# NVIDIA CUDA Hacker Approved page: https://hackerapproved.com/p/cuda/ 65 people recommended NVIDIA CUDA in Hacker News comments, 22 had mixed feelings and 69 criticized it, November 2025 to October 2026. It was recommended in 12 of 12 months, in 76 threads. It does not meet the Hacker Approved badge rule. **Verdict:** Slightly more people criticize it than recommend it. Developers recommend CUDA for its mature ecosystem, easy defaults and performance on NVIDIA GPUs. Main caveat: it only runs on NVIDIA hardware and is not portable. Summary written by Claude from the comments below. ## What people warn about - Build and toolchain friction (4 comments in the sample): "I do mind that all of the SDKs don't dynamically load the various CUDA shared libraries at runtime" ([source](https://news.ycombinator.com/item?id=49735545)) "it was so niche that you ended up compiling from source literally everything since no wheels were available for it" ([source](https://news.ycombinator.com/item?id=49429922)) "nvidia libs are broken with clang-21 and recent glibc due to stuff like rsqrt() having throw() in the declaration" ([source](https://news.ycombinator.com/item?id=45993598)) - Locked to NVIDIA hardware (3 comments in the sample): "I'm asking you to imagine how much easier everything would have been if Cuda was cross-platform without any performance or ergonomic penalties." ([source](https://news.ycombinator.com/item?id=48072670)) "Trying to bake CUDA assumptions into model design, and pushing the software ecosystem to be as Nvidia first as they can." ([source](https://news.ycombinator.com/item?id=49460458)) "Just because it's still called CUDA doesn't mean it's portable over a not-that-long of a timeframe." ([source](https://news.ycombinator.com/item?id=46075228)) - Breaks backward compatibility over time (2 comments in the sample): "Newer hardware may require a CUDA toolkit version that is newer than what the project maintainer intended." ([source](https://news.ycombinator.com/item?id=45990798)) "nvidia is not interested in maintaining any semblance of backwards compatibility when it comes to CUDA" ([source](https://news.ycombinator.com/item?id=49263360)) - Proprietary closed stack (2 comments in the sample): "everything is open source except CUDA itself. So any issue that happens on the CUDA level…" ([source](https://news.ycombinator.com/item?id=49733954)) "the worst privacy, transparency, and FOSS integration due to their insistence on a heavily proprietary stack" ([source](https://news.ycombinator.com/item?id=47746068)) ## Why people recommend it - Easy defaults with optional complexity (6 comments in the sample): "Cuda has shown that, when done right, defaults and a convenience layer can make for a well received API without sacrificing performance." ([source](https://news.ycombinator.com/item?id=49739229)) "With a design like Cuda where there is always a default easy path, and a complex but optional path." ([source](https://news.ycombinator.com/item?id=48910260)) "Which is way easier because Cuda always provides the simple API and makes complexity opt-in." ([source](https://news.ycombinator.com/item?id=46968919)) - Mature ecosystem and large library (5 comments in the sample): "The CUDA runtime coming with a gazillion reasonably decent kernels (DNN, BLAS, CUTLASS) and a concurrency system (NCCL) is a big deal" ([source](https://news.ycombinator.com/item?id=49267597)) "training requires the mature ecosystem and numerical stability of cuda to handle mixed-precision operations" ([source](https://news.ycombinator.com/item?id=46070441)) "it's still the easiest to use, with most available resources when you inevitable need to dig through stuff" ([source](https://news.ycombinator.com/item?id=45878395)) - Long backward compatibility across GPUs (4 comments in the sample): "CUDA focused not just on keeping one backward compatible one, but also made it work from cheapest NVS card to high end parts" ([source](https://news.ycombinator.com/item?id=45941521)) "the compute capabilities from 2014 are still instrumental for accelerating training and acceleration workloads" ([source](https://news.ycombinator.com/item?id=47755144)) "The next 20 years of cuda support weren't luck, as anyone trying to use AMD will know." ([source](https://news.ycombinator.com/item?id=49514367)) - Strong performance on NVIDIA hardware (3 comments in the sample): "CUDA is far-and-away the better hardware architecture, not to mention better-supported by the community." ([source](https://news.ycombinator.com/item?id=47092024)) "its not super easy to use, but unlocks orders of more magnitudes of power for the use cases it excels at" ([source](https://news.ycombinator.com/item?id=49265154)) "got the same stuff done in less than a day that I could not do in a whole week in Vulkan" ([source](https://news.ycombinator.com/item?id=47749201)) ## Other GPU compute frameworks people recommend - [AMD ROCm](https://hackerapproved.com/p/rocm/): 26 recommend, 17 mixed, 86 criticize - [BarraCUDA](https://hackerapproved.com/p/barracuda/): 15 recommend, 5 mixed, 4 criticize - [Triton](https://hackerapproved.com/p/triton/): 13 recommend, 5 mixed, 2 criticize - [ZLUDA](https://hackerapproved.com/p/zluda/): 10 recommend, 3 mixed, 6 criticize ## The badge A product earns the Hacker Approved badge when at least 10 people recommend it, recommenders outnumber critics at least 2 to 1 (mixed feelings count toward neither side), it was recommended in at least 4 of the 12 months and in at least 5 different threads, and no single thread supplies more than half of the people who recommend it. Health and money products are never badged. ## Cite as Hacker Approved, "NVIDIA CUDA", https://hackerapproved.com/p/cuda/ (65 recommend, 22 mixed, 69 criticize; Hacker News comments, November 2025 to October 2026). Counts updated 2 Nov 2026, with data through October 31, 2026.