Hacker News Reader

> For one, all shaders would need to be converted from HLSL to GLSL

There are transpilers (cross-compilers?) that translate from HLSL to GLSL and back and into every other shading language out there, there's really a lot of such tools.

> Secondly as I understand it the Vulkan API is much more low level and barebones than DirectX

Vulkan is about as low-level as Direct3D 12, which has not much in common with Direct3D 11 and lower, just like Vulkan has not much in common with OpenGL. OpenGL is (very roughly) equivalent to Direct3D 11, and Vulkan is more like Direct3d 12. The thing is that switching between the similar APIs is a lot of work, but does not require large refactorings, whereas switching from OpenGL/Direct3D11 to Vulkan/D3d12 requires more fundamental work.

"For one, all shaders would need to be converted from HLSL to GLSL, which could become quite a big task depending on the amount of shaders, and testing of all changes."

Ironically Microsoft have open-sourced a Clang-based HLSL compiler that could be used as a front-end for a Vulkan backend.

https://github.com/Microsoft/DirectXShaderCompiler

https://blogs.msdn.microsoft.com/directx/2017/01/23/new-dire...

That's not irony, just circumstantial.

I think the irony is in Microsoft, traditionally thought of as a "embrace, extend, extinguish" proponent, being the one providing an open-source tool facilitating escaping their own DirectX vendor lock-in.

> And why is this even a concern for a game started in 2011? It's mind boggling that they are making these kind of decisions 6 years into development, surely just finishing the damn gameplay should be the focus?

That would force them to show the results of the effort and the $texas-sized dollars contributed. As long as they can keep pointing to "well, we had to make a major architectural change.." they can continue pushing the idea that they'll deliver.

Dig into the criticism of them the last couple years. This one is from ~20 months ago: http://www.pcgamer.com/derek-smart-star-citizen/ or ~4 years into development.

Ah yes, the neverending criticism from Derek Smart, who was so sure the project would completely fall apart in "60-90 days" back in September 2015. He has this really weird obsession with wanting the project to fail.

https://archive.is/y1kMK#selection-3799.0-3964.0

Not to mention his even weirder obsession with Sandi Gardner, Star Citizen's community manager.

https://www.reddit.com/r/DerekSmart/comments/4w0w51/has_doct...

Wow. Okay that guys is nuts.

Still, star citizen is a train wreck. Wasn't the big thing last year to "rewrite all the net code".

They're just rearranging the deck chairs over and over waiting for the company to go under

Admittedly, "rewrite all the net code" is still a thing. Part of this is supposed to land in 2.6.2 ("soon"), while the bulk of it is targeted at 3.0 later this year. They're perpetually behind schedule in a lot of areas, but at the same time they're showing of some pretty impressive stuff.

I was a $20 backer on the kickstarter campaign and have since gone in for about $100 total. Obviously I wish it had come out years ago, but I'm still optimistic about the project, and fine with the drastically expanded scope. The alpha releases are far from perfect, but I've already had more fun with it than some other games that I paid much more than $20 for. That's what pushed me to back it further. There's just something incredible about the level of detail of the world, going from first person through a space station, out the airlock, walking over to the landing pad, hopping into a fully detailed spaceship (not just a disembodied camera in a cockpit), and flying off to fight space pirates.

This is a game that wouldn't be possible to make with the traditional publisher driven funding systems. Maybe it'll turn out to be impossible to make this way too, but I don't think so.

If you have a couple of hours to put videos on in the background, the last three "Around the Verse" studio updates have been pretty good:

March 2 (Frankfurt) - Hurricane and Character Customization https://www.youtube.com/watch?v=NGzDI2wUqf0

March 9 (Austin) - Weapons: The Game's Arsenal https://www.youtube.com/watch?v=c_v-hMu_h5s

March 16th (Manchester) - Level Design https://www.youtube.com/watch?v=-c2DogQL95o (watch past the credits for in-progress ships)

Or for a quicker coverage, here's a list of short excerpts from the last video. Quality isn't great, but it gets the point across: https://www.reddit.com/r/starcitizen/comments/5zu3of/atv_gfy...

Elaborating on the netcode, that comes back to their choice of CryEngine back in 2012. It's really geared toward smaller scale shooters where "Update every client with literally every state change" is a reasonable choice, and can't scale to an MMO. If CIG had known then that they'd get funded to the tune of $145,000,000 and make a much bigger game, they might have decided to make their own engine instead.

On the other hand, CryTek's financial trouble over the last several years was an enormous benefit to Star Citizen, because it let them build the Frankfurt office with a bunch of CryEngine experts who were tired of not getting paid on time. They're the folks who made the full scale and crazy detailed procedural planets possible.

When alpha 3.0 lands later this year (hopefully...) you'll be able to fly down and land on these from orbit: https://www.youtube.com/watch?v=uU0hDriMqXo (short excerpt from one of the studio updates linked above)

Missed the edit window, but Sandi Gardiner is actually the VP of Marketing. My bad!

