AMD dropping the ball. Nvidia and Apple pushing for walled APIs. Consoles going extinct. Valve moving away from native APIs.Vulkan is fragmenting and it shows. This is a dire situation, only new manufacturers from China can salvage this!
>>109106133Why is he missing a tooth? Is he poor?
>>109106173They got it fixed after getting a DX12 contracting job
>>109106133It won't get popular for as long as new versions require new hardware. MS learned this, and DX12 runs on 12 y/o hardware.
>>109106133>Valve moving away from native APIswhat?>Nvidia and Apple pushing for walled APIsDoes NVIDIA have a secret DX12 alternative I don't know of?Vulkan runs everywhere and is the backbone of a lot of critical software. It's not going anywhere.
>>109106225This problem is a natural part of the hardware industry. People want features, and it is often easier to make a physical circuit for it than it is to find an algorithm that runs on general purpose shaders.A weird in between zone however is compiler-only features like coopmat2 from Nvidia that AMD still hasn't implemented for years now
>>109106209Why stefan molyneux? or is that someone that looks like him
>>109106284 >coompat2
>>109106225You're fucking retarded. Vulkan runs on GCN 1.0 GPUs from 15 years ago. No they don't support every single extension and instruction, but neither do 12 year old GPUs do that for DX12. Maxwell GPUs were known for notoriously poor DX12 performance even 10 fucking years ago due to missing hardware support for certain features, let alone with modern games making full use of the API. Stop shilling for corporations and learn something, you pathetic cuck.
>>109106272valve has demonstrated through their actions that they do not believe native development is viable anymore on Unix/Vulkan. That is why Proton exists and why they've stopped funding MoltenVK (which translated Vulkan to native Metal). They should be funding KMSDRM/X11, but they've deliberately chosen not to. That's fine, I'm not asking for their money, but that is their opinion.Nvidia obviously is pushing CUDA over Vulkan for AI/ML. That DX12 games exist is just a happy accident for them.>>109106288it's stefan yes
>>109106225>DX12 runs on 12 y/o hardwareLike absolute shit.
>>109106225It's not popular, because it's not designed for humans, it's designed for corporations with infinite resources.Large teams were supposed to use it to create middlewares for actual human consumption, but it never happened.
>>109106354I am not at all saying DX12 is good. I am saying that CUDA and Metal are going to win, which is the opposite of what I want. DX12 and Vulkan are in the same boat>>109106317lmao
this is a very low quality thread, even by nu-4chan standards
>>109106133stop getting filtered and just take the triangle slow, you'll get it eventually! :)
>>109106443>>109106451ive written more vulkan renderers than youve written hello worlds. message me back in 5 years when the API is abandoned by hardware vendors-- 2026 computex is going to be 2031 vulkanized
>>109106474AI can one shot a vulkan rendererno one, and I do mean no one caresmaybe that is what your thread is secretly about? you coming to terms with that fact
>>109106479send me the github link go your one shot vulkan renderer
>>109106479>AI can one shot a vulkan rendererCool. Too bad writing a simple single triangle Vulkan renderer has almost nothing in common with writing an efficient large-scale rendering solution. And even less so with developing novel rendering algorithms...There's a complete disconnect with what graphics developers want and what (open source) graphics APIs provide and Vulkan makes it worse instead of better.
>>109106367
>>109106555i didnt know diddy was on that skooma. i feel for him
>>109106133Vulkan needs to be rewritten in rust.
>>109106367I don't follow what would they actually be using then? What would be the alternative
>>109106584The most straightforward way to render on Linux is KMSDRM + Vulkan/OpenGL (or a software renderer lol). It's unbelievably simple and powerful, and it's literally all the world would ever need. Fortunately nothing is stopping anyone from making applications like this. However, what I speak of is the future. There is going to be a future where Vulkan goes away, where KMSDRM/Linux itself is blocked. And all you will be left with is Apple Metal, CUDA, and if you're lucky, DX12, but even that isn't a given anymore with the literal implosion of the gaming/budget semiconductor industry.
