Skip to content
This repository has been archived by the owner on Mar 28, 2024. It is now read-only.

[BUG-229609] The death of Open GL Vulkan dilemma and photorealism vs end of rasterization #7439

Open
sl-service-account opened this issue Nov 2, 2020 · 6 comments
Open

[BUG-229609] The death of Open GL Vulkan dilemma and photorealism vs end of rasterization #7439

sl-service-account opened this issue Nov 2, 2020 · 6 comments
Labels
project:BUG

Comments

@sl-service-account
Copy link

sl-service-account commented Nov 2, 2020 •
edited by sl-issues-importer bot

How would you like the feature to work?

Everytime a new feature is requested, the core thing that is always considered is if it will break somehting existing.
But now somethign existing is broken forever Open GL, so this cannot apply.
2/3's of SL users have Vulkn comatible hardware.
By the time that at Linden Lab You have coded a new render engine on it, the remaining 1/3 of SL users that cannot run vulkan API will be way less.
So moving to Vulkan becuase it is crossplatform is a no branier...
the problem is how to offer support for that 1/6s that wont be able to run it when you will have a new Render engine supporting Global Illuminaiton at 144hz.
The answer is very simple...it is on how You frame htings like everyone does in politics, iat school, iat work , and in ones family.
You make a notice WAY before one year of expected moving of SL on to Vulkan stating that You wil cease support for Open GL runnign hardware, and that those who wil have such hardware wil nto be able to run second life.
Yhsi notice has to be sent out MINIMUM 18 month before the day of conversion, and teh date must either be a coupl eof weeks after black friday OR mid march, as these are the periods of the year which the economy in genral is mostly running at full tilt, and Second life renveues are at their highest.

Now moving to Vulkan is not the issue, the isses are more than this and more important.
Teh first issue is Phtorealism...whatever render engine You will opt touse, has to be capable NOW today of absolute phtorealism, if teh content is properly made, and My apologies, You guys at Linden Lab worked very hard for Sansar, but Umbra 3d like engines are by no means close to anythign usable.
You need to aim at Vray, redshift, corona render absillute feature movie phtorealism , if You want Your transition to be future proof.
this comes with teh following isses and consideratiosn to take into account:

  1. memory, mainstream graphics Vram memory will be way over 16Gb in 2023 because of MCM design and new generation memory, and soem techonolgies like ones used at AMD, will have shared memory between CPU and GPU for.
    Thsi Implies, that You wil need to make a roadmap, in which You will downscle textures renderd in during run time allowing uploads up to 8k .
    EG the first year u eg keep the 1k standard for diffuse and 512 for al the other shaders maps needed on teh specular workflow , including reflecion maps refraciton maps, fresnel IOR etcetc.
    and as Yoy gather data thatmore and more cleints movigninto new generation hardware You scale automatically teh texture size... OR you coudl actually make this a feature in the client prefernces grpahcis settings, so some wil run downscaled textres an dothers will run instead full resolution textures.
    it is a very easy thing to implement.

2 The render engine, needs to be otusourced tocompanie slike Chaos group or Otoy or similar...chaos gorup woudl be teh ideal choice, as they woudl be very willing to offer support for Vulkan APi , and develop a Vray version for gmaing, that opposed to Unreal engine does not need to bake the scene.... Vray supports specular, metalness workflows, has support, for substances, and has an excellent IBL or SUN sky system.
I do not think it would cost You more than 5 million to outsorce this to be built, and half a million a year for thei rengineers development and support, while your staff could instead focus on the Vulkan API.

3 AMD Smart access feature...
this feature is NOT to be underestimated, AMD with the purchase of xilinx, as things stand now creates a massive threat to the very existance of Nvidia and Intel in a very near future.....thi sis somehtign that for hwo dramtic i put it has to be taken into consieration, and a further reason to use Vulkan API.
Even if at Nvidia they wil come with amn amazing gpu release with hopper mcm architecture on a new node, it might not be good enough to compete with an all amd system, and Intel is now not credible enough anymore to have anyhtign tangible enough.
This might be a worse case scenario bu tit is a scenario that is abig issue an dthat has to be definitely taken into account, as whatever render engine Youwill use, it wilhave to be capable to support Intel NVIDIA and AMD hardware, and it is very veyr hard to say how things wil turn out,,, the good thing is that at Nvidia they do support vulkan.
SO please do not fall into the RTX marketing stunt By Nvidia like it happened for VR as RTX and Optix and AI denoiser are defintely a dead end and will not have a future

Why is this feature important to you? How would it benefit the community?

this feature is important to me becuase 3d is all about photorealism and Photorelaism is the ultimate goal and what Id like for You to give to SL.
This feautur eis iportan tto teh collunity becuase phtorealism is what they want, even if it wwas for fantasy like Pixar scenes, which are stil photorealsitic and belivable in theyr stylized way.

