version 4.5 feels like a massive "free" upgrade.

toddw said:
Slightly off topic, but I've never understood the need for FG outside compeititve play or whatever. Don't you already need to be pushing near 60 to use it? If so, what's the point? I've used DLSS upscaler, and uderstand that use case... but not FG.
Click to expand...
Frame gen works OK if your base FPS is around 45-50 range and makes the rendering much smoother. The cost is higher input latency and some artifacts. Some artifacts may not be noticeable in gameplay while others can be quite noticeable. If you run around 55 fps without FG then you can usually get around 100 fps with 2x FG, with a rendering smoothness that you will not get unless you are in the 75+ FPS range. FG is a tradeoff and I often leave it off, but it is a nice option to have and I can absolutely understand why a lot of people love it for SP games.
 
yinzerniner said:
Are you on 1440p or 4k? Don’t know if you answered it before but just wanted to confirm.
Reason to ask is with dlss4 found that at 1440p dlss balanced was the min setting I’d use for image quality not distractingly worse than native. At 4k with the added pixels found that performance (or one setting lower than balanced) was the min. This was more noticeable in high image quality games at lower fps, so below 120fps.

With the new dlss 4.5 if the settings could be bumped one step lower while maintaining or even slightly improving image quality then the compute penalty will be worth it on 40 and 50 series cards. Should get a nice fps or power efficiency bonus.
Click to expand...
Running at 4K here.
 
DLSS is fantastic for taking a game that normally runs around 50-60fps and pushing it to 80+. Depending on the game, sometimes it can take a game from just under 60 to 100 or more. That's at 4K.
Sometimes it looks like hazy ass or does have input lag, but most of the time it's pretty fantastic for that scenario. It's one I encounter pretty frequently with my 4090 with the details cranked, too. In most cases, I don't really even notice it's on. Menus tend to be an area that suffer more than most others, and luckily most games don't use them super often.
No clue about those higher multipliers (I'm capped at 120Hz), but I appreciate how the base technology has gotten better and better.
 
Domingo said:
DLSS is fantastic for taking a game that normally runs around 50-60fps and pushing it to 80+. Depending on the game, sometimes it can take a game from just under 60 to 100 or more. That's at 4K.
Sometimes it looks like hazy ass or does have input lag, but most of the time it's pretty fantastic for that scenario. It's one I encounter pretty frequently with my 4090 with the details cranked, too. In most cases, I don't really even notice it's on. Menus tend to be an area that suffer more than most others, and luckily most games don't use them super often.
No clue about those higher multipliers (I'm capped at 120Hz), but I appreciate how the base technology has gotten better and better.
Click to expand...
Are you taking about dlss frame gen or dlss upscaling? I know it’s confusing on the part of both being classified as the same technology even though they perform completely different functions, but would help in verifying your claim that it’s only good for taking games upwards from a base fps of 60
 
yinzerniner said:
Are you taking about dlss frame gen or dlss upscaling? I know it’s confusing on the part of both being classified as the same technology even though they perform completely different functions, but would help in verifying your claim that it’s only good for taking games upwards from a base fps of 60
Click to expand...

Both. I'm not saying it's only good in boosting your FPS above 60, but it's definitely good for that. It can work the same way to take your FPS from 40'ish to 60 as well. My wife does that with her 3090 setup that's capped at 4K/60. The upscaling works great for essentially running a game that (mostly) looks like 4K, but is really running at 1440p or 1080p. The frame generation function works almost like free extra FPS for whatever resolution, but it's situational from game to game. Sometimes it looks just like "real" FPS and sometimes not. I find that it works best when you're playing something that's pulling around 60-70% of your maximum refresh rate and doesn't quite look ideally smooth.
 
Although image quality is increased, the performance uplift is smaller and in some cases minimal over native. A big reason to use DLSS was the large performance jump, that seems to be gone. Of course you can always set the older version for newer games if you don't think the improved image quality is worth it. But at this point you may have to compare the visual quality/performance with other AA methods. I know some of those have their own blurring downsides some maybe DLSS 4.5 might still be a better option in some cases. At least the VRAM usage looks similar though.
 
apparently 4.5 is for performance level of DLSS, not so much for quality level, DLSS 4.5 can look a bit better than 4.0 quality in some case, you still have good performance boost specially on lovelave and up for its optmised for scenario.
 
Flogger23m said:
Although image quality is increased, the performance uplift is smaller and in some cases minimal over native. A big reason to use DLSS was the large performance jump, that seems to be gone. Of course you can always set the older version for newer games if you don't think the improved image quality is worth it. But at this point you may have to compare the visual quality/performance with other AA methods. I know some of those have their own blurring downsides some maybe DLSS 4.5 might still be a better option in some cases. At least the VRAM usage looks similar though.
Click to expand...
There is a significant difference in rendering time between the models on low end and lower mid range GPUs so the M model is probably not that appropriate when trying to run max framerates, but it is good when you want to push the FPS into around 60-70 FPS. Probably not so good, unless you have a 5090, when you are trying to push 120+ fps.

Which model to use probably depends a lot on the GPU and the frame rates one is targeting. Did some quick benchmarking and the K model is around 5-6% faster than the M model on my 5070 ti in CP 2077 in the built in benchmark with RT. Without RT it is around 8% faster when running the heavier GPU loads while that difference will move to around 13% faster when pushing close to 150fps on the K model, using DLSS performance. Keep in mind that DLSS performance at 4k renders at 1080p, which has around 25% more pixels than the render resolution of DLSS quality at 1440p.

Native rendering doesn't get close at all when you are at 4k. At worst the DLSS quality on M model is 25% faster than native with no RT and that increases to 65% or higher when doing the comparison with RT on.

Their results in CP 2077 would probably look quite different if they turned on RT. Their results all show much larger differences when trying to render 100+ FPS compared to when they are below 90 FPS. Based on what I have seen and their results, I would conclude that the lower the frame rate you are targeting, the less the impact of running the M model.
 
LukeTbk said:
apparently 4.5 is for performance level of DLSS, not so much for quality level, DLSS 4.5 can look a bit better than 4.0 quality in some case, you still have good performance boost specially on lovelave and up for its optmised for scenario.
Click to expand...
Yeah, I think this wasn't well communicated by Nvidia, but this is their recommendation to game devs use the new 4.5/M model on Performance mode DLSS, and J/K for for Quality and Balanced DLSS, and L for Ultra Performance mode. These recommendations make sense given that the new model is more expensive than the previous ones, and if we're generally getting better image quality from a low native resolution image on Performance mode than on the higher native resolutions on Balanced/Quality on the previous models, then the trade-off is def worth it.

It would be nice if they annotated these recommendations in the DLSS app if they're going to give us the granularity for us to force the DLSS models with specific games.

Citation:

View: https://youtu.be/36zbGPECzDI?si=p3Qnqe_Vv9CpaqVJ&t=913

Edit: Just saw this guy just posted this saying exactly what I just said as well while showing performance and image quality differences on a 2060 Super using the K and M model:


View: https://www.youtube.com/watch?v=TLl3BeKpbZQ
 
Last edited:
You must log in or register to reply here.
Back
Top