>>109106367Source2 uses Vulkan, and they prefer Vulkan because the steam deck is Linux. Proton only exists as a necessity because of the giant backlog of Windows only games that will never get proper ports. Targeting DX11/DX12 and hoping that DXVK works well enough for your game is not the way. And NVIDIA is greedy but they're not going to drop Vulkan support ever, but yes they've been pushing monopoly technology (RTX,CUDA,PhysX,DLSS,Gsync) their entire existence.
>>109106133>Valve moving away from native APIs.What the fuck are you even talking about?
>>109106615>>109106629forgot photo sorry >>109106584Yes and their offline baker for maps uses the proprietary Nvidia Vulkan RT core extensions. Of course this is just an optimization, nothing is stopping the math from being done on general purpose shaders. So I'm not saying Valve is all bad, they are obviously our best supporter right now. Be careful in assuming that Nvidia will support Vulkan forever. 10 years ago, did we ever think there would be a world without Xbox? We are staring that possibility down right now. Did we ever think we'd have Apple laptops that explicitly block Linux? Did we ever think we'd have Sega using hordes of AI assets?The centralization is intense. Plan accordingly
>>109106667Feel free to stop spamming your AI slop any time, garbage.
>>109106666>>109106367MoltenVK defunding and pushing of Proton are not accidents. What's going to happen when people target Proton for new development?
>>109106679Proton is not a standalone API. Shut the fuck up already.
>>109106679proton and moltenvk aren't interchangeable. Proton isn't something you target for development. MoltenVK is for people who want to make a crossplatform renderer without having to write an entire metal backend on their own. Vulkan already writes natively on Linux / Windows / Switch. It makes no sense to target DirectX and rely on proton for compatibility on Linux.
no one gives a fuck about metal
>>109106133Vulkan has won, in ai. Nvidia has its own, of course, and it's fine.But the #2 is Vulkan. Vulkan actually won because Nvidia will never open theirs, so eventually people will just quit using Nvidia, just because it's too expensive.
>YOU WILL BUY A RADEON SOON OR LATER, GOY
>>109106578no
quite possibly the most retarded thread on /g/
At this point, for a small indie dev, who wants to avoid unreal unity and all this other shit, it's probably better to make your own software renderer, draw the 10k triangles at 540p then upscale to 1080p or 4k.The game is going to be playable forever.
>>109106822Incorrect. Nvidia will be the only company making GPUs left. AMD is already going bankrupt in two weeks.
>>109106133...why is Peppersteps in this picture?? WHAT IS THIS FROM I LOVE PEPPER
>>109107068respectfully vulkan is about as complex as that
>>109107101it's a secret
>>109106209wtf is molymeme doing on a g thread and yes I know the cunt used to do IT
>>109106548>There's a complete disconnect with what graphics developers want and what (open source) graphics APIs provide and Vulkan makes it worse instead of better.what do they want that vulkan doesn't provide properly?
>>109107116not an argument
>>109106367CUDA and Vulkan are entirely different things with different purposes. Yeah, doing CUDA type stuff on Vulkan or other generic graphics API sucks, but CUDA predates them by a long ass time anyway and is already well established everywhere and was basically built for that (and other scientific computing type) problem(s).
>>109106977how many filters do you have for this to be the "most retarded" thread?
>>109107143WRONG
i'll update the list EVs: >China pls save usRAM: >China pls save usSSDs: >China pls save usGPUs: >China pls save usCPUs: >Taiwan pls save usMotherboards: >China pls save usSmartphones: >China pls save usSolar panels: >China pls save usBatteries: >China pls save usE-bikes: >China pls save usDrones: >China pls save usRare earth magnets: >China pls save usConsumer electronics: >China pls save us3D printers: >China pls save usPower banks: >China pls save usUSB cables: >China pls save usSteel: >China pls save usIndustrial machinery: >China pls save usLinux handhelds: >China pls save usBudget gaming laptops: >China pls save usMini PCs: >China pls save usHome appliances: >China pls save usVulkan drivers: >China pls save us
>>109107209i mean, who in america is going to manufacture chips affordably? they cant even manufacture pizzas affordably
>>109107209At some point, we'll ask China to save us keeping track of the list.
