Steam is going to be heavily skewed Nvidia since many games either need it or run poorly without it. Gamers seem to tolerate an otherwise nearly unusable computer for Nvidia graphics acceleration where most users won't.
Steam merely confirms what I already knew as a long-term Linux user: when it comes to discrete GPUs, NVIDIA has long been the biggest player. This isn't just Linux-specific either, NVIDIA commands a large portion of the GPU market period. Forgetting gamers, NVIDIA has the most robust options for hardware acceleration in many cases, especially for AI use cases like PyTorch, and a lot of productivity tools like Blender. ROCm and HAP exist, but AMD has often struggled to gain widespread support.
If there's any strong evidence that NVIDIA still isn't winning in Linux desktop marketshare, I'd certainly be interested to hear it, and it would be news to me. I can imagine Intel technically has a lot of marketshare due to laptops, but then again, Intel and NVIDIA users are hardly mutually exclusive on laptops, either.
I don't really believe the majority of desktop Linux users have Steam installed. I do believe that the majority of Linux users historically (at least leading into the 2010s) used NVIDIA GeForce graphics cards, for a multitude of reasons. Even today where AMD and Intel GPUs are clearly the better option for stability, that doesn't mean they're the best option for doing work with. If anything, gaming on AMD is pretty comparable today, in part thanks to Valve. Productivity on AMD is where things are not in good shape. I wish it weren't so, but compare the CUDA and OpenCL/ROCm ecosystems and it's pretty obvious. Most recently, when AMD released some pretty enticing GPU options into the market, ROCm users were surprised to find that their cards either were entirely unsupported, or support had not yet been stabilized. That's just not how things work with CUDA.
I'd pay for things to get better here. I desperately hope that Vulkan as an option for compute can mature and compete with CUDA, but I also think that to do so, quite an ecosystem will need to be built up around it, even if the API and drivers prove sufficient to work as a solid runtime.
I notice it I just don't care at all. Latency is a much bigger deal to me and it may not be technically part of Wayland or the widget toolkits used by compatible apps but it is consistently a problem.
If latency is a real problem for you, buy a monitor with higher refresh rate. That’s the only meaningful way to decrease latency on modern software. Mind you, literally every single program you use will double buffer, unless you are using XMotif or so exclusively.
I actually do use a lot of XT/Motif apps. Mostly because of the lower latency but also they weren't infected with whatever brain damage generated the gtk3 design language.
I used plan9 extensively (Almost all my machines including my mail server and primary desktop) before I went to college. That doesn't mean its ready for most people to use.
Yeah. It won't matter. Self hosting powerful models has become trivially easy. Training them is quickly becoming similarly easy.
You're just about in a situation where stopping widespread use of AI requires confiscation/monitoring of everyone's personal computers. It won't be enforceable.
For a long time I renewed my domain name with crypto because my bank's anti-fraud system got hung up on it for some reason (it does that with crypto too but buying more crypto is less time sensitive.)
I used it when originally trying jmp.chat out so I wouldn't have to worry about recurring charges/fraud etc.
I've bought some electronic components with it just to see it work.
Meh. Webassembly has polyfills. I think an incremental approach wouldn't be as hard as people make it out to be but someone does have to sit down and do it.
