Nvidia ,Rtx2080ti,2080,(2070 review is now live!) information thread. Reviews and prices

[PeterScott]

Hardware Unboxed talks about DLSS demo data.

This reveals something I was wondering about before. Does DLSS have a significant overhead that is being buried in the boost from running at lower resolution? The answer is Yes. FF Demo at 1440P is 72 FPS vs 4K DLSS (assumed 1440p) at 57 fps.

So DLSS is NOT lightweight AA method, and from what I remember, DLSS 2x uses a bigger DL Network as well, so it will likely end up being a VERY expensive AA method.

Also, the more that gets revealed the worse DLSS looks. If running TAA at 1800p gives essentially the same quality, and performance without the need for new GPUs, and network training, that makes DLSS look like questionable smoke and mirrors marketing.

All AMD has to do is encourage devs to do some mix of decent generic AA + 1800P, and they have a DLSS killer that doesn't need new GPUs.

 

[LTC8K6]

Perhaps DLSS has to work harder on the poor AA of that game, as described in the video?

DLSS could also get much better as time goes by.

 

[PeterScott]

Perhaps DLSS has to work harder on the poor AA of that game, as described in the video?

DLSS could also get much better as time goes by.

That doesn't work. DLSS starts with the AA off, and does it's own thing.

 

[LTC8K6]

That doesn't work. DLSS starts with the AA off, and does it's own thing.

I must have misunderstood the youtube guy.

But it seems certain to me that Turing will improve a lot.

 

[coercitiv]

DLSS could also get much better as time goes by.

But it seems certain to me that Turing will improve a lot.

AMD would like their marketing slides back please.

 

[LTC8K6]

AMD would like their marketing slides back please.

When you bring out the shill card, you probably have no arguments.

Why not just tell me why drivers won't improve Turing performance over time?

BTW, the best NV card I have ever owned is a 750ti, and I don't even have it now.

 

[PeterScott]

When you bring out the shill card, you probably have no arguments.

Why not just tell me why drivers won't improve Turing performance over time?

BTW, the best NV card I have ever owned is a 750ti, and I don't even have it now.

Turing will likely improve over time as SW matures.

But, the more info that comes out, the worse this looks for NVidia. This is from someone a lot of people accused of being Pro-NVidia.

DLSS originally marketed as performance/quality boost. Finally it gets revealed that there are two DLSS modes:

DLSS Performance version that gets performance from running at low resolution, something of a letdown on the quality front.

DLSS 2X high quality version at higher performance penalty. Which is completely unknown quality/performance, but likely very high performance hit.

The latest info indicates, that even basic DLSS has a significant performance hit, hidden in the low resolution performance boost.

1800P + TAA, is essentially the same as DLSS for both performance and quality. Which means you can do the same thing with current cards. No Tensor cores, no network training needed. This is really damning.

On top of that DLSS sounds fragile. It only appears on canned demos. It needs specific training for each game. What happens to DLSS if you download reshade mods, new textures? Probably it breaks.

All along I was holding out some hope that DLSS would be a game changer. Now it looks more like smoke and mirrors.

AMD has a massive opening here, if only they had something to take advantage of it. Especially in doing something to to equal to DLSS that actually works on old cards, and is easier to implement without needing network training and wouldn't be as fragile.

This release is really turning into a mess for NVidia.

 

[LTC8K6]

Turing will likely improve over time as SW matures.

But, the more info that comes out, the worse this looks for NVidia. This is from someone a lot of people accused of being Pro-NVidia.

DLSS originally marketed as performance/quality boost. Finally it gets revealed that there are two DLSS modes:

DLSS Performance version that gets performance from running at low resolution, something of a letdown on the quality front.

DLSS 2X high quality version at higher performance penalty. Which is completely unknown quality/performance, but likely very high performance hit.

The latest info indicates, that even basic DLSS has a significant performance hit, hidden in the low resolution performance boost.

1800P + TAA, is essentially the same as DLSS for both performance and quality. Which means you can do the same thing with current cards. No Tensor cores, no network training needed. This is really damning.

On top of that DLSS sounds fragile. It only appears on canned demos. It needs specific training for each game. What happens to DLSS if you download reshade mods, new textures? Probably it breaks.

All along I was holding out some hope that DLSS would be a game changer. Now it looks more like smoke and mirrors.

AMD has a massive opening here, if only they had something to take advantage of it. Especially in doing something to to equal to DLSS that actually works on old cards, and is easier to implement without needing network training and wouldn't be as fragile.

This release is really turning into a mess for NVidia.

We haven't even really got started with Turing, though.
It just came out.

Perhaps it will turn out that DLSS is a bust, but I don't understand slamming it at this point.

Besides, the biggest problem NV has is the price, not the features.

 

[PeterScott]

We haven't even really got started with Turing, though.
It just came out.

Perhaps it will turn out that DLSS is a bust, but I don't understand slamming it at this point.

