RTX continues to seriously disappoint me

[Stuka87]

Alright, ran the benches myself.

Ultra 56.43
Ultra +200% shade- 42.19
Extreme- 31.93
RTX no DLSS- 47.82
RTX- 54.71

Throw away the DLSS numbers, I just ran them because that's now default RTX mode so for comparison. RTX is faster than ultra with a 200% shading rate and nothing else and significantly faster than extreme. There was obviously a performance patch for Metro in the last few months, heh.

Is there no way to run RTX with Extreme? Its obvious Extreme moves the bottleneck to someplace other than the RTX hardware.

But I am not sure what the point of comparing RTX (that isn't using extreme preset) to non-RTX that is. if RTX is using the Ultra preset, then it is obviously performing worse, which is what we have all come to expect. Not sure why this is an argument?

 

[Hitman928]

Alright, ran the benches myself.

Ultra 56.43
Ultra +200% shade- 42.19
Extreme- 31.93
RTX no DLSS- 47.82
RTX- 54.71

Throw away the DLSS numbers, I just ran them because that's now default RTX mode so for comparison. RTX is faster than ultra with a 200% shading rate and nothing else and significantly faster than extreme. There was obviously a performance patch for Metro in the last few months, heh.

What resolution are you running?

 

[Qwertilot]

Its has been proven on every game that has DLSS support that basic up-scaling provides better image quality and nearly identical performance. Several places have run these tests. AMD's new Radeon Image Sharpening also provides superior image quality than DLSS, and works with any DX9/DX10/DX12 game (DX11 support is coming). Now obviously AMD doesn't have RT support yet, so its only handy for people wanting to boost performance of regular rasterized games. But it shows that nVidia's tech behind DLSS is inferior at this time.

Oh, look. In one corner you've got a fixed upscaling algorithm, in AMD's case a slightly fancier one.

In the other corner you've got people spending considerable lengths of time to train a game specific neural net to do upscaling tailored specifically for that game. There isn't even beginning to be any competition in terms of the technology used here.

There logically will be IQ gains to be had, and moreover NV absolutely wouldn't be doing this if they weren't there. Yes marketing but super computer training time is truly not cheap. They're not insane.

 

[BenSkywalker]

1080p, it's a little 2060 after all.

The reason for the numbers, increasing the shader rate is a larger performance hit than using RTX, in other words arguing that sticking with higher shader loads will be better than moving to ray tracing is certainly not universally applicable. I used what most people say is the best looking PC game to date, and adding RTX is, by far, the largest visual improvement over ultra, and very notably, the *fastest* performing of the better than ultra settings.

My stance is simple, ray tracing is the best way forward, others are saying more shader hardware is better. In Metro doubling the traditional shader load, which is a miniscule visual difference, is slower than using RTX which in some situations is a staggering difference.

Upscaling, I don't like any of them, DLSS might end up being good, but it ain't there yet, and the Radeon sharpening tools have been available on the green side for a while now, too many artifacts imo. I'm not bashing any of them, but I like clean and that they aren't.

Forgot to mention, the metro bench is torture test, in game performance is quite a bit higher(although a 2060 isn't doing 4k no matter what).

Edit- I ran numbers with extreme and then turned on DXR and it came back higher then just plain extreme, the shader resolution auto lowered to 100%, re ran with the shader resolution manually set to 200%, extreme with RTX on

RTX extreme 100%- 36.73
RTX extreme 200%- 26.22
Extreme- 31.93
RTX- 47.82
Ultra- 56.43
Ultra 200%- 42.19

DLSS off for all those numbers.

 

[Ranulf]

Quake 2 RTX is using ray tracing for everything right now, not sure what you're talking about taking a decade? With resolution shifting at a snail's pace again two generations out and the xx50 series will probably handle 1440p with ray tracing at solid performance levels.

I love your enthusiasm there Ben but this is Nvidia we're talking about. They were too cheap to put the new turing encoder on the 1650 cards.

 

[Hitman928]

1080p, it's a little 2060 after all.

The reason for the numbers, increasing the shader rate is a larger performance hit than using RTX, in other words arguing that sticking with higher shader loads will be better than moving to ray tracing is certainly not universally applicable. I used what most people say is the best looking PC game to date, and adding RTX is, by far, the largest visual improvement over ultra, and very notably, the *fastest* performing of the better than ultra settings.

