DLSS on Battlefield 5 tomorrow 12th February

That's truly horrible. So much hype for this tech and yet it turns out to be another useless gimmick.

 

I wouldn't just judge it on the one game and call the tech useless. There's another game coming out in a couple of days. We need more comparisons to make an informed decision.

 

They better not skip a machine learning day

 

Agreed. I've been skeptical of DLSS since the announcement but I didn't expect it would look at bad as it does.

It's not just BF, the same negative effects are present in FF XV and Metro Exodus.

 

What a mess, great showing for RTX cards once again.

 

Did Novidia or Dice released any screenshots of what DLSS suppose to look like in-game?

 

I seem to remember that DLSS running on FFXV was more detailed than TAA. That was from one of the first videos, but this game never interested me, so i never payed to much attention.
BFV looks pretty awful from what they have shown so far.

As for Metro, i can't speak for that as i have only seen It with Ray Tracing but no DLSS.

 

The whole launch has not been a good one that's for sure.

 

Then you dont remember it well. Check it out.

 

It's pretty rough from what I've seen.

Thanks for this. Interesting to see that DLSS has problems in motion too .

 

Thanks for putting that up. Ok, i was wrong saying DLSS looks sharper. I'd say as long haired Baby Jesus said, they both look about the same, you would need a magnifying glass to spot the difference in this case. Yes there was annoying flickering on some thin objects, but I have seen this sort of thing in some forms of AA in the past. I don't have this game nor a 20 series video card and unless AMD released a magic driver for my 480, I don't think I will ever be able to run this so can't say which one I would choose. I still think the technology is still in its infancy and needs time to mature. If you disagree, then that's fine.

 

They did it other way around. Decreasing rendering resolution to achieve same/similar fps. Yes, DLSS-OFF at 75% looked better than DLSS-ON at 100%. But let's face it, you would not play at 75% rendering resolution either. (Or unlikely.)
Proposed scenario is to run DLSS-OFF 100% against something like DLSS-ON at 125%.

 

I'm happy with the performance boost and the full use of my card tbh.

 

I'm surprised as well; I expected a lot more based on previous comparisons using the FFXV demo. I'm guessing DLSS works better for fixed path scenarios than actual gameplay - most benchmarks and synthetics run the same loop every time so training would be more efficient on them. A human analogy would be like playing the same level over and over again - eventually, you're going to become really efficient at it since you know exactly what's coming at every turn. You won't be nearly as efficient on a level that you've only played once or twice.

I suppose more training would produce more accurate results, but that could take time and might be very costly for the developers (they might miss the all-important holiday window or be forced to release the game alongside a much more popular game of the same genre). Future driver updates could also improve things, although I doubt many gamers would be holding their breath for it.

The initial results from BFV (as well as FFXV and Metro Exodus) are certainly underwhelming, but this is only the first-gen of RTX. I view this current gen as a test phase more-or-less - the GTX 1660 shows that Nvidia is not aiming at the mainstream just yet (probably a good idea).

 

qft

 

Actually the good part about DLSS is that once its integrated it doesn't cost anything for developers to improve it. Nvidia is going to continually train the games on it's system and supply updates to the games via Geforce Experience.

That being said the image quality clearly sucks right now lol

 

I'm starting to wonder if DLSS is limited to "scale" for gaming. First impression - is better suited at non-gaming implementations. These posts showing Gaming IQ are interesting; the "3D Mark DLSS" benchmark was in more of a "controlled-environment" than the average online-multiplayer game/open-environment.

With random real-time physics calculations on buildings etc; I wonder how more "shifting real time-variables" affect the DLSS framework? Dev > DLSS pipeline > SAT-V > (drivers/tensor cores , gpus/games) > all in milliseconds? mmm I dunno... some information will be localized to drivers, still I feel there's a lot of ambiguity from NV themselves on DLSS. I read DLSS results involve training & time (ML, so makes sense) but I have questions:

1. What will turn-around be per game, on avg post-launch, for DLSS beating conventional AA?
2. Should the goal be matching the aforementioned at Day-1/launch eventually?
3. If DLSS improves for one game does that affect all DLSS games?
4. Can/Will DirectML produce better/equivalent results faster?

Curious: Does SatV talk to RTX owners' gpus in real-time at all? I've tried to research Nvidia's release papers on DLSS in attempts to find this to no luck. Anyone know?

 

Benchmark loops are generally short too which I presume cuts down on training time.

I'm hopeful for improvements with DXR but DLSS I'm not so sure, it depends on how long Nvidia has been training it. If it takes several months of training per game to achieve what we've seen so far I don't think DLSS is going to stick around.

I presume Nvidia will want to push RTX into the mainstream within another gen or two, they can't keep it a halo feature if they want it to take off.

They posted about that later in the thread, resolution scale and TAA are locked out with DLSS on.

 

That really sux. They try hard to prevent objective comparison.