Besides, the biggest problem NV has is the price, not the features.

Actual beneficial features, can mitigate the pricing problem. The more that features fail to do anything initially, the more glaring the pricing problem becomes.

 

[Malogeek]

We haven't even really got started with Turing, though.
It just came out.

That's part of the problem. Outside of enthusiasts with cash to spend who always want the flagship, it's very difficult to justify buying these GPUs right now. There's zero games using the touted features which are there taking up huge amounts of die space and apparently justifying the massive price increase.

As you said, we have not started with Turing. The 2080 and 2070 is basically a gamble purchase. No one has any idea of what they're getting for their money.

 

[SirDinadan]

NVIDIA just saw the opportunity to re-purpose the same tech for both professional users and gamers. AMD can't attack back with "traditional" gaming GPUs so they are pretty much covered on this front.

They are shaping the future now in gaming as well and it seems that hybrid raster-raytraced-AI pipelines will emerge in the not too distant future. This DLSS magic and ray-traced shadows / reflections are just the first baby-steps in that direction.

 

[jpiniero]

AMD can't attack back with "traditional" gaming GPUs so they are pretty much covered on this front.

The problem is getting developer support... and that's going to be tough without the consoles supporting HW RT. And if the consoles support HW RT, that obviously means AMD will too.

 

[crisium]

DLSS may have the same performance hit as 1800p + TAA but it surely looks better. We know this because DLSS is being compared to 2160p + TAA and it's pretty much a draw.

 

[dlerious]

NVIDIA just saw the opportunity to re-purpose the same tech for both professional users and gamers. AMD can't attack back with "traditional" gaming GPUs so they are pretty much covered on this front.

They are shaping the future now in gaming as well and it seems that hybrid raster-raytraced-AI pipelines will emerge in the not too distant future. This DLSS magic and ray-traced shadows / reflections are just the first baby-steps in that direction.

If AMD comes out with a raster only GPU that's within 5%-10% of the 2080Ti for around $750 (or less), they'll probably get my money.

 

[LTC8K6]

NVIDIA just saw the opportunity to re-purpose the same tech for both professional users and gamers. AMD can't attack back with "traditional" gaming GPUs so they are pretty much covered on this front.

They are shaping the future now in gaming as well and it seems that hybrid raster-raytraced-AI pipelines will emerge in the not too distant future. This DLSS magic and ray-traced shadows / reflections are just the first baby-steps in that direction.

Except that so far, we've seen no sign of the 2080ti being able to do RTRT at any decent resolution and frame rate, and that's what these cards were celebrated for at intro.

 

[sze5003]

Except that so far, we've seen no sign of the 2080ti being able to do RTRT at any decent resolution and frame rate, and that's what these cards were celebrated for at intro.

Tomb raider has been out for a few weeks now and I haven't seen an rtx patch yet. Unless one came in yesterday or the day before. Been a few days since I've played. There are not many games that support rtx yet anyway. Besides all these features these cards are hyped for are features devs will have to implement so it will take time. Until Nvidia works with all the game companies to train and provide feedback on the process, most of these features will come with patches for now. Until it becomes standard rendering procedure to include these bell and whistle features, we can't expect them to just work immediately.

 

[PeterScott]

DLSS may have the same performance hit as 1800p + TAA but it surely looks better. We know this because DLSS is being compared to 2160p + TAA and it's pretty much a draw.

That isn't what the HW UB video linked above says.

HWUB summary:
FF TAA is garbage, so DLSS can look better
Infiltrator has good TAA, so it looks better than DLSS.
Infiltrator 1800p + TAA is about equal to DLSS.

I downloaded original 4K videos (no youtube compression) of both demo videos and I concur the first two statements of HWUB, so I believe they have it about correct for the third.

 

[SirDinadan]

The problem is getting developer support... and that's going to be tough without the consoles supporting HW RT. And if the consoles support HW RT, that obviously means AMD will too.

I'm pretty sure that NVIDIA does everything in their power to get a console deal or to create a cloud service for console gaming. The latter is much more attractive proposition for them.

Except that so far, we've seen no sign of the 2080ti being able to do RTRT at any decent resolution and frame rate, and that's what these cards were celebrated for at intro.

It's marketing - they always promise the stars. And the whole release seems rushed, the tech is early, developers are only experimenting with it. Currently only great looking RTRT demos exist: Epic's StarWars and Porsche and Project Sol and Atomic Heart. SotTR and BF5 will utilize RT cores but just for shadows / reflection, etc.

 

[LTC8K6]

Tomb raider has been out for a few weeks now and I haven't seen an rtx patch yet. Unless one came in yesterday or the day before. Been a few days since I've played. There are not many games that support rtx yet anyway. Besides all these features these cards are hyped for are features devs will have to implement so it will take time. Until Nvidia works with all the game companies to train and provide feedback on the process, most of these features will come with patches for now. Until it becomes standard rendering procedure to include these bell and whistle features, we can't expect them to just work immediately.