My stance is simple, ray tracing is the best way forward, others are saying more shader hardware is better. In Metro doubling the traditional shader load, which is a miniscule visual difference, is slower than using RTX which in some situations is a staggering difference.

Upscaling, I don't like any of them, DLSS might end up being good, but it ain't there yet, and the Radeon sharpening tools have been available on the green side for a while now, too many artifacts imo. I'm not bashing any of them, but I like clean and that they aren't.

Forgot to mention, the metro bench is torture test, in game performance is quite a bit higher(although a 2060 isn't doing 4k no matter what).

Edit- I ran numbers with extreme and then turned on DXR and it came back higher then just plain extreme, the shader resolution auto lowered to 100%, re ran with the shader resolution manually set to 200%, extreme with RTX on

RTX extreme 100%- 36.73
RTX extreme 200%- 26.22
Extreme- 31.93
RTX- 47.82
Ultra- 56.43
Ultra 200%- 42.19

DLSS off for all those numbers.

I don't know why you are so stuck on this 200% shader rate kick. If you doubled the shader resources you wouldn't just start supersampling everything to improve visual fidelity, that's essentially what you are suggesting.

 

[BenSkywalker]

The argument over dedicating die space to shader hardware versus ray tracing cores. I'm not the one who came up with this, it's part of the default in game settings. Using the preset graphics options in the game is what gave me the idea in the first place.

Increasing the detail on the shader routine increases visual fidelity, as does adding RTX. Adding RTX incurs less of a performance hit than increasing shading loads in Metro.

My argument, diminishing returns on shader routines will certainly reach a point where adding RTX like hardware is absolutely going to be better for improving visuals. Metro gives us the option of comparing 'next gen' shader loads versus 'next gen' ray tracing.

The performance numbers indicate that shifting die space to ray tracing now is a choice backed by fairly convincing real world examples.

Arguing for more general purpose shader hardware at this point gets us to, 'for what' territory. Higher resolution isn't going much further and the super majority is gaming on a 1080p or lower resolution screen, a resolution that the entry level ray tracing part is *already* handling in the best looking game out.

With diminishing returns a fairly long standing reality, I'm pulling for the paradigm shift in rendering.

 

[Hitman928]

The argument over dedicating die space to shader hardware versus ray tracing cores. I'm not the one who came up with this, it's part of the default in game settings. Using the preset graphics options in the game is what gave me the idea in the first place.

Increasing the detail on the shader routine increases visual fidelity, as does adding RTX. Adding RTX incurs less of a performance hit than increasing shading loads in Metro.

My argument, diminishing returns on shader routines will certainly reach a point where adding RTX like hardware is absolutely going to be better for improving visuals. Metro gives us the option of comparing 'next gen' shader loads versus 'next gen' ray tracing.

The performance numbers indicate that shifting die space to ray tracing now is a choice backed by fairly convincing real world examples.

Arguing for more general purpose shader hardware at this point gets us to, 'for what' territory. Higher resolution isn't going much further and the super majority is gaming on a 1080p or lower resolution screen, a resolution that the entry level ray tracing part is *already* handling in the best looking game out.

With diminishing returns a fairly long standing reality, I'm pulling for the paradigm shift in rendering.

What I'm saying is that if developers used additional rasterizer resources just to increase shader resolution to 2 samples per pixel, it'd be a stupid and wasteful use of resources. There are much better ways to use the additional resources which would also add additional things beyond shaders.

What you're arguing is like turning tessellation factor to 128x, getting no visual fidelity increase but a big drop in performance and then saying see, adding additional resources wouldn't increase visual fidelity. Your resource allocation makes no sense.

Developers like the idea of ray tracing because if/when it does become the standard for all cards to be able to use it with acceptable performance, they can stop a lot of the tricks they have to use with traditional rasterization to make things looks as good as they do. This will hopefully save them a lot of time and money. The problem is, the hardware has to be vastly more powerful than it is today in order to support such ray tracing without greatly degrading other visual quality features that we're used to having now. Because the hardware is not ready, developers still have to develop the game with all the same time consuming tricks as they did before, plus add ray tracing which still takes time as it has to be tweaked a lot to fit into current engines running on current hardware.

