[Micky C]

Artificial neural networks. Artificial intelligence so to speak.

[Hank]

^ I think, the stuff icecoldduke is talking is more like:
Applications built on the RTX platform bring the power of real-time photorealistic rendering and AI-enhanced graphics, video and image processing, to enable millions of designers and artists to create amazing content in a completely new way.

Reference:
https://developer.nvidia.com/rtx

I doubt he plans to get into the complex, programming level of Nvidia itself, it's more like trying to make that technology, from them, work for EDuke32

[Daniel]

Damn... im on AMD Radeon R9 290X and Win7 x64 :/

What about Radeon Rays are they usable somehow ?
RadeonRays 2.0

Baikal tech-demo Github:
Baikal

[icecoldduke]

Micky C, on 16 September 2018 - 04:18 PM, said:

Artificial neural networks. Artificial intelligence so to speak.

The idea of rendering to a subset of pixels and filling in the blanks with narual network is an interesting idea. AI to assist real time computer graphics is an area of untapped research that has some potential, and maybe utilizing ANN becomes a optimization feature for PolymerRTX.

[TON]

https://twitter.com/...472096268337152

[icecoldduke]

TON, on 19 September 2018 - 11:29 AM, said:

...

How many of you guys plan on buying a 20xx card? Vulkan RT extensions are only supported on 2080/2080 ti(and soon 2070) hardware right now, and I don't see that it has a fallback layer. I have no problems switching over if you guys are planning on buying the hardware, otherwise I can create a Vulkan RT based renderer but you guys won't be able to run it .

[icecoldduke]

No examples yet for GL_NVX_raytracing --> https://github.com/K..._raytracing.txt

[Mark]

For the last 2 years or so I have been wanting to buy a whole new system but laziness keeps me using this same 9 year old Dell and a 750ti card. Eventually I'll upgrade but I don't see myself buying cutting edge tech. So I can't say for sure if I'll have what it takes to run your new project when the time comes.

[icecoldduke]

Mark, on 19 September 2018 - 12:41 PM, said:

For the last 2 years or so I have been wanting to buy a whole new system but laziness keeps me using this same 9 year old Dell and a 750ti card. Eventually I'll upgrade but I don't see myself buying cutting edge tech. So I can't say for sure if I'll have what it takes to run your new project when the time comes.

That's why I'm inclined to say on Optix,at least initially so more people can use PolymerRTX. I'd really love to use Vulkan, but if you guys don't have the hardware to run it, kind of makes it meaningless. Some of you might not get more then simple textured geometry with PolymerRTX but at least you'll be able to run it.

What is everyone's thoughts? If everyone wants vulkan I'll switch over this weekend when I get my new hardware, but again, you won't be able to run RT without the hardware.

[Romulus]

Seeing how GTX 1080 Ti runs at 5 FPS with Shadow of The Tomb Raider's RTX effects enabled, I am convinced that it's best not to worry about supporting hardware that doesn't come with RayTracing acceleration.

[TerminX]

Do something that runs on my 980 Ti and 1070... not planning on upgrading any time soon.

[icecoldduke]

TerminX, on 19 September 2018 - 02:08 PM, said:

Do something that runs on my 980 Ti and 1070... not planning on upgrading any time soon.

I personally think this is the way to go. I figure PolymerRTX will have two levels, one just does primary rays and you get geometry with textures and that's it. Then if you have 20xx series, you can try and turn on advanced lighting, reflections and such. Some of the Optix demos that have reflections work just fine on my 1080 and hold > 30 fps at 1080p.

[Mark]

Does the simple geometry of most Build maps make it possible that the game will run much faster than the fancy demos you have run so far?

[icecoldduke]

Mark, on 19 September 2018 - 02:38 PM, said:

Does the simple geometry of most Build maps make it possible that the game will run much faster than the fancy demos you have run so far?

I don't have the build engine up and running yet, but I have made level viewers for Build and Quake 3 running on DXR FL. Both viewers rendered only primary hit only without textures, rendered at over 100 fps on my 1080 at 1080p.

[icecoldduke]

Quake 3's q3dm1 map 100% raytraced running on a 1080 at 130fps primary ray hit only. This is also tris only no patch tessellated surfaces, was too lazy to do that .

[lamduck]

icecoldduke, on 19 September 2018 - 02:18 PM, said:

I personally think this is the way to go. I figure PolymerRTX will have two levels, one just does primary rays and you get geometry with textures and that's it. Then if you have 20xx series, you can try and turn on advanced lighting, reflections and such. Some of the Optix demos that have reflections work just fine on my 1080 and hold > 30 fps at 1080p.

I have a 1080 and plan on waiting until more DX RT and Vulcan RT stuff is out before getting a new card, because I think the cards will be cheaper with 7nm process and I will probably need a whole new system to have a faster CPU for the new GPU to be fed.

