NVIDIA CUDA

“Cuda has shown that, when done right, defaults and a convenience layer can make for a well received API without sacrificing performance.”

“messes with your build system, ties your build system to platforms that have their toolchain binaries available”

“I do mind that all of the SDKs don't dynamically load the various CUDA shared libraries at runtime”

“everything is open source except CUDA itself. So any issue that happens on the CUDA level, you are forced to wait for them to fix it.”

“All wannabe replacements fail, because they focus only on the CUDA C++ compiler and nothing else.”

“The next 20 years of cuda support weren't luck, as anyone trying to use AMD will know.”

“Trying to bake CUDA assumptions into model design, and pushing the software ecosystem to be as Nvidia first as they can.”

“it was so niche that you ended up compiling from source literally everything since no wheels were available for it”

“The CUDA runtime coming with a gazillion reasonably decent kernels (DNN, BLAS, CUTLASS) and a concurrency system (NCCL) is a big deal”

“its not super easy to use, but unlocks orders of more magnitudes of power for the use cases it excels at”

“GPU architectures are supported by the full tool suites for at least 10 years by CUDA”

“nvidia is not interested in maintaining any semblance of backwards compatibility when it comes to CUDA”

“I had actually a ton of success using Strassen turboplicattion kernels with some extra custom CUDA”

“With a design like Cuda where there is always a default easy path, and a complex but optional path.”

“No ridiculous descriptor management and binding in cuda, just passing pointers and handles via launch arguments.”

“Ease of programming and a giant leap in performance is one of the key reasons for the CUDA platform's widespread adoption”

“NVIDIA developed CUDA and (perhaps more importantly) an entire GPU software ecosystem around it.”

“CUDA has support for C, C++, Fortran, Python JIT, and anything else targeting PTX, graphical GPU debugger”

“I'm asking you to imagine how much easier everything would have been if Cuda was cross-platform without any performance or ergonomic penalties.”

“the compute capabilities from 2014 are still instrumental for accelerating training and acceleration workloads”

“got the same stuff done in less than a day that I could not do in a whole week in Vulkan”

“the worst privacy, transparency, and FOSS integration due to their insistence on a heavily proprietary stack”

“CUDA’s moat isn’t just writing kernels, it’s the whole ecosystem + hard earned perf intuition.”

“CUDA feels really cozy, because like Microsoft, NVidia understands the Developers, Developers, Developers mantra.”

“CUDA is far-and-away the better hardware architecture, not to mention better-supported by the community.”

“You don't need advanced C++ knowledge anymore to write and run a CUDA project anymore.”

“Which is way easier because Cuda always provides the simple API and makes complexity opt-in.”

“Cuda allows to easily get stuff done first then check the more complex stuff to optimize”

“Raw CUDA works for the heavy lifting but I suspect it gets messy once you implement things like grammar constraints or beam search.”

“CUDA, by comparison, is having zero expense spared in it's development at every layer of the stack”

“This is exactly yet another reason why researchers prefer CUDA to the alternatives.”

“there is a huge library of CUDA code that will run out of the box on NVIDIA hardware is a big advantage”

“Just because it's still called CUDA doesn't mean it's portable over a not-that-long of a timeframe.”

“training requires the mature ecosystem and numerical stability of cuda to handle mixed-precision operations”

“Cuda shows that allocation of GPU memory and copy from host to device can be one-liners”

“CUDA isn't all that and a bag of chips. It just is the Facebook/Twitter of the data science and from that LLM space.”

“nvidia libs are broken with clang-21 and recent glibc due to stuff like rsqrt() having throw() in the declaration”

“Newer hardware may require a CUDA toolkit version that is newer than what the project maintainer intended.”

“CUDA focused not just on keeping one backward compatible one, but also made it work from cheapest NVS card to high end parts”

“it's still the easiest to use, with most available resources when you inevitable need to dig through stuff”