I think everyone in this thread would agree that ray tracing is the future, I think most just think that it's a waste on a consumer card right now. Personally I wish Nvidia wouldn't have hyped it anywhere close to what they did because it's no where near what they made it out to be, not yet. I also would have liked to have seen them roll RTX out first in professional cards, let developers start to play with it, then introduce a Titan card with it for the crazy gamers with cash to burn to tap into it. Then a little later bring out a Ti version to make it more accessible. Then maybe next generation start trickling it down further as developers have more time to develop their engines with ray tracing in mind.

 

[BenSkywalker]

Ok I wasn't very clear, I'm not saying to use the end visual results to compare when looking to the future, I'm saying look at the relative performance difference. Doubling the shader resolution allows us a clear look at how that stacks up in the context of an actual game engine and it's real world impact, as does RTX.

What would developers actually use that extra shader power for? Why I keep harping on diminishing returns is that we are already into very small incremental improvements with shader hardware, the devs have all the easy stuff done, and the moderate stuff, now we're getting into the expensive stuff and the really expensive stuff. So double the hardware shaders for a marginal increase, or move to ray tracing which is already quite playable today using a mid range part that's occasionally flirting with$300?

Now you think they should've rolled it out differently, and that is another completely valid discussion, I don't agree but I see that as far less of a black and white thing than discussing ray tracing versus more shaders.

On the rollout, I needed a new card during the mining boom and the only choice I had under $700(1070 non ti) was the 1060, I paid $350 for it and a year later gave that to my wife and grabbed a 2060(waiting for 7nm big boy parts) for the exact same price, oh yeah, that came with Metro for free. The 1060 was upgrading my old 780GHZ, it was a solid upgrade, but it just ran the same old things faster.

The 2060 was just fundamentally different, it could do things that I'd never seen in real time, and not just for tech demos, but in actual games. Metro is completely playable with RTX on, Q2 RTX the same, Tomb Raider also very playable(albeit only some shadows really benefit, nothing like the other two).

Real world, if you factor in the free game, the 2060 cost less than the 1060 and was fundamentally game changing. Now I get what you are saying, but as an owner of the slowest RTX card, I can play all the games they hyped with ray tracing on and completely playable. For a first gen part offering radical new technology? Straight up, the Voodoo 1 is the only part I can compare it to.

 

[Ranulf]

I find it amusing that you're comparing the 1060 to the 2060 using the inflated coin mining prices. Yes, real world indeed and in the summer of 2015 I got a custom 970 for $350 with 2 free games, bringing it down to the same price or a bit more than a R9 290 custom card (that had also come with free games). You overpaid for the 1060 and you did again for the 2060.

 

[BenSkywalker]

I overpaid for the 1060 to be sure, the 2060 came with a $59 game I was going to buy anyway which makes the effective price $290.

I just went back through my history, the Radeon 9500 Pro was the last time I spent less on a video card for myself. Do I fell like I overpaid? Lol, no. I'm comparing them because that is the real world, there's a reason that the 1060 has such a staggering lead over every other card for gamers, it's the only thing we could buy remotely sanely.

 

[Hitman928]

A few months after launch I picked up a 1060 for under MSRP. $350 was definitely a mining inflated price. Not the top of the inflated market, that was probably ~$400.

 

[Glo.]

Looks like not only users are disappointed in RTX:

https://www.techpowerup.com/258340/assetto-corsa-competizione-dumps-nvidia-rtx

Now que the smear campaign over the developers 😉.

 

[Krteq]

Dafuq? NV itself was working on RTX implementation and studio ditched it due to performance reasons.

RTX in a nutshell.

 

[DrMrLordX]

Wow, didn't expect that. Where was the NV money hat?

 

[BenSkywalker]

The dev said they are fixing the rest of the game, which honestly is what they should be working on right now, before DXR is added. Not quite what the sensationalist headlines read.

Competizione review.

I have a racing seat/ff wheel setup, I really enjoyed Assetto Corsa, I probably have about the same time into that as project cars, but the current game isn't ready for release(which already happened).

 

[nurturedhate]

The dev said they are fixing the rest of the game, which honestly is what they should be working on right now, before DXR is added. Not quite what the sensationalist headlines read.