[icecoldduke]

lamduck, on 19 September 2018 - 06:27 PM, said:

I have a 1080 and plan on waiting until more DX RT and Vulcan RT stuff is out before getting a new card, because I think the cards will be cheaper with 7nm process and I will probably need a whole new system to have a faster CPU for the new GPU to be fed.

This is why I don't want to make PolymerRTX in Vulkan RT. Turing launch was terrible, and from a consumer perspective, I completely understand why you would skip this gen. Only ONE RTX powered game at launch(Final Fantasy 15). The rest of the games consumers play, performance is on par with a 1080 ti at a 30% higher price markup. Then with 7nm GPU's in a year or two? Makes sense just to wait it out.

My mission with PolymerRTX is to have something for everyone. For those that don't upgrade, you should be able to get Polymost style visuals with the development freedom of the classic renderer(provided your GPU isn't a total pile of shit ). For those that do upgrade to the 20xx series, you'll have things like reflections, lighting, etc.

To deliver on that goal, I have to either use DirectX Raytracing or Optix. Optix has NEVER been used in games, and I'm probably going to find out why soon enough. If you guys do purchase the RTX series, I'll revisit creating a Vulkan RT renderer, but I just can't do it if no one on here has the hardware to run it.

[enderandrew]

icecoldduke, on 19 September 2018 - 11:51 AM, said:

How many of you guys plan on buying a 20xx card? Vulkan RT extensions are only supported on 2080/2080 ti(and soon 2070) hardware right now, and I don't see that it has a fallback layer. I have no problems switching over if you guys are planning on buying the hardware, otherwise I can create a Vulkan RT based renderer but you guys won't be able to run it .

The current Steam hardware survey says 3% of users have a GTX 1080 currently.

This is your project. Do as you will. But if you target the 2080, then you're likely only targeting a very small percentage of EDuke32 players.

[icecoldduke]

enderandrew, on 19 September 2018 - 07:13 PM, said:

The current Steam hardware survey says 3% of users have a GTX 1080 currently.

This is your project. Do as you will. But if you target the 2080, then you're likely only targeting a very small percentage of EDuke32 players.

Let's just kill this conversation then. I'm going to focus on 970, 980, 1070, 1080 and 2080 series, sound good? That means I'm punting on Vulkan for the time being.

[MusicallyInspired]

Actually, I've been looking into the whole realtime raytracing thing and it's pretty neat. Regardless of whether people have the cards yet or not I like the idea of pursuing it for the future. I'd like to see Duke get attention in that area especially if Q2 and other source port renderers are already doing it. Even if I can't actually enjoy it for many years. Could you make two forks of your engine?

[lamduck]

icecoldduke, on 19 September 2018 - 06:49 PM, said:

This is why I don't want to make PolymerRTX in Vulkan RT. Turing launch was terrible, and from a consumer perspective, I completely understand why you would skip this gen. Only ONE RTX powered game at launch(Final Fantasy 15). The rest of the games consumers play, performance is on par with a 1080 ti at a 30% higher price markup. Then with 7nm GPU's in a year or two? Makes sense just to wait it out.

My mission with PolymerRTX is to have something for everyone. For those that don't upgrade, you should be able to get Polymost style visuals with the development freedom of the classic renderer(provided your GPU isn't a total pile of shit ). For those that do upgrade to the 20xx series, you'll have things like reflections, lighting, etc.

To deliver on that goal, I have to either use DirectX Raytracing or Optix. Optix has NEVER been used in games, and I'm probably going to find out why soon enough. If you guys do purchase the RTX series, I'll revisit creating a Vulkan RT renderer, but I just can't do it if no one on here has the hardware to run it.

I like that plan, I'd like to be cutting edge although I think they priced a few too many people out with the 2080 TI, the regular 2080 doesn't even seem to be much of an increase from my 1080 for the money. That being said I really like raytracing and can't wait to be able to run it, the price is just too much for me because I'd have to build a whole new PC and then have only a few RT titles, although it will get better in the coming months.

[Romulus]

FF15 doesn't have any RTX implementation within the game yet, just a benchmark which has been dubbed bogus by many people due to having bugs with culling, draw distance, LOD, etc. DLSS does seem to provide the RTX cards with a significant performance boost without any reduction of image quality.

I am in possession of 2080 FE and 2080 Ti FE for the next few weeks and right now and the only thing that stands out is the Unreal Engine 4 Star Wars demo, that runs around 22~30 ish FPS on 2080 Ti without DLSS and 30~37 with DLSS, whereas it barely gets 6~7 FPS on a GTX 1080 Ti. But it's not a game.. and then there's 3DMark Dandia but I don't have a way of obtaining that, and it's WIP.

[MusicallyInspired]

Will Blender take advantage of raytracing accelerated cards I wonder to make rendering faster?

[Romulus]

MusicallyInspired, on 20 September 2018 - 04:19 AM, said:

Will Blender take advantage of raytracing accelerated cards I wonder to make rendering faster?