FWIW this is actually an announcement about decreasing scope, not increasing it. It's more work relative to sticking with existing OpenGL/DX11 support, but they are basically saying instead of supporting two next gen graphics API's, they are just supporting one. In theory that should bring it closer not to release, not farther.

> For one, all shaders would need to be converted from HLSL to GLSL

Not necessarily. Vulkan now supports HLSL input too, which is compiled into SPIR-V, same way GLSL is. You don't need HLSL → GLSL translation step.

> For one, all shaders would need to be converted from HLSL to GLSL

GLSL is just one option to write shaders for Vulkan (the only inmportant thing is that SPIR-V code can be generated). I can even imagine quite well that one could write a compiler which generates SPIR-V code from HLSL shaders.

> Secondly as I understand it the Vulkan API is much more low level and barebones than DirectX, meaning it'll be a lot of work to shore up the differences.

DirectX 12 is similarly low level.

> I may be wrong here, but it seems to be like the developer is saying that a DirectX -> OpenGL change is no big deal

DirectX 11.x is similar in "level" to OpenGL 4.x (though the difficulties are a little different - I don't want to go into details here). Similarly DirectX 12 and Vulkan are similar in level (much more low level than their predecessors). No wonder since both Vulkan and DirectX 12 are "strongly inspired" by Mantle. Additionally Vulkan was intended as the succesor of OpenGL and DirectX 12 as the succesor of DirectX 11.x.

> surely just finishing the damn gameplay should be the focus?

That is what you wish. :-)

>That is what you wish. :-)

Star Citizen definitely appears more like an ongoing crowd-funded video game research and graphics research project at times.

> I can even imagine quite well that one could write a compiler which generates SPIR-V code from HLSL shaders.

Khronos are working on building this into the official SPIR-V compiler: https://github.com/KhronosGroup/glslang/issues/362

Trying to remember the last time a game deep in development announced something like this. Oh yes, Duke Nukem Forever and Daikatana...

You shouldn't be downvoted - that is a legit criticism of a game that has been rather long in development. At least it's not development hell like DNF or Daikatana, the people building it seem to genuinely make progress, even if they rebuild parts more often than might be necessary. The question is if these symptoms will sink the ship before they reach shore - their situation is obviously different, but no matter whether you throw a bowling ball, or a paper airplane, both eventually hit the ground.

They announced plans to use new generation of graphics APIs a long time ago. It's only now they made it clear they prefer Vulkan to DX12. So I'd say it's their work as usual, except they at last made the clear choice (or at least made it public, they probably chose Vulkan in the past already).

If Vulkan was "much lower-level" than DX12, then DX12 performance wouldn't even be in the same ballpark as Vulkan performance. Both Vulkan and DX12 were made more "lower-level" to gain that extra CPU and multi-core efficiency. I think Vulkan still beats DX12 in performance, if all else is equal, but I don't think the difference is huge.

DX12 is not DX11+1. It's a clone of AMD's Mantle, which was eventually given to Khronos, so they make it into Vulkan. The main advantage of Vulkan is that it runs almost everywhere. DirectX 12 doesn't.

> The main advantage of Vulkan is that it runs almost everywhere. DirectX 12 doesn't.

DirectX 12 runs on XBox One, while Vulkan does not:

> http://www.redgamingtech.com/directx-12-for-xbox-one-is-here...

And for OSX/iOS, you have to use OpenGL or Metal.

and Vulkan does not run on windows uwp apps, which MS is strongly pushing forward.

and everyone else ignoring

> DirectX 12 runs on XBox One, while Vulkan does not:

And the Switch uses Vulkan; while Vulkan on top of Windows (including 7 and 8/8.1) supports Linux and Android. For iOS and macOS there's MoltenVK, but I can't speak for its quality.

While the Switch supports Vulkan, it isn't the main graphics API, that role belongs to NVN.

https://blogs.nvidia.com/blog/2016/10/20/nintendo-switch/

So it remains to be seen which one gets the love of game developers.

> So it remains to be seen which one gets the love of game developers.

That's actually a fairly interesting question, given the percentage of games that come out for Nintendo platforms which are exclusive to that platform (or at least to the Nintendo family of platforms). There are all the first party titles, and then there are all the esoteric games which don't really fit the control schemes of the other platforms (but the switch is less unique in this aspect). I imagine games targeted purely at Nintendo platforms might use their proprietary API, especially first party games, which might get the API changed based on their needs.

> The main advantage of Vulkan is that it runs almost everywhere.

It doesn't run on Sony PS4, Nintendo Wii U, XBox ONE, UWP, OS X, iOS, tvOS.

Even on Android it is going to take a while until most people exchange their handsets for ones having it, because the majority will never get an upgrade for Android 7.

Lock-in freaks will crumble gradually. The times of balkanized graphics will be in the past, same way as happened with the Web. And the slogan of "industry does it that way" is irrelevant. Luckily the industry is now changing for the better. Expect further change.

