you are viewing a single comment's thread
view the rest of the comments
[–] [S] 11 points 1 day ago* (2 children)

A lot of distros enable ZRAM by default. Some don't (such as Ubuntu apparently). If not, no effect

Otherwise, the worst case scenario is thousands of times improved latency, and improvements across the board

This means faster response time. For gamers, this might mean less stutters. For DJs it might mean the music doesn't skip.

It's hard to know what impact on which apps. It might be negligable, and memory constrained systems with fewer cpus are probably the biggest winners. But if the benchmarks don't have regressions elsewhere, every system will benefit

In practice though, I'm seeing so many posts for kernel 7.4 with huge improvements, that it seems that kernel 7.4 will be overall noticeably faster for most it feels like

And whilst the FuckAI crowd would moan about it (and I'm not really a fan of AI), in reality, many of these optimisations were likely caught by AI making test harnesses and benchmarking sections of code, analysing the profiling data and identifying slow algorithms.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 5 points 1 day ago (1 child)

    many of these optimisations were likely caught by AI making test harnesses and benchmarking sections of code, analysing the profiling data and identifying slow algorithms.

    Do you have a particular reason to think that, or is it speculation? If the former, can you share some links?

  • source
  • parent
  • hideshow 1 child comment
  • [–] [S] 3 points 1 day ago

    Yeah.. Absolutely. I have a few references:

    1. https://www.phoronix.com/news/Faster-Kernel-Builds-AI-v2 (this only affects building speed)
    2. We know its being HEAVILY used for everything: https://www.phoronix.com/news/Linux-7.3-rc6-Released
    3. We know Linus is using it: https://github.com/torvalds/linux/commit/818bebeb63dd6bf5f4e07e145f6cdbace520a34c

    I also recall seeing another article, but admittedly unable to find it where AI isolated a kernel bug that reduced latency too (but maybe someone else had it). I thought it had something to do with file handling. Without that though, it could be speculation, and bad memory on my side.

    That being said, we (unfortunately) use AI at work, and it is good at finding performance bottlenecks which even profiling won't find. I'd be very surprised if no devs used it kernel side for the same reason

  • source
  • parent