Competizione review.

I have a racing seat/ff wheel setup, I really enjoyed Assetto Corsa, I probably have about the same time into that as project cars, but the current game isn't ready for release(which already happened).

That is not what they said. Close, but you seem to mischaracterize a lot. Since linking what they said wouldn't line up with what you wrote we can just post it for you.

"Our priority is to improve, optimize, and evolve all aspects of ACC. If after our long list of priorities the level of optimization of the title, and the maturity of the technology, permits a full blown implementation of RTX, we will gladly explore the possibility, but as of now there is no reason to steal development resources and time for a very low frame rate implementation,"

Just for clarification's sake.

 

[Glo.]

I told you: now que the smear campaign over the devs 😉.

 

[BenSkywalker]

The text was linked two posts up, why would I link it a second time? We know they have the DXR code base running, the game itself needs more work before they worry about that.

If a game is pushed out too early, IME it's *always* the publishers fault, not the developers.

I already stated their last game was great, how many hundreds of hours do you have into it? Want to compare who supports this developer more? Nobody has trashed the devs except some wishful troll delusions.

 

[BFG10K]

Looks like not only users are disappointed in RTX:

Remember folks, we were repeatedly told ray tracing just needs to "flip a simple switch and it'll automatically work everywhere with no developer effort!" 😵

Meanwhile BF5, the AAA RTX launch partnership title that cost $millions and had massive R&D and software engineering resources from nVidia, gave us horrendous performance, framespikes and completely blew out VRAM usage.

In exchange? Slightly better reflections, and ten second animated lighting GIFs on tanks.

What's interesting is that streaming, VR and RTX are diametrically opposed to each other, yet all three are being touted as the next big thing(tm). I personally think all three are garbage.

 

[nurturedhate]

The text was linked two posts up, why would I link it a second time? We know they have the DXR code base running, the game itself needs more work before they worry about that.

If a game is pushed out too early, IME it's *always* the publishers fault, not the developers.

I already stated their last game was great, how many hundreds of hours do you have into it? Want to compare who supports this developer more? Nobody has trashed the devs except some wishful troll delusions.

It has nothing to do with dev, or hours played, or money spent, or fanboyism, or epeen size. You stated something misleading, reason for doing so also does not matter. I simply provided clarity to people who would otherwise just read the posts and not click on the link. Context is important. For you to ignore that and go the epeen route.. well... I don't know, I''m happy for you?

 

[nurturedhate]

Remember folks, we were repeatedly told ray tracing just needs to "flip a simple switch and it'll automatically work everywhere with no developer effort!" 😵

Meanwhile BF5, the AAA RTX launch partnership title that cost $millions to make with massive R&D and software engineering resources from nVidia, gave us horrendous performance, framespikes and completely blew out VRAM usage, In exchange? Slightly better reflections, and ten second animated lighting GIFs on tanks.

What's interesting is that streaming, VR and RTX are diametrically opposed to each other, yet all three are being pushed as the next big thing (tm). I personally think all three are garbage.

They are only opposed to each other over the short term and it's going to be a long time but I want my holodeck you hear me! All three of those are tips on an iceberg that we'll eventually see all of. The tech and similar should continue to be developed. I do agree that as of today, from a consumer perspective, they are garbage with a couple nice exceptions.

 

[BenSkywalker]

WTF.... What did I say that's misleading?

You've been following the development of this game for years, right? You already watched the DXR footage they showed off? You were reading the thread on the official forums this post came from, correct?

This has absolutely nothing to do with epeen, this is I'm not a troll who read a single paragraph and thinks he's more informed than anyone else could possibly be.

 

[Glo.]

What's interesting is that streaming, VR and RTX are diametrically opposed to each other, yet all three are being touted as the next big thing(tm). I personally think all three are garbage.

All of them are good ideas, just released way too early for prime time. And I mean WAY too early.

 

[BenSkywalker]

You worked in UE4 with ray tracing yet BFG?

Instead of working through specific shaders for every freaking element in the game, you turn on ray tracing for the light source. It's exponentially easier than using 1,472 different shader routines.

Also, you keep harping on Frostbite, why not tackle Metro Exodus instead? It's very clear BFV is the poorest implementation of DXR by a comfortable margin, why not go after the best looking game to date instead?