>>109107360ACCELERATEWHY CONTAIN IT
>>109107094GPUs are a side gig for AMD. Their real focus is CPUs, and they're not really facing death on that front.
>>109107209GLORY TO THE CCP.
>>109107209China isn't saving anyone from anything. The problem that nobody wants to face is that yes, we're all going back to being poor. Not poor on paper but actually-poor, where you cannot afford luxuries or services of any kind without seriously setting back your future.This is solvable, but only in two ways. Either we dramatically restructure our society to a non-profit driven model that nobody has even conceived of yet (Unlikely, and no, it can't be Socialism because for that to work we have to stop being human) OR we dramatically reduce the amount of people to drastically improve the value of the remaining labor. Basically, you nuke China and India and probably Africa and South America for good measure. Since nobody is willing or able to control their feelings long enough for this to happen, its the long, slow descent into everybody being poor and so it will continue. But the very idea of China saving anyone from anything is laughable - 70% of their population is still illiterate, rural and has no access to electricity or indoor plumbing. If they're not wanting to save their own people from being truly poor, what the fuck makes you troglodytes think they're going to try and save US from being poor?
>>109107659AMD has some huge contracts for their datacenter GPUs especially with Oracle.One of AMD's core businesses is still APU with custom SoCs for Sony and Microsoft and because of that alone will they keep developing GPUs.
>>109107162no he's right this thread is utterly retraded
>>109106284> coopmat2 is literally just a compiler flag and amd still can't get it righthardware is hard but sometimes it's just lazy driver teams
>>109106667> muh hypothetical nvidia vulkan doomsdayvalve literally funds vulkan devs and pushes open source drivers. fearmongering about sega ai assets isn't a tech argument.
>>109107704>where you cannot afford luxuries or services of any kind without seriously setting back your future.nonsenseI just found out this week I actually own several luxury goods
>>109107659Dumb. nVidia's data center revenues are 100X consumer gaming. And they're entering CPU side of the business as well. AMD is going extinct. AMD should be focusing 80% of their core business on GPU, instead, they're off in lala land
>everything will always work fine with OpenGL 4 at the least because otherwise all the software stops workingWhy use anything else? let all these gay companies veer off in their own direction with their own special sauce, knowing they all support opengl and will for decades? lol just ignore 'em
>>109107789>>Dumb. nVidia's data center revenues are 100X consumer gamingall fake money they literally give money to companies and then the companies buy data center use with the money nvidia gave them it's all a scam
>>109107756NOT AN ARGUMENTif valve is funding vulkan dev, why did they kill moltenvk and push everyone towards an emulation layer?native development isn't hard, it's just that 99.999% of corpos programmers are not good at programming. "muh derisking" is all senior software engineers can think about so that they can pay their overinflated goy mortgage
>>109107792OpenGL4 is fundamentally flawed. It is not possible to reason correctly about synchronization, memory allocation, and render passes in OpenGL4.
>>109106133Did you come from a alternative dimension or are you just retarded? Nvidia has improved their Vulkan implementation a lot these past few months and Android will replace OpenGL with ANGLE next year.
>>109107912so how much are we betting that Nvidia says oopsies!! please use CUDA instead HEHE!!!! and dont act like gaming is going to be around in 5 years. theyre going to be giving us all thin clients
>>109106667hearty kek>>109106544is that Stephen Hawking?
>>109108483lost between borders... born in norvey... living in indiaim confused how you mistook caitlin clark for stephen hawking, unless you mean the guy
>>109107773better buy nintendo stock
>>109106667incomplete without the head wobble.
