Announcement

Collapse
No announcement yet.

Clear Linux Offering Performance Advantages Even With Low-Power IoT/Edge Hardware

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • coder
    replied
    Originally posted by Spam View Post
    You should try Gentoo Linux. It's all about optimising for your current hardware, if you want. Default is generic i686 or amd64, but can be set to GCCs 'native' CPU detection.
    Thanks for the advice, but I'm more interested in seeing a paradigm shift than simply tweaking my personal environment. Besides, my existing setup is more than fast enough for most of what I'm doing, these days.

    Leave a comment:


  • Spam
    replied
    Originally posted by coder View Post
    After that shitstorm over Fedora maybe requiring AVX2, I've come around to the position that what we really need is a framework for natively optimizing software for the host. I wish Intel would invest their time and resources into a framework for doing that, rather than playing in their own little Intel-exclusive sandbox.

    However, I see the main benefit of what they're doing as being the patches and optimizations they're publishing to various upstream projects. So, the community certainly still benefits.

    BTW, I never used gentoo, but my sense is their idea isn't completely off. I'd make it a lot more automated and demand-driven, however - compiling & optimizing on-demand. Also, it could involve some amount of profile-driver re-optimization.
    You should try Gentoo Linux. It's all about optimising for your current hardware, if you want. Default is generic i686 or amd64, but can be set to GCCs 'native' CPU detection.

    Leave a comment:


  • coder
    replied
    Originally posted by arjan_intel View Post
    I can only speak for Clear Linux, I don't have enough visibility in "everything" the many software parts of Intel do.
    Clear Linux does not have any special Genuine Intel checks as you imply.
    I know the Linux kernel (upstream) has a few, mostly for cosmetics (printing model numbers etc) and for "on Intel cpus this is how you load microcode" kind of things and maybe some other upstream projects might as well, but we don't add any on top of this in the distro.
    We do use cpuid for things like "is avx2 supported, then use the avx2 version of this", which is vendor independent and is about supported instructions.
    Thank you for the info. Much appreciated.

    Leave a comment:


  • arjan_intel
    replied
    Originally posted by coder View Post
    Thanks.

    Can you make any more general statement about "Genuine Intel" CPU checks, either in Clear Linux or individual software packages maintained by Intel?
    I can only speak for Clear Linux, I don't have enough visibility in "everything" the many software parts of Intel do.
    Clear Linux does not have any special Genuine Intel checks as you imply.
    I know the Linux kernel (upstream) has a few, mostly for cosmetics (printing model numbers etc) and for "on Intel cpus this is how you load microcode" kind of things and maybe some other upstream projects might as well, but we don't add any on top of this in the distro.
    We do use cpuid for things like "is avx2 supported, then use the avx2 version of this", which is vendor independent and is about supported instructions.

    Leave a comment:


  • coder
    replied
    Originally posted by arjan_intel View Post
    fwiw clear linux does not contain or use MKL
    Thanks.

    Can you make any more general statement about "Genuine Intel" CPU checks, either in Clear Linux or individual software packages maintained by Intel?

    Leave a comment:


  • hwertz
    replied
    "After that shitstorm over Fedora maybe requiring AVX2, I've come around to the position that what we really need is a framework for natively optimizing software for the host. I wish Intel would invest their time and resources into a framework for doing that, rather than playing in their own little Intel-exclusive sandbox."

    I think Debian has some in-built hoo-hah (maybe multilib related?) that lets it load different variants of a binary based on CPU capabilities; the original intended use was for ARM. Apparently with the armel port (version for ARMs older than ARMv7), they are now using this so binaries support both "hard float" and "soft float".

    Leave a comment:


  • arjan_intel
    replied
    Originally posted by coder View Post
    Ugh. My first reply is awaiting moderation, but anyway consider:



    The "discriminating" one will only run optimized code on Genuine Intel CPUs. This needs to be patched, for AMD hardware to have a fair shot on Clear Linux. Or to run MKL on it, anywhere.

    Source: https://www.agner.org/optimize/blog/read.php?i=979&v=t
    fwiw clear linux does not contain or use MKL

    Leave a comment:


  • ms178
    replied
    Originally posted by coder View Post
    After that shitstorm over Fedora maybe requiring AVX2, I've come around to the position that what we really need is a framework for natively optimizing software for the host. I wish Intel would invest their time and resources into a framework for doing that, rather than playing in their own little Intel-exclusive sandbox.

    However, I see the main benefit of what they're doing as being the patches and optimizations they're publishing to various upstream projects. So, the community certainly still benefits.

    BTW, I never used gentoo, but my sense is their idea isn't completely off. I'd make it a lot more automated and demand-driven, however - compiling & optimizing on-demand. Also, it could involve some amount of profile-driver re-optimization.
    You might remember my day dreaming some time ago on this topic in another thread to squeeze every bit of performance out of the underlying technology in a way that does not discriminate against less-capable hardware. And Raja just stressed the importance of software optimizations himself to get the hardware to its limits. Fingers crossed that they come up with something more clever than the status quo, just as you said.

    Leave a comment:


  • coder
    replied
    Originally posted by betam4x View Post
    I was not aware there were SFF PCs built on Celeron chips. I assume that from the picture of it they are not the same form factor as raspberry pie. That's what Intel really should focus on. It can be hard to find cases and accessories for proprietary devices. I would love to get my hands on one of those devices. Install a skeleton Windows install and you suddenly have a great little SFF PC.
    For general usage, you really want a SATA SSD. So, I'd steer in the direction of a proper mini-PC, like one of these:

    https://www.asrock.com/nettop/
    https://www.anandtech.com/show/13734...cff-pcs-review

    There are some cases for the ODROID-H2, but they're kind of industrial-looking:

    https://www.hardkernel.com/shop/odroid-h2-case-1/
    https://www.hardkernel.com/shop/odroid-h2-case-3/
    https://www.hardkernel.com/shop/odroid-h2-case-4/

    Leave a comment:


  • coder
    replied
    Originally posted by ms178 View Post
    But I very much welcome their philosophy to bring a performance-optimized distro to the market.
    After that shitstorm over Fedora maybe requiring AVX2, I've come around to the position that what we really need is a framework for natively optimizing software for the host. I wish Intel would invest their time and resources into a framework for doing that, rather than playing in their own little Intel-exclusive sandbox.

    However, I see the main benefit of what they're doing as being the patches and optimizations they're publishing to various upstream projects. So, the community certainly still benefits.

    BTW, I never used gentoo, but my sense is their idea isn't completely off. I'd make it a lot more automated and demand-driven, however - compiling & optimizing on-demand. Also, it could involve some amount of profile-driver re-optimization.

    Leave a comment:

Working...
X