NVIDIA GeForce 511.23 WHQL driver download & Discussion

Those are different scaling methods so I wouldn't be comparing them directly. On other hand I wonder about sharpening, if someone with new driver wants to use just sharpening without scaling, would that reduce framerate by larger margin than older driver?

 

Instant black screen during install this driver

 

It doesn't "look exactly the same" at all. 2.25x DL does look pretty close to the old 4x while the old 2.25x never looked anywhere close to that.

 

'Lossless Scaling' uses the official 'NVIDIA Image Scaling platform-agnostic open source SDK' for NIS support.

The NVIDIA Image Scaling SDK provides a single spatial scaling and sharpening algorithm for cross-platform support. The scaling algorithm uses a 6-tap scaling filter combined with 4 directional scaling and adaptive sharpening filters, which creates nice smooth images and sharp edges. In addition, the SDK provides a state-of-the-art adaptive directional sharpening algorithm for use in applications where no scaling is required - https://github.com/NVIDIAGameWorks/NVIDIAImageScaling​

Could you explain why the driver NIS and NIS official SDK are 'different scaling methods' when used on Maxwell and Pascal based GPUs?

 

Actually even DLDSR 1.78x looks much more sharpened than old DSR 4x with smoothness 0. They really should have made the smoothness slider control texture sharpen with DLDSR, I find the effect way too strong. Might look ok with trash resolution textures like in RDR2, but in e.g. Witcher 3 it's really way too much.

 

This feature has nothing to do with bandwidth, or 60hz.

 

You had me doubting myself so I just ran a series of controlled tests.

IL-2 BOS, 3440 x 1440 100 MHz Gsync Monitor, RTX 2070 Super

I started with legacy 4x DSR, and pixel peeping at 200x, it's the clear winner, but it crushed my FPS down to 25-27 with significant stutters.

Then I did the legacy 2.25x DSR. It brought the FPS up to 42-44, and pixel peeping at 300x (to see it at the same scale as the 4x DSR image at 200x) I can see more evidence of aliasing pixelation.

Lastly I tried the new 2.25x DLDSR and again the FPS was 42-44. Layering that image in Photoshop over the one from the 2.25x DSR run confirmed my earlier observation that they were pixel-for-pixel matches.

Running at my native 3440 x 1440 I get 75-80 FPS, but there is a fair amount of aliasing shimmer from the more obvious jaggies. Those are reasonably suppressed by using either the old SGSSAA, or DSR/DLDSR, but with a significant hit to FPS.

At least for IL2-BOS, on my system, I'll stand by my original observation that DLDSR is not an enhancement over legacy DSR (yet).

p.s.
As far as the Nvidia downscaling goes, to observe each of these live in game at 3440 x 1440, my experience is that 4x DSR has almost no shimmering and angled horizon lines are dead smooth, while the level observed of artifacting with both legacy 2.25x DSR and 2.25x DLDSR is perceived as being exactly the same as each other, equally more aliased. With native 1440 the shimmering as you move through the scene is even worse, and becomes quite noticeable.

 

It worked perfectly fine for me. Also, can you elaborate on the "fundamentally flawed" part?

 

You always send native res to the display. That's what DSR is. It's downsampling to your native monitor res. It places no requirements on your display whatsoever.

 

This doesn't make sense to me -- even if this is so, shouldn't the driver be smart enough to know when two simple conditions are met? E.g. #1) Application is running at native output resolution (the control panel itself shows that it knows what's native under the resolutions/refresh tab) #2) Sharpen is set to 0%.

Assuming both of these conditions are met and NIS is enabled globally in the control panel (with 0% sharpening since you can tweak that upwards per profile if you so desire -- globally you can set it to 0% I mean), there is just no reason for NIS to be running any operations whatsoever. If it doesn't run scaling and if it doesn't run sharpening then there should be no perf impact. When those conditions are both met Nvidia's driver should just do whatever it was doing before automatically. Obviously sharpening and spacial scaling aren't free, but the driver should be smart enough to know when neither is being used.

If they refuse to fix this then the obvious work around is to just let users enable NIS per specific game profile no? This way we could manually turn it on for games we know we'll be upscaling/sharpening and leave it OFF for games that we know we'll be running at native and not sharpening. I hate using GeForce overlay and in the tests I've seen online using NIS / sharpen via GeForce overlay may even have a slightly larger perf impact so I do not want to go that route (and what about users that just manually install the driver and don't use GeForce?).

Maybe I'm missing something, I just don't really understand your comment/why you'd say this won't or shouldn't be fixed -- at least as I understand things seems like it absolutely should be in either of the ways I described above.

 

I could be wrong, but I assume he's saying that they're not bringing back their old GPU scaler and their old sharpening filter because they were worse/caused worse quality and/or more artifacting VS the new sharpening filter and NIS spatial scaling.

That said, unless i'm missing something, I definitely think the performance loss of NIS when gaming at native resolution with 0% sharpening should be fixed. Either by having the driver check for those conditions automatically or at least they should let users only turn on NIS manually per game application profile.

Interestingly with NIS enabled globally (Sharpening set at 0%), of course I do see the NIS overlay indicator even when gaming at native resolution -- however, if I switch to a resolution above native and make use of the new "smart" downscaler (replacement for DSR) the NIS overlay icon disappears so if it is turning NIS OFF when DL DSR is used, then it seems to me all they'd have to do is that exact same thing when the application is being run at native and sharpen is set to 0%. Really there's no reason the sharpen and upscaling passes should even be forced together, one should easily be able to run without the other.

Again, if I'm missing something correct me, but it definitely seems like the implementation should not work like it does for that case (native res/0% sharpen). The spatial scaling should only come into play when at sub native resolutions and when another better scaler is not being used (such as DLSS). The sharpening pass also shouldn't run ever unless sharpening is set to a value greater than 0.

 

In FH5 i got the same impression so far, on 1440p UW there was no difference in IQ between 2.25 DLDSR and regular DSR i could discern. People on Reddit report same thing as far as i can tell.

 

That's not true. NIS has more artifacting compared to the bilinear upscaler of my GPU. And I can still use it by disabling NIS. Problem is sharpening is then locked away from me for no valid reason whatsoever.

 

I dunno what to tell you, looks fine to me. Smoothness at 0 for both as I find it counter productive to the whole idea of downsampling to slap a blur filter on top of it.

 

Maybe NVIDIA will come to its senses and return a separate sharpness setting without binding to the NIC. Since it does not work in conjunction with DSR and DLDSR.

 

DLDSR vs 4x DSR Tests at 1440p

So wrapping up, I think 1.78x DLDSR at 1440p is an absolute great tool to get a better image quality in older games, not so much in newer or more action focused games where performance is more important.

-> https://www.reddit.com/r/nvidia/comments/s4qr69/dldsr_vs_4x_dsr_tests_at_1440p/

Note:
I have similar experience, x1.78 (3413:1920) FTW.
Guys but please remember to change in game reolution to DLDSR

 

So far I haven't gotten any freezes in Guardians of the Galaxy with this driver. Maybe they silently fixed something

 

I could not get GForce filters to work for me, they pop up but the sliders have no effect

 

You could have done literally a few seconds of research to know that DLDSR requires tensor cores.

 

It had functionality problems rendering it incompatible with many games either in non function or artifacts.