>>109106133>new API with uniform features and well-though out design comes out>everyone loves it>has support for extensions>over time vendors keep coming up with more and more extensions, which they force on users to fuck over their competition>they're badly designed because they don't care or want to soft lock in people by making it a pain to rewrite>that then gets upstreamed because it's "too late to change it now" and/or the API people have no power to override vendors "no this is how it should be">shit gets worse and worse, everyone hates it, until...>new API with uniform features and well-though out design comes outThe circle of life
that's a problem for unreal and blender3d people, not for us.
never expected to see a thread full of ai generated images of people holding vulkan water bottles what the fuck lmao
>>109107704China already saved the world from an energy crisis by driving the cost of sollar and battery storage into the fucking ground. Say "thank you for saving me from power rationing and energy wars, Chiba"
>>109110291Actually, their chink garbage panels, batteries (besides BYD) and charge controllers are a fucking fire hazard.
>>109110414lets see paul allens solar panels. oh wait, he doesnt have any
>>109107793You fell for midwit doomerfag engagement bait.
>>109107789>Nvidia CPUsUnless they use some sorcery on Prism, there is zero chance of x86 going out of style for at least another decade
>>109110891all the servers that nvidia uses, can ship with nvidia cpu inside them. its an all in one package they are going with. instead of putting in 32 core Ryzen in their data centers, they may put in their own 64 core ARM chip which saves them money and extracts the extra profit in developing their own cpu wing of the department. They could also put this out to any nvidia device in the future, like local nvidia inference chips DGX minis. Any future console might go with nvidia combo of nvidia ARM chip + RTX. They can already do x86 and arm. Any program runs there.
>>109106133not gonna lie, i've been porting over some of the AI inferences that are trapped in python over to vulkan in windows executable mode so they run effectively on all gpus. So vulkan is the best right now for AI imo, especially without the hard CUDA requirements and the ROCM nonsense or any other gpu vendor dependent libraries.
>>109110988yes i agree but im saying nvidia is going to turn that off in future hardware releases
>>109106548This>>109107133Vulkan is a Fake low level api. It hallucinates, it believes that the GPU has cores and processes that does not exist. It is a schizo, and in its imagination, has made a crazy web of complexities that the gpus dont abide to. OpenGL from 40 years ago is better at creating a abstraction for general parallell computing. Emphasis on General.
>>109106133>Consoles going extinct.Not enough people seem to understand this.
>>109111704>it believes that the GPU has cores and processes??????????????All it does it give you access to blocks of heap memory and a bunch of attributes about what you can or can't do with it.Stop getting filtered and making up nonsense
>>109106133OpenGL chads just can't stop winning
>>109106667Heck, did we ever conceive a world where government agencies would try and price us out of the gaming toy consumer market?
Just realized I missed an opportunity
>>109113663LMFAOOOOO
>>109113663formerly chud's
>>109113036Oh god shut the fuck up. If you have ACTUALLY tried to learn Vulkan, then you would have seen 20 different graphs that contain the same complexity as pic related, you GAY Jew. Do you think that this represents something real in the actual hardware??? This is lower level than the hardware itself. Gpus, are a row of cash-memory, ready to be sendt to the output, and some processors that are good at multitasking, and some ram that stores data sendt from the cpu, with some shit bolted on to make it faster. Fake low level
>>109114613LE HECKIN RESOURCE TRANSITION AND MEMORY ALLOCATIONS ARENT REAL, THEY CANT HURT YOUJUST SWITCH PIPELINE STATE AND THROW 8 THREADS AT THE GPU SIMULTANEOUSLY, NOTHING CAN GO WRONGIF YOU SEE SOMETHING WRONG, YOUR USE CASE?????
>>109110291????Evs are more expensive than gas.
>>109115432???