They'll have to push an update to specifically take advantage of the tensor cores, which is something I am looking forward to in the near future.

[icecoldduke]

Romulus, on 19 September 2018 - 09:27 PM, said:

FF15 doesn't have any RTX implementation within the game yet, just a benchmark which has been dubbed bogus by many people due to having bugs with culling, draw distance, LOD, etc. DLSS does seem to provide the RTX cards with a significant performance boost without any reduction of image quality.

I am in possession of 2080 FE and 2080 Ti FE for the next few weeks and right now and the only thing that stands out is the Unreal Engine 4 Star Wars demo, that runs around 22~30 ish FPS on 2080 Ti without DLSS and 30~37 with DLSS, whereas it barely gets 6~7 FPS on a GTX 1080 Ti. But it's not a game.. and then there's 3DMark Dandia but I don't have a way of obtaining that, and it's WIP.

I'm picking up my 2080 pre-order from Best Buy either tomorrow or saturday. Whats the performance delta between the star wars demo on the 2080 vs the 2080 ti?

[icecoldduke]

Found this video on reddit, Star Wars demo on the 2080 vs 2080 ti vs 1080 ti

[Romulus]

Once you're running native 4K and you've uncapped the 24 FPS limit, the RTX 2080 FE runs it about 16~24 FPS depending on what's on screen. Once DLSS is enabled, I have seen the 2080 FE hits 30 FPS. even going as high as 33 FPS.

The 2080 Ti FE runs the demo natively at 22~30 FPS and once DLSS is enabled, 30~37 FPS, the highest recorded FPS from the timedemo stats tells me is 42. Both my cards are marginally overclocked.

Edit: The video you posted seems to be showing comparison of the cards running with DLSS on except for the 1080Ti which doesn't have support for it. DLSS renders at below native resolution, though it's very difficult to tell the difference in motion, and UE4 being post processing heavy is good at hiding the loss of details from the reduction in pixel count.

[icecoldduke]

Romulus, on 20 September 2018 - 08:51 AM, said:

Once you're running native 4K and you've uncapped the 24 FPS limit, the RTX 2080 FE runs it about 16~24 FPS depending on what's on screen. Once DLSS is enabled, I have seen the 2080 FE hits 30 FPS. even going as high as 33 FPS.

The 2080 Ti FE runs the demo natively at 22~30 FPS and once DLSS is enabled, 30~37 FPS, the highest recorded FPS from the timedemo stats tells me is 42. Both my cards are marginally overclocked.

Edit: The video you posted seems to be showing comparison of the cards running with DLSS on except for the 1080Ti which doesn't have support for it. DLSS renders at below native resolution, though it's very difficult to tell the difference in motion, and UE4 being post processing heavy is good at hiding the loss of details from the reduction in pixel count.

Would you be able to run a similar test at 1080p with both the 2080 FE vs the 2080 FE TI without DLSS(not that it matters at the resolution)?

[Romulus]

I did that yesterday, but these were done at stock speeds with nothing altered.

RTX 2080 FE: Maximum - 92 FPS, 1% Low (Minimum) - 55 FPS
RTX 2080 FE: Maximum - 102 FPS, 1% Low (Minimum) - 58 FPS.

The reason 1080p isn't ideal for testing these card is because it becomes very CPU dependent, and I am currently unable to push my 8700K past 4.8 GHz due to sending my AIO in for RMA.

What DLSS does is it takes the resolution down a notch, then scales it from there. So if you're running 2160p, it's actually sampling at 1440p. If you're running 1440p, it's sampling at 1080p. I never bothered running 1080p DLSS because it'd likely run at somewhere around 900p internally, which is not what I want. DLSS 2X is still yet to be implemented, we'll have to wait and see how that works.

[icecoldduke]

Romulus, on 20 September 2018 - 09:37 AM, said:

I did that yesterday, but these were done at stock speeds with nothing altered.

RTX 2080 FE: Maximum - 92 FPS, 1% Low (Minimum) - 55 FPS
RTX 2080 FE: Maximum - 102 FPS, 1% Low (Minimum) - 58 FPS.

The reason 1080p isn't ideal for testing these card is because it becomes very CPU dependent, and I am currently unable to push my 8700K past 4.8 GHz due to sending my AIO in for RMA.

What DLSS does is it takes the resolution down a notch, then scales it from there. So if you're running 2160p, it's actually sampling at 1440p. If you're running 1440p, it's sampling at 1080p. I never bothered running 1080p DLSS because it'd likely run at somewhere around 900p internally, which is not what I want. DLSS 2X is still yet to be implemented, we'll have to wait and see how that works.

Did you happen to run the star wars demo on a 1080 or 1080 ti at 1080p? Sorry for just throwing a bunch of questions at you, I don't have the hardware yet, and I'm trying to figure out how PolymerRTX will scale. These numbers are helping me out tremendously.