That’s fair, it has changed a bit since then and I’m hoping we get another filesystem benchmark to see if it has improved, and the caching features might offset that on frequently used data, but I don’t know how hard that would be to benchmark.
In my use it has been pretty stable so far with 7 disks participating (3 caching SSDs, 4 mechanical disks, with 3 copies of metadata and 2 of data), but I’m not using the more experimental features like erasure coding, I will note the on-disk-format has changed twice since it has been in the kernel, and it hasnt been there long, but it has succeeded both on-disk-format upgrades without obvious data loss, and it recently got self healing for some checksum errors, Id say its probably ready for use if the data is backed up, replaceable, or can be gone without (so for me games are all I have that fits this). Otherwise I would use caution if you use it, but I am very optimistic about the future of the FS, as Kent Overstreet (the creator) has taken a lot of care with it.
I use bcachefs for my games, I like that it lets me have multiple disks with redundant data copies, plus ssd caching of frequently accessed files, this fs is linux specific for now as far as I know, and is still experimental. I use ext4 for everything else, and FAT32 for flash drives.
Iirc they added raid1c3 raid1c4 etc to make raid 1 work with more copies.
As you mentioned Ubuntu’s Night light, f.lux, and Redshift all work more like a color temperature adjust than like a red only mode, I found some people mentioning if you are running X you can use xcalib to set the color channels individually, but couldnt find a tool for it, not entirely sure wine would make that function work correctly but it is worth a shot, as for wine, if the version in the package manager isnt new enough there is also a PPA for Ubuntu for more recent wine versions, but I havent used those in a long time and cant strongly advise them, YMMV in installing them and keeping the system working long term, but I was always the sort with too many PPAs so I switched to arch to not deal with that.
Link to xcalib discussion here.
If it isvalready equipped with an Intel WiFi card you should be good out of the box. Glad I could help!
Do want to add, on the CF-30 when I replaced the WiFi card with a newer Intel card I had to shave off the power control mini pcie pin so that the BIOS whitelist couldnt deactivate it, no clue if they still whitelist WiFi cards in newer models or the CF-33.
No experience with the CF-33, but I did use a CF-30 and as far as I could tell, outside the WiFi (back when drivers for that were a problem) everything on it worked out of the box, never had any accessories that plugged into the more proprietary connectors but everything I ever plugged in worked, including RS-232
A program that can run on 1GB but uses 2GB is more wasteful, OS and FS level caching and memory reclamation only work if the memory is available, and a program wasting it takes it from everything else, unused RAM is wasted, but so is RAM being used for no actual function.
Not to say programs cant use large amounts, but they should provide a level of functionality for the amount of memory used, and some programs of late have been more than a bit inefficient, in short, filling the RAM is good, but do make sure its actually being used.
Exactly, your program using the minimum of RAM allows more for other programs to run and gives more memory for the OS to cache literally anything that isnt their web app, likely the filesystem, and that is a much better use of the RAM then letting electron or some such eat it all.
I remember under pulse I would have issues of programs like discord and my headset breaking the connection over the switch between A2DP and HFP or HSP or whatever the mic mode was. Havent had any issues since pipewire came along and supposedly took over handling that, but I havent used a Bluetooth device with a mic to test with since, so I’m just quoting hearsay that pipewire fixed that.
A good portion of this though is the CUDA stranglehold nvidia has. Good luck getting a neural net accelerated on OpenCL or Vulkan Compute.
Oh, didnt know that was posted, some of the tests seem to be different from last time, it hasnt regressed but hasnt improved much yet either (from the ones that were the same). It does seem to have pulled ahead of BTRFS since the last test, doubling score in the DBench test, but it still varies a lot compared to the other filesystems it seems, doing worse pretty regularly.