"They've been relegated to pretty insignificant roles in the PC business."

Intel doesn’t think that Arm CPUs will make a dent in the laptop market::“They’ve been relegated to pretty insignificant roles in the PC business.”

  • @bamboo@lemm.eeEnglish
    5•1 year ago
    • Gaming is indeed a big weak point for M series chips. Some games run, but it’s a pretty bad experience overall. Personally I’m insulated from this because I have a gaming PC I can use.
    • I’ve not heard this at all recently. This sounds like it was probably true in late 2020 but afaik creative workflows are quite flushed out at this point.
    • No idea what you’re talking about here. Tools like PyTorch fully support metal for training and inference. llama.cpp fully supports Apple silicon. Apple’s shared memory model gives their GPUs access to huge pools of memory compared to even high end discreet GPUs, and this allows working with models otherwise not possible in a laptop form factor. No other laptop GPU is getting shipped with up to 96GB of memory.
    • Really? This seems like a generally easy thing to fix, what projects (that ran on x86 macOS) are known to be bothersome?
    • This is a barrier to the Asahi Linux folks, and I hope someday the situation improves for them. Otherwise, it’s irrelevant since nearly everything for macOS is gonna be built through Xcode. It’s the system toolchain.

    I don’t know exactly what software you use for work, but for simple cases docker desktop uses binfmt-misc to enable Rosetta and qemu-user for containers. This actually makes it really easy to build and test for a bunch of different architectures, x86 but also ppc64le, mips, etc. With x86-64 specifically you get Rosetta for very high performance. I know tools like gdb don’t work right in this environment, but thats not usually part of a typical ci/cd system anyways.

    • @just_another_person@lemmy.worldEnglish
      -3•1 year ago

      Quick responses, sorry.

      • gaming is a problem for people who DO want this type of machine, and not a second or console. It’s only become an issue with M*
      • Apple has literally thrown an entire engineering group behind their ability to import, export, and transcode media. So much so they even have their own fork of ffmpeg. It’s atrocious.
      • PyTorch is CPU with GPU acceleration where applicable, and most devs want direct GPU access. Not possible on Apple hardware. LLMs are kinda dumb, and most people work on imaging inference. Direct access to the hardware for local development is a must.
      • Not sure what you mean, but Rosetta is essentially QEMU emulation, which is insanely slow. I can run a 2m build for something on a native x64 machine, or a 1 hour build through Rosetta. No thanks.
      • This is a power play by Apple thinking people will still buy their stuff and just deal with the inconveniences, but turns out the sales numbers show that is not the case.

      In general, the “use containers for everything” is not a good workflow. It’s also very subjective to performance on the platform you run it on. Containers all the time is exhausting and problematic for a number of different reasons.