>>109115432>JUST SWITCH PIPELINE STATE AND THROW 8 THREADS AT THE GPU SIMULTANEOUSLY, NOTHING CAN GO WRONG>THROW 8 THREADS AT THE GPU
>>109114613yet another retarded faggot who thinks he understands hardware better than those making the api. another level of delusion is need for you to talk such retaded shit. the entires in the graph are literally structs. nearly everything in vulkan is just fucking structs with configuration. you just make a bajillion of structs, allocate some memory, bind it to some buffers some and do synchronization shit before upload things to the queue with command buffers from a command buffer pool. vulkan isn't inherently complex, its pedantic and anal, its just that nearly all things you would do with vulkan are complex. i can't fathom how you can be so retarded that you both get filtered and then go spewing shit. go back to hello world kike
>>109116386>>109116409throwing 8 threads at the GPU in this context means having 8 threads recording into the same command buffers simultaneously which are being submitted to the GPU. while some buffers are being read by the GPU, their state is overwritten. bad things happen as a result
>>109106133'jak 'em
All I know about vulkan is it was going to make gaming on linux easier and faster but since I can't afford a gaming PC anyway it really doesn't matter.
>>109117475>the entires in the graph are literally structs. nearly everything in vulkan is just fucking structs with configurationHow the fuck exactly is using their arbitrary strucks "low level".>>109117671>throwing 8 threads at the GPU in this context means having 8 threads recording into the same command buffers simultaneously which are being submitted to the GPU.Okay, Rabbi, so I should let 8 cpu threads do drawcalls to the gpu in parallel, and then "nothing can go wrong"
Why is there no graphics api that's as simple as:draw this vertex buffer with this shader pleaseI tried following the vulkan tutorial and noped out long before I was even close to drawing a triangle. I want to draw polygons. I don't want to write a hardware driver
>>109118051OpenGL
>>109118014>How the fuck exactly is using their arbitrary strucks "low level".it is low level because you have control, and structs are low level, they are literally just regions of memory that you label in the code. the api is very consistent and by all means low level. nobody in this thread has even done hello triangle in vulkan, only fucking retards
>>109118257Okay, mrs. Expert. Is pic related also "low level structs"?
>>109118257How about this aswell
>>109118257You really have to explain this shit to me, because I dont get it, and I really dont think this represents something real in the gpu>>109118293>>109118298
I like opengl a little bit more>>109118293larp
>>109118293>>109118298yes it is, you have control over everything, low level doesn't mean simple, in most cases it means the reverse. video cards are inherently complex pieces of hardware and require complex configuration. what do you want "low level" to be for a graphics card. these graphs show dependencies between the different structs and object handles. it is never the less an abstraction over the graphics card, and PipelineLayout can be considered a bit "higher level", but vulkan is lowest level graphics API for all means and purposes. for lower level you start considering individual series of GPU, but they are so closed down that you have no access.
>>109118293Usecase for tessellation and geometry shaders?
>>109118014i explicitly said things WILL go wrong
>>109118377WRONG, that is the graphics pipeline, and it is a incomprehensible mess and a web of technical dept. I love using structs btw, but in openGL the stucts are not knee deep in shit, just up to the ankle.
>>109118318>dont get it, and I really dont think this represents something real in the gpuwell it is an abstraction over the hardware that gives access to all practical uses for a gpu and beyond. it is the gpu's job to do things with pipelines, buffers, etc as these are well known concepts in graphics progrmaming. you don't have hardware access, because gpus are closed down and you code becomes gpu speicific. vulkan is implemented on all gpus for the last 15 years and is a graphics api. its job is to give low level access and configuration options to for mainly graphics and compute and it does its job well. you can consider nvidia cuda lower level, but good luck doing graphics with cuda.
>>109118298>>109118318it's right there in the top left of your image. things that are physically in the GPU are heaps and queues. heaps let you store data, aka textures and arrays of numberd. queues let you submit data/code/commands from the CPU over PCIE. the only thing not mentioned here would be compute-- instead it is implicit.everything else in the vulkan API is a logical, non physical construction intended for the programmer to express their intentions to the GPU for how they want things to be processed.
