• 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.

DLSS 5 - Generative AI

Half the performance with increase power for something one would after a few minutes in a game forget about it but will see more artifacts, and a jerkier presentation.
 
MistaSparkul said:
In the same video not even a minute later he literally says it is not quite transformative enough in NBA. I still think Nvidia could have chosen a better game to showcase DLSS5 where the difference is much greater, but if no other game was ready for release then I guess they had no choice but go with what was ready for launch.
Click to expand...
He specifically mentioned that for game play. For broadcast style close ups, he implied it did.

After watching it, it made me think if there was a way for DLSS 5 to be dynamically turned on or off for those kinds of games. On only for cut scenes/cinematic scenes and off during gameplay, particularly when the difference isn't that obvious. That said, I have seen some examples where DLSS 5 was transformative even during gameplay so it would be a case to case basis thing.
 
Ok so I got DLSS5 setup in Control Resonant. So unlike the UE5 games, this game with DLSS5 actually looks fine even if DLSS is used instead of DLAA before the NR step, maybe it's because this game has proper path tracing along with RR 4.5 while Stalker 2 and Dawnwalker do not? I believe those two games are software lumen only and no RR support so maybe that's why using upscaling doesn't play very nicely with DLSS5 but I could be wrong.

Anyways, with full path tracing + DLAA + no DLSS5 at 1440p I can get about 70fps, turning on DLSS5 drops me below 60 but then using DLSS Quality I then get the same frame rate as path tracing + DLAA + no DLSS5. Kick on 2x FG and I'm at about 130fps, and since image quality doesn't take a significant hit this time around I've decided this is how I'll play the game (Path Tracing + DLSS Quality + DLSS5 + 2x FG). The DLSS5 upgrade definitely feels less significant in this game which is not what I was expecting since I thought giving it a path traced image is supposed to yield even more of an uplift as opposed to just pure raster. Here's a quick look:


View: https://youtu.be/SUAqLaXK-g8
 
Last edited:
  • Like
Reactions: erek
like this
MistaSparkul said:
Ok so I got DLSS5 setup in Control Resonant. So unlike the UE5 games, this game with DLSS5 actually looks fine even if DLSS is used instead of DLAA before the NR step, maybe it's because this game has proper path tracing along with RR 4.5 while Stalker 2 and Dawnwalker do not? I believe those two games are software lumen only and no RR support so maybe that's why using upscaling doesn't play very nicely with DLSS5 but I could be wrong.

Anyways, with full path tracing + DLAA + no DLSS5 at 1440p I can get about 70fps, turning on DLSS5 drops me below 60 but then using DLSS Quality I then get the same frame rate as path tracing + DLAA + no DLSS5. Kick on 2x FG and I'm at about 130fps, and since image quality doesn't take a significant hit this time around I've decided this is how I'll play the game (Path Tracing + DLSS Quality + DLSS5 + 2x FG). The DLSS5 upgrade definitely feels less significant in this game which is not what I was expecting since I thought giving it a path traced image is supposed to yield even more of an uplift as opposed to just pure raster. Here's a quick look:


View: https://youtu.be/SUAqLaXK-g8
Click to expand...

So far seems like UE5 games are the ones that are making the most of DLSS5 in the IQ department.
 
If you do not fuck up your settings, by tweaking stuff to 2.0 and 10 x passes DLSS 5 does wonders in games:
1790573525858.png
 
It think we need to get performance at a more acceptable level. Ray Tracing is already a huge performance hit, and this on top of that makes it a luxury feature only for those who want to spend 5K on a video card...
 
KazeoHin said:
It think we need to get performance at a more acceptable level. Ray Tracing is already a huge performance hit, and this on top of that makes it a luxury feature only for those who want to spend 5K on a video card...
Click to expand...
That's its real issue at this point. nVidia is working on it, we'll see what they are able to do. If they can get it integrated with the upscaler it may become a lot more useful.
 
KazeoHin said:
It think we need to get performance at a more acceptable level. Ray Tracing is already a huge performance hit, and this on top of that makes it a luxury feature only for those who want to spend 5K on a video card...
Click to expand...
Unustified entitlement these days.
People want their entry level GPU to do max settng with PT @ 4K and cannot lower settings to macth their hardware oh not it must be "bad optimization".
No GPU could run Doom3 at maxium settings at a high resolution monitor on release.
Neither couldCrysis.
Or the Witcher 3.