IIRC, Huang said "it just works" several times in regards to Ray Tracing being implemented in gaming.

'The architecture is "the biggest generational leap in the history of computer graphics," Nvidia CEO Jensen Huang said at Gamescom 2018 in Cologne, Germany. Demonstrating what that means for gaming, he said, "Lighting, reflections, shadows -- it just works... Everything just works because ray tracing just works." The new features deliver 4K HDR gaming at 60 frames per second on even the most advanced titles, Nvidia says."

 

[sze5003]

IIRC, Huang said "it just works" several times in regards to Ray Tracing being implemented in gaming.

'The architecture is "the biggest generational leap in the history of computer graphics," Nvidia CEO Jensen Huang said at Gamescom 2018 in Cologne, Germany. Demonstrating what that means for gaming, he said, "Lighting, reflections, shadows -- it just works... Everything just works because ray tracing just works." The new features deliver 4K HDR gaming at 60 frames per second on even the most advanced titles, Nvidia says."

Yea I know he said it just should work. But I'm also a developer, not for games though. You know how many times architects or leads said "oh yea you can implement that, it will just work!"

Wish I had a dollar every time I heard that. Every time there are some snags you run into. He's marketing, of course he's gonna say it just works..they aren't going to say anything ambiguous that could halt sales of their launch product.

 

[LTC8K6]

He says it just works, I pay $1200 for it, and it doesn't work...

I can almost hear the class action lawyers tuning up...

 

[PrincessFrosty]

IMO, there's no limitation of DLSS as it is not an AA method and shouldn't be categorized as such. It is an AI guided super-sampling / upres method. The confusion stems from the fact that the official slides compared the image quality of DLSS to native resolution + TAA. And rightly so, as both way of producing a rendered frame introduces various amount of blur.

I'm still hazy on the details but I think I agree with this. The original methods of AA were to increase the sample rate of the scene and average those additional samples to generate a smooth gradiant in areas where there's sharp contrast, such as on edges. Smoothing out the jaggies. Yet none of the post processing methods can do that because the scene has finished rendering and the so called AA method is applied to an already completed scene, you cannot sample the scene anymore you can only deal with the output pixels. Which is why FXAA and similar methods are not averaging but bluring. The key difference is that averaging adds information to the scene where as bluring removes it.

I hate that FXAA has become like a new standard, I'd rather run with it off at 4k because jaggies a lot of the time are really hard to see, especially if you have a small 4k monitor. I've seen some 24" dell 4k panels out there, I bet they look amazing for games without AA.

 

[krumme]

I'm still hazy on the details but I think I agree with this. The original methods of AA were to increase the sample rate of the scene and average those additional samples to generate a smooth gradiant in areas where there's sharp contrast, such as on edges. Smoothing out the jaggies. Yet none of the post processing methods can do that because the scene has finished rendering and the so called AA method is applied to an already completed scene, you cannot sample the scene anymore you can only deal with the output pixels. Which is why FXAA and similar methods are not averaging but bluring. The key difference is that averaging adds information to the scene where as bluring removes it.

I hate that FXAA has become like a new standard, I'd rather run with it off at 4k because jaggies a lot of the time are really hard to see, especially if you have a small 4k monitor. I've seen some 24" dell 4k panels out there, I bet they look amazing for games without AA.

Its a matter of taste if you like aa. I guess most gamers think fxaa looks better. But one thing Is looks another is beeing competitive.
What works is 4k. 144hz on an ugly tn panel and no aa. Low or medium detail. Like it or not. It's effective.

 

[PrincessFrosty]

Its a matter of taste if you like aa. I guess most gamers think fxaa looks better. But one thing Is looks another is beeing competitive.
What works is 4k. 144hz on an ugly tn panel and no aa. Low or medium detail. Like it or not. It's effective.

I don't think it's a matter of taste. FXAA lowers overall IQ, edges are one thing that people focus on, but the IQ of the rest of the scene also matters. FXAA is indiscriminately lowering the quality of textures in the middle of polygons, as well as blurring the edges.

 

[IntelUser2000]

HWUB summary:
FF TAA is garbage, so DLSS can look better
Infiltrator has good TAA, so it looks better than DLSS.
Infiltrator 1800p + TAA is about equal to DLSS.

It almost feels like they should have kept the Tensor cores out and just had Ray Tracing as a feature. In that way, the space for the Tensor cores could have been used to increase performance with increased SMs, ROPs, caches, etc. That would have also boosted the performance in Ray Tracing.

Maybe there's a reason why they did not do so. Moore's Law is being jittery nowadays, but we're still getting scaling, so more transistors in an area. What's really being slowed is in performance gains, and worse, power reductions. For the high end CPUs and GPUs though the design becomes entirely power and thermal limited.

So what do you do? You add accelerators to take advantage of the abundant area. Power limits are "solved" since accelerators only run sometimes, not always, which allows that part of the chip to power down.