Attachments

Original Jira Fields
Field Value
Issue BUG-229609
Summary The death of Open GL Vulkan dilemma and photorealism vs end of rasterization
Type New Feature Request
Priority Unset
Status Accepted
Resolution Unresolved
Reporter COCA Yven (coca.yven)
Created at 2020-11-02T14:32:30Z
Updated at 2020-11-05T10:03:03Z 
{
  'Build Id': 'unset',
  'Business Unit': ['Platform'],
  'Date of First Response': '2020-11-02T10:06:09.251-0600',
  'How would you like the feature to work?': "Everytime a new feature is requested, the core thing that is always considered is if it will break somehting existing.\r\nBut now somethign existing is broken forever Open GL, so this cannot apply.\r\n2/3's of SL users have Vulkn comatible hardware.\r\nBy the time that at Linden Lab You have coded a new render engine on it, the remaining 1/3 of SL users that cannot run vulkan API will be way less.\r\nSo moving to Vulkan  becuase it is crossplatform is a no branier... \r\nthe problem is how to offer support for that 1/6s that wont be able to run it when you will have a new Render engine supporting Global Illuminaiton at 144hz.\r\nThe answer is very simple...it is  on how You frame htings like everyone does in politics, iat school, iat work , and in ones family.\r\nYou make a notice WAY before one year of expected moving of SL on to Vulkan stating that You wil cease support for Open GL runnign hardware, and that those who wil have such hardware wil nto be able to run second life.\r\nYhsi notice has to be sent out MINIMUM 18 month before the day of conversion, and teh date must either be a coupl eof weeks after black friday OR mid march, as these are the periods of the year which the economy in genral is mostly running at full tilt, and Second life renveues are at their highest.\r\n\r\nNow moving to Vulkan is not the issue, the isses are more than this and more important.\r\nTeh first issue is Phtorealism...whatever render engine You will opt touse, has to be capable NOW today of absolute phtorealism, if teh content is properly made, and My apologies, You guys at Linden Lab worked very hard for Sansar, but Umbra 3d like engines are by no means close to anythign usable.\r\nYou need to aim at Vray, redshift, corona render absillute feature movie phtorealism , if You want Your transition to be future proof.\r\nthis comes with teh following isses and consideratiosn to take into account:\r\n\r\n1. memory, mainstream graphics Vram memory will be way over 16Gb in 2023 because of MCM design  and new generation memory, and soem techonolgies  like ones used at AMD, will have shared memory between CPU and GPU for.\r\nThsi Implies, that You wil need to make a roadmap, in which You will downscle textures renderd in during run time allowing uploads up to 8k .\r\nEG the first year u eg keep the 1k standard for diffuse and 512 for al the other shaders maps needed on teh specular workflow  , including reflecion maps refraciton maps, fresnel  IOR etcetc.\r\nand as Yoy gather data thatmore and more cleints movigninto  new generation hardware You scale automatically teh texture size... OR you coudl actually make this a feature in the client prefernces grpahcis settings, so some wil run downscaled textres  an dothers will run instead full resolution textures.\r\nit is a very easy thing to implement.\r\n\r\n2 The render engine, needs to be otusourced tocompanie slike Chaos group or Otoy  or similar...chaos gorup woudl be teh ideal choice,  as they woudl be very willing to offer support for Vulkan APi , and develop a Vray version  for gmaing, that opposed to Unreal engine does not need to bake the scene.... Vray supports specular, metalness  workflows, has support, for substances, and has an excellent IBL or SUN sky system.\r\nI do not think it would cost You more than 5 million to outsorce this   to be built, and half a million a year for thei rengineers development and support, while your staff could instead focus on the Vulkan API.\r\n\r\n3 AMD Smart access feature...\r\nthis feature is NOT to  be underestimated, AMD with the purchase  of xilinx, as things stand now  creates a massive threat to the very existance of Nvidia and Intel in a very near future.....thi sis somehtign  that for hwo dramtic i put it has to be taken into consieration, and a further reason to use Vulkan API.\r\nEven if at Nvidia they wil come with amn amazing gpu release with hopper mcm architecture on a new node, it might not be good enough to compete with an all amd system, and Intel is now not credible enough anymore to have anyhtign tangible enough.\r\nThis might be a worse case scenario bu tit is a scenario that is abig issue an dthat has to be definitely taken into account, as whatever render engine Youwill use, it wilhave to be capable to support Intel  NVIDIA and AMD hardware, and it is very veyr hard to say how things wil turn out,,, the good thing is that at Nvidia they do support vulkan.\r\nSO please do not fall into the RTX marketing stunt By Nvidia like it happened for VR as RTX and Optix and AI denoiser are defintely a dead end and will not have a future\r\n",
  'ReOpened Count': 0.0,
  'Severity': 'Unset',
  'Target Viewer Version': 'viewer-development',
  'Why is this feature important to you? How would it benefit the community?': 'this feature is important to me becuase 3d is all about photorealism and Photorelaism is the ultimate goal and what Id like for You to give to SL.\r\nThis feautur eis iportan tto teh collunity becuase phtorealism is what they want, even if it wwas for fantasy like Pixar scenes, which are stil photorealsitic and belivable in theyr stylized way.',
}
@sl-service-account
Copy link
Author

