Bolt

t story compares? https://github.com/HSA-Libraries/Bolt https://github.com/NVIDIA/thrust in the last few years they have (once again) dumped everything and started over, ROCm supported essentially no consumer cards and rotated support rapidly even in the CDNA world. It offers no binary compatibility support story, it has to be compiled for specific chips within a generation, not even just "RDNA3" but "Navi 31 specifically". Etc etc. And nobody with consumer cards could access it until like, six months ago, and that still is only on windows, consumer cards are not even supported on linux (!). That's fine for non-commercial and HPC, but it makes it difficult to imagine ever shipping commercial products to end-users on it. I guess you... ship as source, and your users have to figure out ROCm and compile it in-situ? Or ship a binary for every single GPU to ever release (including NVIDIA, Intel, ....) https://geohot.github.io/blog/jekyll/update/2023/06/07/a-div... This is on top of the actual problems that still remain, as geohot found out. Installing ROCm is a several-hour process that will involve debugging the platform just to get it to install, and then you will probably find that the actual code demos segfault when you run them. AMD's development processes are not really open, and actual development is silo'd inside the company with quarterly code dumps outside. The current code is not guaranteed to run on the actual driver itself, they do not test it even in the supported configurations, because all the development is happening on internal builds etc, not release code. oh and AMD themselves isn't using the open driver of course... they only test and validate on AMDGPU-PRO. it hasn't got traction because it's a low-quality product and nobody can even access it and run it anyway, let alone ship commercial products on it. and now everyone regrets it because it's turned into a money fountain.

ll kinds of 2d/3d/higher-dimensional memory access). And AMD has a history of abandoning these projects after a couple years and leaving them behind and unsupported... like their Thrust framework counterpart, Bolt, which hasn't been updated in 8 years now. https://github.com/HSA-Libraries/Bolt The old bit about "Vendor B" leaving behind a "trail of projects designed to pad resumes and show progress to middle managers" still reigns absolutely true with AMD. AMD has a big uphill climb in general to shake this reputation about being completely unserious with their software... and I'm not even talking about drivers here. This is even more the widespread community perception with their GPGPU/ML efforts than with their drivers. http://richg42.blogspot.com/2014/05/the-truth-on-opengl-driv... AMD doesn't have a library of warp-level/kernel-level/global "software primitives" like Cuda Unbound or Thrust either. So instead of writing your application, you are writing the primitives library, or writing your own poor implementation of them. https://github.com/NVIDIA/cub https://github.com/NVIDIA/thrust It's just a fractal problem of "the software doesn't exist and AMD would really rather you write it for them" all the way down and nobody wants to do that instead of doing their own work. AMD is the one who benefits from the rewrite, for everyone else it's a "best case scenario it works the same as what we've already got", so if AMD isn't gonna do it then pretty much nobody else is gonna leap on it. And then AMD has poor adoption and no software and the cycle continues. AMD really really just needs to get serious and hire a half dozen engineers to sit there and write this software, cause it's just not going to happen otherwise. It's a drop in the bucket vs the sales to be realized here even in the medium term, like one big ML sale would probably more than pay those salaries. They're not doing it because they're cheap or they're doing it because they're not really serious, take your pick, but, AMD is no longer that broke, they can afford it and it makes financial sense. Again, not a "nice" thing to say but it's the cold truth here. I feel like I've made some variation on this post about every 6 months for like 5 years now but it's still relevant. If you as a vendor don't care about writing good code for key features/libraries for your product, nobody else is either, and you'll never get uptake. It's the same thing with AMD/ATI not embedding developers with studios to get those optimizations for their architectures. Console lock-in will only get you so far. If you don't care about the product as a vendor, nobody else will either. It's remarkable how much flak Jensen got for "NVIDIA is a software company now" back in 2009, and how people still don't get it, AMD is not a software company and that's why they keep failing. Writing the framework that turns into StableDiffusion and sells a billion dollars of GPUs is the NVIDIA business model, AMD keeps trying to jump straight to the "sell a billion dollars of GPUs" part and keeps failing.

stically annoying but not really complex. But in general it's also up to the vendor to implement their own SDK, compiler, etc, and there's no guarantee that what runs on one OpenCL platform will run on another. What is commonly referred to as "OpenCL" is actually the AMD APP SDK and the NVIDIA OpenCL SDK, which both implement their own version of parallel C++ dispatched according to the OpenCL function interface/semantics. So in other words, OpenCL suffers from a bit of Architecture Astronaut syndrome. AMD solved the problem of making a framework that can dispatch tasks to any coprocessor that implements their interface, NVIDIA solved the problem of making a GPGPU language. As such, AMD sometimes struggles to fulfill the cross-platform promise of their language and suffers in conciseness. For learning CUDA, I really like the Thrust language. You can write GPGPU programs that are literally 10 lines or less. It looks like regular C++ STL but you can execute it on the GPU. You can easily call __device__ functions using functors iterating across your dataset, or get raw_pointers into a device_vector and then pass them into a normal kernel launch, so there's a lot of flexibility there to tune stuff as you want to make it more complex. http://docs.nvidia.com/cuda/thrust/index.html For a non-trivial program, here's my master's thesis, which I did in Thrust. Not really machine learning but at least you can see what you're getting into. https://github.com/pholvs/PandemicThrust For AMD, there is supposedly an equivalent called "Bolt Framework" but I have not used it personally. As is customary for AMD's libraries, after release it was abandoned and it has not been updated in several years, best of luck. https://hsa-libraries.github.io/Bolt/html/ https://github.com/HSA-Libraries/Bolt