If you're comfortable coding to something as low-level as Vulkan, it shouldn't be hard to use whatever API the PS4 or XBox has. You'll want to customize for the hardware regardless, since it's a bit underpowered compared to PC, so the only difference will be your API functions don't begin with vk_

Comfortable doesn't mean you always have time and resources to do it for N target APIs which are completely incompatible. Game studios clearly benefit from avoiding reinventing the wheel and duplicating their work. That's the whole point of lock-in proponents. They want to tax developers, to make cross platform development more expensive, forcing at least part of them to limit their release targets. That's a dirty anti-competitive tactic.

Well, we can scratch off one of those, no one cares about Wii U :)

Even on Android 7, Vulkan is not mandatory (devices must support Level 0, i.e. have a stub lib and return 0 surfaces, they should support Level 1, i.e. normal implementation).

Only OpenGL ES 3.2 is mandatory.

This was probably made for chipset producers; if Vulkan was made mandatory, you would see exactly 0 updated devices.

Thanks for the clarification, I wanted to mention it as I got the feeling of having read about it somewhere, but I failed to find the right spot in the documentation.

I remembered it wrong too. Not even OpenGL ES 3.2 is mandatory, only if you claim that the device is capable of high-perfomance VR. Otherwise, OpenGL ES 2.0 is fine (but 3.1 lib and stub symbols must be available).

Source: Android 7.1 Compatibility Definition Document

Chapter 7.1.4:

Device implementations MUST support both OpenGL ES 1.0 and 2.0, as embodied and detailed in the Android SDK documentations. Device implementations SHOULD support OpenGL ES 3.0, 3.1, or 3.2 on devices capable of supporting it.

Chapter 3.3.1:

Note that device implementations MUST include libGLESv3.so and in turn, MUST export all the OpenGL ES 3.1 and Android Extension Pack function symbols as defined in the NDK release android-24. Although all the symbols must be present, only the corresponding functions for OpenGL ES versions and extensions actually supported by the device must be fully implemented.

Chapter 3.3.1.1:

Device implementations, if not including support of the Vulkan APIs:

- MUST report 0 VkPhysicalDevices through the vkEnumeratePhysicalDevices call.

- MUST NOT delare any of the Vulkan feature flags PackageManager#FEATURE_VULKAN_HARDWARE_LEVEL and PackageManager#FEATURE_VULKAN_HARDWARE_VERSION .

Chapter 7.9.2:

Android handheld device implementations MUST identify the support of high performance virtual reality for longer user periods through the android.hardware.vr.high_performance feature flag and meet the following requirements.

...

- Device implementations MUST support OpenGL ES 3.2.

- Device implementations MUST support Vulkan Hardware Level 0 and SHOULD support Vulkan Hardware Level 1.

....

Support depends on whether the GPU can handle compute shaders. That's a starting point.

No, the starting point is that Google doesn't require a Vulkan capable GPU on Android 7 devices, thus leaving everyone that cares about the size of customer base to keep using OpenGL ES.

If the engine needs two rendering interfaces, the effort is exactly the same as OpenGL ES + whatever else.

Unless you are suggesting game developers should be happy to sell only to LG V20, Samsung S7 and Google Pixel owners.

I'm talking about hardware capabilities. Actual support depends on the drivers. And why would Qualcomm, Imagination, Nvidia and etc. not want to provide them, if they can? Google has nothing to do with it (at least not in the sense of developing them).

Turnaround in the mobile hardware space is even faster than on the desktop (let alone stagnating consoles), so Vulkan availability is much less of an issue there.

The OEMs building those handsets get to choose which SOCs are used on those devices.

It doesn't matter if the drivers are available if those GPUs aren't on the motherboard.

Currently Vulkan is constrained to 2.8% of the Android market.

> Turnaround in the mobile hardware space is even faster than on the desktop (let alone stagnating consoles), so Vulkan availability is much less of an issue there.

Only in US, not everyone around the world, specially in countries where pre-pay is the way to go, changes their mobile every two years.

> It doesn't matter if the drivers are available if those GPUs aren't on the motherboard.

My point is, this situation will be shorter than on the desktop. Because as I said, the turnaround of mobile hardware is faster.

> ly in US, not everyone around the world, specially in countries where pre-pay is the way to go, changes their mobile every two years.

In US too, not everyone is going to buy every new model. I'm talking however about manufacturers making them. That happens pretty fast. So Vulkan support on new hardware is probably already a given and this whole thing is a non issue. Any new technology has adoption period.

Vulcan doesn't have a shading language. It has a shader IR.

6 years into development, rather huge tasks will feel like weekend projects ...

"And why is this even a concern for a game started in 2011? It's mind boggling that they are making these kind of decisions 6 years into development, surely just finishing the damn gameplay should be the focus?"

Well, I would say, it is not a ordinary game. And even though I don't know about their Engine, but normally it should be possible to make those changes on the engine without any effect for the gameplay, since they wanted to support OpenGL before.

> Well, I would say, it is not a ordinary game.

No indeed. Neither was Duke Nukem Forever.