Mac shipments grow 10%, as all major PC brands see downturns

I definitely remember seeing a tv+ signup with I first got my M2. After setting up the device.

If you're not signed into icloud, well you still have all the apps that want you to sign in. Same goes with the tv app.

I use the music app on iOS with mp3s and 3 of the 4 main app tabs are screens trying to sign me up for apple music.

I do see the windows stuff as more upselling to other microsoft apps like onedrive which to me doesn't seem so different form the icloud integration. I use both an haven't noticed much in win11 though.

Just the other day, my M1 Mac at work slowed to such a crawl that mouse movement badly lagged due to runaway 'triald' (ML learning on user inputs) and Siri processes. They relaunched every time I killed them, and they were running even though Siri was explicitly disabled. Turns out there's no way to actually disable Siri without disabling SIP, which I cannot do on my work computer.

So on macOS I'm still plagued with cloud-connected crap I cannot control (or sometimes even expect) that at worst can make my computer completely unusable.

Is it different from Windows bullshit in interesting ways worth talking about? Definitely. Can the two things be compared? Definitely!

This only makes sense to iCloud users. I've never intentionally put anything in iCloud and I get a constant parade of alerts about it being full. On macOS it doesn't matter where you click on the alert widget, it takes you to the page where you can pay for iCloud regardless. It is very spammy.

Whenever I hit the play button it's like 50/50 if Spotify will continue or if apple will open up the music app inviting me to sign up for music.

I did get pestered about joining icloud pretty regularly in the past but haven't seen those in a bit. I prefer onedrive and Google photos since it works on my non apple machines too.

It’s just a question of whether it’s good enough for you or not. And honestly, for laptops, the answer so far is probably still "not". Thunderbolt not working is kind of a deal killer, for example.

https://github.com/AsahiLinux/docs/wiki/Feature-Support

Gotta say the other day I was thinking about the Mac Pro and even at only 192GB of memory, 20P/4E with some Optane SSDs and some flash behind it would absolutely tear through postgres workloads, I bet.

Everyone is dumping on Mac Pro for regressing on memory capacity (after spending the last 2 years dumping on the "$50k mac" and comparing it against Threadripper builds that top out at 256GB), and asking why Mac Pro instead of Studio. Since Asahi doesn't support Thunderbolt yet, you can't use enclosures on Studio. Mac Pro is the only option for a PCIe card that Apple won't sign a driver kext for (NVIDIA cards, for example - and they do have an aarch64 driver thanks to the Tegra family!).

NVIDIA glomming onto Asahi and giving it formal support instead of going their own way with L4T/Jetson Linux would be an interesting course of events. It is, after all, just ARM64/AArch64 under the hood.

WSL has different install requirements for Windows 10 and Windows 11. On Windows 10, it requires a special Windows Optional Component to be enabled, which is a change that requires local admin rights. On Windows 11, WSL can be installed from the Microsoft Store with no need to set up the Windows Optional Component. Shortly after that version of WSL for Windows 11 went live in the Microsoft Store, Microsoft removed the hooks in their installer which enabled that Optional Component in the case it was disabled— on all versions of Windows. So on systems where end users have limited or no admin rights, Windows 10 users cannot install WSL, but Windows 11 users can. On systems which use whitelists for specific installers and processes to be able to elevate privileges, installing WSL on Windows 10 broke because the whitelisted process no longer attempts to set up WSL's dependencies. My organization was bitten by this.

Like with other aspects of WSL, the deeper your engagement is the more bugs and quirks are revealed.

> many need to step outside and realize that Windows has made great strides since Windows 8.

Never ran Windows 8, thankfully. All I know having returned to Windows for a year for the sake of a job is that it is much worse than it was the last time I used it. I'm sorry to hear that Windows 8 so lowered your expectations. It must have been pretty bad.

> Acting like the entire OS is now adware is just complete nonsense.

I'm way more interested in setting realistic expectations about WSL for fellow Linux users than the overall value of Windows, which is probably already a settled question for most of us. But this statement strikes me as pretty funny. 'Complete nonsense' is a pretty strong phrase. If you ever figure out what makes something 'truly adware', send a tip to the lexicographers, because I've just checked 3 dictionaries and Windows qualifies as adware by all 3 definitions they had! :P

More seriously: Windows is definitely different from a lot of adware in that advertising isn't the only revenue strategy for Windows, and Microsoft also sees Windows fulfilling other purposes for them than selling ad space. But generally, 'adware' doesn't mean much more than 'annoying software which shows me ads I don't want to see'. How can you argue that Windows is not that for many, many people?

MPS support is lacking in many ways, i.e. BF16 support is only arriving with Sonoma. One of the biggest things Nvidia has going for it is software maturity, CUDA is far beyond any alternative, particularly MPS. Many PyTorch optimizations only support Nvidia GPUs.

And the tensor core equivalents on Apple SoCs are in the ANE, which necessitates using CoreML. While unified memory bandwidth is impressive on Apple SoCs, it is still slower than recent gen Nvidia cards.

While the PPW certainly is impressive and the overall ML performance is impressive for laptop SoCs, Apple Silicon simply does not hold a candle to discrete GPU performance. You're correct, it can certainly run local inference on the memory-hungry, demanding LLMs that are en vogue currently, but you won't see the performance of a discrete GPU.

The unified memory lends itself to a cheaper, yet slower, alternative to DC-grade Nvidia cards with massive amounts of HBM, but there's a reason you don't see racks of Apple Silicon for LLM training, or even inference at scale.

I'm hopeful the software will vastly improve as Apple Silicon adoption grows, because I'd love to run some of the stuff I build for school on my MacBook, but for now I offload it to my desktop PC because the raw compute brings more value to me than the efficiency when due dates approach.

Perhaps the revenue my code generates isn't "serious", but I'm happy to sit down at any Mac with Visual Studio Code and drag and drop my way to the bank instead of spending hours getting my dotfiles to artisanal perfection.

I'd be pretty shocked if they didn't.