Ansariel Hiller commented at 2020-11-02T19:28:07Z

Maybe we should simply face the death of Mac instead - cheaper and way less user are affected even if it means the way overpriced Apple workstations at the Lab become useless. Nobody should fall for Apple's false claim that OpenGL is deprecated technology because they were a) plain simple incompetent to fix their broken implementation for years and are b) now using the deprecation claim to distract from that fact and work further on a vendor lock-in by promoting their proprietary rendering API.

@sl-service-account
Copy link
Author

Ansariel Hiller commented at 2020-11-03T07:34:48Z, updated at 2020-11-03T07:44:28Z

Sorry to say, but there are quite a bit of errors and misunderstands in what you wrote. Some examples:

I detest Apple products, but this being said, Apple is a Market Maker as Microsoft as AMD and just as Intel was in the past.
Apple is a market maker for overpriced lifestyle products self-proclaimed hipsters pay any price Apple is calling for. With their neglectance of the workstations and pro-users for many years and blatant ignorance of those user's needs, Apple should be glad there are still pros using their overpriced crap - many of those already switched to other platforms.

Think of top tier graphic cards with 200 Gb of vrame memory that is 16 times faster of what u had the past decade, with up to 8 gpus in a single package working as one
Not within the next decade,

In 2024 mainstream graphic cards will likely ship starting from minimum 32 Gb Vram
Unlikely - even if it does, lots of users run SL on a better toaster.

a whole game couldbe completely running on the shared memory of the gpu and the gpu.
I assume you mean CPU and GPU. That shared memory already exists - it's slow as hell because GPU memory is much faster than DRAM. If you're on Windows, run DXDiag and you will notice that Windows also virtualizes VRAM and can swap it out, if VRAM actually gets exceeded - which is painfully slow. But maybe you meant the so called "Shared Memory Feature" that AMD has presented with their new 6000 series GPUs: That just means the CPU can write to and read from directly GPU memory - it doesn't extend your memory because as already stated: That is slow as hell!

Nvidia struggling with RTX to replace CUDA
RTX has nothing to do with CUDA

that as they HAVE NO OTHER CHOICE other than to move to Vulkan
You always have a choice: Spend tons and tons of money and implement a new rendering pipeline that currently 30% of the users can't use or spend less money and drop support for only about 6% of users (= Mac). Not to mention they will need to put additional money into Mac support thanks to Apple's choice to switch to ARM CPUs which have a different instruction set. If everything works as easy even if they offer something like Rosetta 2.0 is a whole different story. Easy choice from financial point of view, especially since the Lab is still a company that needs to make profits. And my personal opinion on that: They should give Apple the finger - which is unfortunately very unlike since the Lindens love their overpriced Apple hardware...

I hope they budget 20 to 30 million spread in the next 5 years and get the platform with the biggest content delivery per day in the world ran number one game in the world
20 to 30 million to replace the rendering engine? By outsourcing it to some other company? Erm... guess not!