>>109118421if you have brain larger than your balls you will see and appreciate the complexity of the graphics pipeline and the amount of control you get by writing 1000 lines of just struct initialization. it is pedantic and a bit annoying but that is its strength. it is inherently complex and your lack of understanding and constant coping is sad at this point.
>>109118432you can argue that the other concepts are at least partly physical because it is the job of a graphics card to have a graphics pipeline, etc, and i am sure there is specialized physical logic on the die that is related to the graphics pipeline. i think the most non-physical part of vulkan is the interfacing with windowing systems and the concept of a surface
>>109106133I never really thought of it that way before. Actually true. Jonathan Blow's new game uses DX12 exclusively, doesn't support macOS, and lets Proton take care of Linux.I can't say where DLSS fits into all of this. It kind of just adds to the chaos with no meaning behind it.
>>109117475Your carefully crafted 2-page structs are then immediately thrown away by the driver that does its own heuristics about what you're actually trying to do, because creating and maintaining legit Vulkan implementation is too complicated and cumbersome even for multibillion-dollar graphics card manufacturers.
>>109118472They were trying to support all existing hardware by putting thousand of switches and toggles for developers to set and in doing so they created a monster that is too big for both sides - software devs and graphics drivers devs - to handle.
this entire thread is 1 schizo who got filtered by vulkan and is making up falsehoods to justify his desire for its demise. I'm sure you know better than the entire graphics community.
>>109119134>too expensive to write a proper vulkan driveryet again the jews stop us from having good things
>>109119487It's not Jews, it's simple maths.Let's say you have 100 boolean switches (in reality Vulkan has a lot more, and in both dimensions - more switches, some of them are more than boolean).Suddenly you get 2^100 of possible states for your graphics card setup.And sure - most of it gets pruned due to not being applicable, but still you get an astronomical number of possible states you - as both software and driver developer need to account for.And a lot of them are difficult, not straightforward decisions:>hmm, it's not really what we expect with how we do things in our hardware, what should we do?>do we return an error immediately?>do we show a warning and do it in half-assed very unperformant way>or maybe in this case we should support it - since it can be emulated and very efficiently, we just need to write 10'000+ lines of code to do so in our driver
>>109120650Nah, ever heard about a COMPILER. This shit should have been taken care of. And yes, it is probably the jews fault as well. Vulkan is fake low level, lower level than the gpus themselves, and a hassle to abstract and work with. The shader the users are writing should just adapt to the gpu, and be "molded" accordingly.
>>109118873DLSS is just the industry's attempt to not pay artists and get you to pay for slop>>109119487>>109120650>>109120815vulkan drivers are the easiest thing in existence to write. it's the application developer's problem now
>>109121390They would be if Vulkan matched what real hardware does.But it doesn't. It doesn't match anything. It's a monster attempting to encompass at a low level* all GPUs with vastly different low-level implementations and in the result creating something that doesn't match anything.*but abstracted low-level, not letting me use native API directly, no, no, it wouldn't be kosher and portable, doesn't matter Vulkan isn't really that portable anyway
>>109121524that hardware vendors don't expose the actual way to program your actual GPU's assembly is a failure of the hardware vendors, aka Nvidia, AMD, Intel, Samsung, Imagination...
>>109121538I think it's Microsoft forcing them to adhere to "standards" as to not create fragmented market like it was during Glide times.But Khronos is a committee of incompetent out of touch retards that does a worse job than a random singular amateur would do at creating library unifying native APIs if given access to.
>>109121570Khronos is themselves mostly AMD, Nvidia, Intel, ARM etc. There is some level of architecture astronautism and fart sniffing, but it's not that bad. The C++ people are way, way worse.The true travesty is that you cannot simply write assembly for your GPU and receive some type of trustworthy information for how it will be executed for you to reason aboutTo fix that, either reverse engineer a device of your choosing and buy 10 of them, or, unironically make your own GPU, which is becoming more and more possible and necessary by the day ^.^
