Comments by "Vitaly L" (@vitalyl1327) on "Game Devs Are Drowning In Complication" video.

So, you're trying to say that in the past developers were competent, but now it is a beautiful egalitarian field where complexity is removed from the developers education and manifests now in the overengineered crap those egalitarian uneducated developers create? Fully agree.
4
There's is SPIR-V though.
2
And, just like pretty much any other branch of this industry, self-inflicted complications. The real complexity of most of the problems is orders of magnitude below what bad engineers make them to be.
2
LLVM IR is not platform-independent at all. There was an attempt to build the previous SPIR standard on top of LLVM IR, it failed. SPIR-V is the way now, and it's got nothing to to with LLVM IR. Now, even if you go a few levels up, shader source code, or OpenCL kernels, while semantically portable, are not performance-portable anyway. GPU architecture specific optimisations have to be done on a very high level and cannot be automated.
1
They are not. Most of the problems code monkeys solve are trivial, but code monkeys cannot think, so they make them orders of magnitude harder than they should have been. Code monkeys cannot build things from scratch, they combine dozens of third party dependencies and then they're lost in tacking the emergent complexity of the Freankestein monster they created of a pile of necrotic pieces.
1
@thomassynths lol, you're funny. Raytracing is algorithmically simpler than the traditional rendering. Modern unified shader GPU architectures are simpler than the old specialised pipelines GPUs. You clearly have no idea of what you're talking about.
1
@thomassynths then you should know how much simpler CADs became now, as we got a lot more parallelism. Now SDF and other implicit solid geometry became viable, no need to do all that mind-blowing complex B-rep geometry. I worked on a CAD core for years too, decades ago. Millions of lines of Fortran code. These days I can reproduce all the same functionality in a few hundred lines of code without any mathematical complexity. All with a photirealistic path tracing rendering, fast FEM simulations and all that jazz.
1
@thomassynths hardware became much simpler - I worked on a few generations of a certain mobile GPU, and with every new iteration we removed a lot of hardware hacks, because we could simpler have larger and wider FIFOs, larger SRAM blocks, more ALUs, and so on. Simpler and more straightforward hardware at a larger transiator count cost. Same for drivers, they are literally snippets of code running on the GPU shader cores now, no complex register setup, no moving data around. Everything is getting massively more simple.
1
@thomassynths Look, I design those chips. They are logically simpler, they are simpler in all known definitions of complexity, Kolmogorov-Chaitin and Shannon included. They are only more "complex" in terms of transistor count. Repeating the.same block many times does not multiply complexity. GPUs are now just simple.arrays of primitive CPU cores with tons of caches, scratchpad memory blocks and interconnect. Gone.are the days of complex geometry cores, specialised single-function pipelines, and all that.
1
@thomassynths you really are hopeless. Look, there are tons of open source designs these days, go, take a look. See the modern OoO RISC-V cores for example. Compare to the open older cores, such as Niagara. You clearly have no idea of what you're talking.about. You have so much more to learn.
1
@thomassynths firstly, everything I say is easy to verify. If you had qualification to read HDL, lol. Secondly, you have reading comprehension issues. I am not claiming that the new hardware is scaled old hardware, quite the opposite. Old hardware is excessively complex, because limited transistor budget forced clever hacks and workarounds. They are not needed any longer, so the new hardware is much more straightforward and streamlined. Even the arm cores, with all the legacy cruft they are still carrying, are getting easiser and easier to comprehend, complexity peaked at the Eagle time.
1
@thomassynths you're truly pathetic. Go, read the BOOM code, and compare it to Niagara. Then come back. Your "experience" is evidently worthless, for you already demonstrated that you have no idea how GPUs work, how GPU drivers are organised, how modern hardware is designed, and, most of all, you do not even understand what complexity is. I even doubt you have any idea of what Kolmogorov metric is and how to approximate it. You should have stayed away from the grown up discussion, kiddo.
1
@thomassynths you're an embodiment of Dunning-Kruger. You have no idea whatsoever how modern GPUs and CPUs work, how drivers are organised, and even what complexity is.
1
@thomassynths you very evidently do not understand how modern hardware is built.
1
@thomassynths you're getting increasingly hilarious. Now this little kid "design HW". Not long ago you could not even read BOOM source code and compare it to Niagara, and now you "design HW". Mind naming that HW cores you "design", kiddie? I worked on Mali and VC GPUs, through a few generations of them. I was on several Khronos committees, overseeing the trend to hardware simplification and moving of the vanishing complexity to software side. What "HW" did you "design", kiddo?
1
@thomassynths what are you babbling about, kiddo? I challenge you to read BOOM and Niagara source code in Verilog exactly for you to compare how much of the complexity is gone in the modern hardware, everything is very straightforward now, no need for clever hacks any longer. Mind naming a single GPU that you designed?
1
@thomassynths so, crickets. What was that GPU you've designed?
1