That's it for now, not even commenting on the original stuff you wrote.

@sl-service-account
Copy link
Author

Ansariel Hiller commented at 2020-11-03T18:05:36Z

You are not informed about imminent coming hardware and technology.
Unfortunately you are wrong.

What i write is prettuy much not what I think will happen but what is on the roadmaps to happen and it will happen precisely in the time frame i wrote
Well... Then provide links to reliable sources, not wild speculations!

I would kindly request of You to please stick to the argument of the API and the next render engine to be utilized by Linden Lab, if You do have any relevant savvy to share.
You cannot simply postulate an API change just because the manufacturer of a platform representing a tiny fraction of the entire userbase decides trying to force people into using their proprietary API. As I mentioned, you also need to take the fincial point of view into account and it's a simple math that from a business perspective it makes way more sense to drop support for something 6% of the people are using than spending a ton of money for performing an API change that as of now would lock out 30% of the users. That money could be easily spend on improving the current rendering engine for the remaining 94% of the users.

Also Vulkan was evaluated as on option not because of its feature set or special abilities, but because it would allow to continue support for OSX - unless Apple decides to cut support for that as well to promote their own API. Please stop daydreaming about all that super photo-realistic stuff that should suddenly be possible then - or by 2024 at latest or whatsoever. We are talking about Second Life with user-created content, which might range from well optimized to absolutely horrible. This is never going to happen, but since you won't believe me, I'll leave it up to whoever from Linden Lab will respond to this ticket.

Please do take some time to learn about what Vulkan API is and why it is important....
You better do so.

And do understand that for most users, it will be possible to run secondlife all maxed out with computers that will cost give or take 5/600 if they are using 1080p/1440p. and 1000 for 4k at above 144Hz
You have clearly no idea of how the viewer works and that simply replacing the "oh so bad" OpenGL with Vulkan won't fix anything at all (Hint: It's not OpenGL's fault everything is slow as hell!). The viewer would have to be completely rewritten from scratch, and in that case it would be way cheaper to rely on already existing engines like Unity or Unreal Engine, which comes at the cost of breaking existing content. From the point of a content creator like you this would of course be nice because everyone has to buy their stuff again, but then you clearly underestimate the reaction of the users if that would render most if not all of their content worth lots and lots of money useless - with or without an 18 month notice you proposed!

Anyway: I'm done with this discussion that will lead to nothing and go back to working on our viewer while waiting for Linden Lab to close this ticket as "won't implement". :D

@sl-service-account
Copy link
Author

COCA Yven commented at 2020-11-03T21:35:35Z

Vulkan is open source.

Please refrain form posting on this Jira, if not savvy of Vulkan and most logic behind the next render engine for SL.

I would kindly ask You  to delete Your posts, as irrelevant to the Jira and apparently self serving.

Thank You

@sl-service-account
Copy link
Author

Kyle Linden commented at 2020-11-04T19:48:47Z

Hello,

Vulkan support is on the roadmap for the future. I cannot speak to any implementation details but we're fully aware of the issues.

Thanks

@sl-service-account
Copy link
Author

COCA Yven commented at 2020-11-05T10:01:53Z, updated at 2020-11-05T10:03:04Z

Dear kyle

I understand how colficting and difficult was thsi decision to move on to vulkan becuase of many second life residents, stil not supporting it becuase of beign on old hardware.

I however want to encourage you and all of your team, thinking very big support and expanded multiplatform for this API with is GI and Raytracing capabilities, in conjunciton with Amazon being one of the small enteprise to move on AMD infrastructure....You will have great longevity and if You will do Your homework right wil definitely, increase SL residents (educated guess) by 700% in 50 months.

Good luck and aim big, with a big state of the art photorealistic scalable capable future proof  render engine!!

You might want to contact

Peter Tsekov peter.tsekov@chaosgroup.com       at

Chaos Group

8500 Steller Dr, Suite 8A

Culver City, CA 90232

 

and see if They are willing to exploit the possibility a Version of Vray for Second Life running on vulkan API

I know  that they have been researching otoy like brigade game render engine, and that becuase of AMD come back in GP GPU computing, they must be definitely be at this moment have   people working on vulkan API.

It is a contact worth spending soem time to have an informal talk with  to maybe see if worth making a small 4 member team, feasibility study.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees
No one assigned
Labels
project:BUG
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@sl-service-account