But now we have generation entitlement crying all the time.
 
Sycraft said:
That's its real issue at this point. nVidia is working on it, we'll see what they are able to do. If they can get it integrated with the upscaler it may become a lot more useful.
Click to expand...

After playing around it for quite a bit, I can see why Nvidia currently chose not to run NR before upscaling. Heck, in some games just even using upscaling at all but still running NR on the final output resolution results in worst image stability compared to DLAA. Once they managed to improve things which I'm sure they will, then I will go back to a 4K monitor. But man I gotta say 1440p with DLAA preset M already looks so clean, I haven't gamed on a 1440p display since 2018 and thought I would never be able to go back once going 4K but the latest DLAA has really proven me wrong. I've been happily gaming at 1440p + DLAA + NR for the past 3 weeks and it's been amazing.
 
Apologies for the vague post, I forgot who I was wanting to reply to


I mean it is a first gen tech. I think this popped off well enough for a first go at it. Idk why people are so hung up on it - Is the power draw awful? Yes. Is it performance heavy? Yes. But like isn't that how all new tech is? That is the price you pay as an early adopter. In a gen or 2 this tech will be more usable and eventually solid.
 
KazeoHin said:
It think we need to get performance at a more acceptable level. Ray Tracing is already a huge performance hit, and this on top of that makes it a luxury feature only for those who want to spend 5K on a video card...
Click to expand...
yes making it work on NVFP4 instead of 8 bbits and at some sort of not on fhe full final resolution could help quite a bit.

It is small enough vram wise (specialy if it can go even lower ot the said "4bits") that it could eventually run on a Groq type portion of the silicon (that can be on a old node) and be extremelly fast. Will see what it look like on Rubin, could already be cut by more than half by then.
 
littlegreenmenfromhome said:
Unustified entitlement these days.
People want their entry level GPU to do max settng with PT @ 4K and cannot lower settings to macth their hardware oh not it must be "bad optimization".
No GPU could run Doom3 at maxium settings at a high resolution monitor on release.
Neither couldCrysis.
Or the Witcher 3.

But now we have generation entitlement crying all the time.
Click to expand...

Yep, you see the consequences of it in games now. Max settings in most games target 60 fps for mid range hardware. And if it isn't people cry that it's "unoptimized".

People praise the new Doom games for being optimized, and they are. The graphics are good for the performance, but it is just that. You remove "for the perofrmance" and the graphics are unimpressive. Things games did a decade ago, ray tracing so scaled back you're not even sure they're using it. A far cry from the brand new next gen real time lighting tech Doom 3 had at the time, along with max settings texture detail that extended beyond top end GPU memory limits.
 
sharknice said:
Yep, you see the consequences of it in games now. Max settings in most games target 60 fps for mid range hardware. And if it isn't people cry that it's "unoptimized".

People praise the new Doom games for being optimized, and they are. The graphics are good for the performance, but it is just that. You remove "for the perofrmance" and the graphics are unimpressive. Things games did a decade ago, ray tracing so scaled back you're not even sure they're using it. A far cry from the brand new next gen real time lighting tech Doom 3 had at the time, along with max settings texture detail that extended beyond top end GPU memory limits.
Click to expand...

Funny how back then when new high end graphics features were made with future hardware being able to run it in mind, we were pretty OK with it. Nowadays if some new tech comes out and it doesn't instantly run at 4K 240fps on an RTX 2060 people will get their pitch forks out. :ROFLMAO:
 
sharknice said:
Yep, you see the consequences of it in games now. Max settings in most games target 60 fps for mid range hardware. And if it isn't people cry that it's "unoptimized".
Click to expand...
60 FPS was good back when we only had 60 Hz monitors. That time has passed, people are aware of 120 FPS+ now.

DLSS and FG were pretty huge because they allowed for better performance than native and good image quality. Now we're starting to get more crap thrown on top and framerates are dropping again...

I don't have a problem with DLSS5 in theory, but in practice it's not worth the massive framerate loss. We'll see if further iterations are more acceptable. A feature that's only feasible if you have a $5,000+ GPU is not a well-conceived feature.
 
sharknice said:
Yep, you see the consequences of it in games now. Max settings in most games target 60 fps for mid range hardware. And if it isn't people cry that it's "unoptimized".
Click to expand...
Or sometime they hide the highest setting and call them unobtanium or wait to release them post launch as many publication will only benchmark the highest setting and for some reason many give importance to those, having some FOMO the moment they do not play at max setting, regardless of what the game actually look like at medium/high.

seem a non brainer that every game should have setting that let you go over any current computer can currently run, so 5-10 years laters you can use them when they require no additional work (like number of rays per pixel, ray bounce, range of LOD views and so on)

MavericK said:
A feature that's only feasible if you have a $5,000+ GPU is not a well-conceived feature.
Click to expand...
that seem a bit of an exageration (of a product of so much of the influencer running 4k monitors to play games).

There is a single game with official support making this a bit hard to tell, but a 5070ti max setting, 1440p was able to run it at 86 fps at launch, 5070 at 1080p was able to run over 100 fps and that without testing if High + dlss 5 look better than max setting instead, which is possible.

It is not currently a really working feature (single game support it, not a selling point for a gpu) of course, but it is one of the most rapid and closer to being workable proof of concept of a major new feature we saw, near movie type real time ray tracing or original DLSS were more rougher than this, this seem a single sram inference away to be way faster than traditional rendering at much higher quality and today it is already okish for a mod type experience.

Would not be surprising if it is already ready and just waiting for good game implementation before Rubin successor launch (and importantly for other game studio to make their own version with different training set/rewards, trained on their own z-brush high quality model and so on)

MavericK said:
DLSS and FG were pretty huge because they allowed for better performance than native and good image quality.
Click to expand...
Or better image quality at same or better performance, there an array of possiblity with DLSS, better performance at the cost of lesser image quality being just one of them.
 
LukeTbk said:
many publication will only benchmark the highest setting and for some reason many give importance to those
Click to expand...

it's fine for benchmarking - you're supposed to find the max capabilities of the hardware - you can't do that if you still give the hardware room to breathe - you have to find its limits by going beyond

it's everyone expecting they should be able to achieve ultimate max in new titles at playable frames with even modern hardware that's really not the norm and the problem
 
Last edited:
kram182 said:
it's fine for benchmarking - you're supposed to find the max capabilities of the hardware - you can't do that if you still give the hardware room to breathe - you have to find its limits by going beyond
Click to expand...
I meant that type, when it is not an hardware review but more a game performance review:
https://www.techpowerup.com/review/silent-hill-townfall-performance-benchmark/5.html

they do change resolution for some reason, but not the settings, only ultra and they are not alone (https://www.dsogaming.com/pc-perfor...-townfall-benchmarks-pc-performance-analysis/)

So I could imagine why some studio decide to hide for the futur settings to not be the metric people see a bit everywhere before buying the game or not
 
LukeTbk said:
I meant that type, when it is not an hardware review but more a game performance review:
https://www.techpowerup.com/review/silent-hill-townfall-performance-benchmark/5.html

they do change resolution for some reason, but not the settings, only ultra and they are not alone (https://www.dsogaming.com/pc-perfor...-townfall-benchmarks-pc-performance-analysis/)

So I could imagine why some studio decide to hide for the futur settings to not be the metric people see a bit everywhere before buying the game or not
Click to expand...


even for that I think it's fine because a 'game performance review' like that is just a limited scope 'hardware benchmark review' by a different name and in one specific title - but I get what you mean - there are those who put out 'playable settings' videos etc that are more in line with what you're thinking/want where they try to establish a baseline of settings/HW combos for 'playable frame rate' like Digital Foundry etc (or old [H] playable settings reviews as an example of) - but the problem there is then people taking reviews like the one you posted and seeing 'unoptimized' when they can't hit playable frames with what they got - because unoptimization does exist and is a thing - but it's more indicative when settings/performance don't scale down properly or even 'lowly settings' aren't achievable/playable on more powerful HW that should be capable of etc
 
You must log in or register to reply here.
Back
Top