Arch Linux Developers Discuss Idea Of Providing An x86-64-v3 Port

[hsci]

I would prefer the v2 approach for simplicity and broader support. Especially a lot of reliable well working hardware is in use with a v2 feature set (my laptops for example). The Gitlab show the important things:

/*
Providing a second architecture would increase our repo size by approximately 66% (~32GB).
Building two architectures will take additional packager time unless automated.
Some developers may not have hardware to debug issues found purely in x86-64-v3 packages. It is likely these issues are very rare.
*/

They are right, it must be automatic. Regarding the size I would even prefer seeing still one package, including multiple executable or libraries.

/*
It would be preferable if pacman on x86_64_v3 could still install packages from x86_64, particularly for non-Arch repositories that
may not want to build for both architectures. This would also allow a transition into x86_64_v3 when firstly [core] gets rebuilt, followed
by other repos one at a time. Your friendly pacman developers may be willing to add the ability to specify multiple architectures in
pacman.conf.
*/

And compatible.

The drawback what we are seeing is the maintenance burden. I'm not sure if the distribution should optimize here so much. Maybe actually the software in the packages should recognize the available CPU features at runtime and dlopen?

Compatibility is probably the most important feature. And then there is already ABS and Gentoo.

[AmericanLocomotive]

So distribution maintainers get decide that they have the right to stop me from using their distribution just because i have a two year old Apollo Lake and Gemini Lake laptop that does not come with AVX?

Yes, that is correct. It's their OS. They can arbitrarily decide whatever they want. Nothing is stopping you from just taking the source and recompiling without the AVX flags

Users with new hardware have every right to assume that their hardware will work on the latest x64-v3 optimized distribution, regardless of whether it's an Atom, Pentium or Celeron. Especially if the hardware is less than a year old.

When I buy a brand new GeForce 1010, I don't cry and complain I can't run Half-Life Alyx in VR on it. It is what it is. It's low end hardware. Doesn't matter that it's brand new. Users buying low-end CPUs should be doing so fully aware that they are making sacrifices. If I can't run a certain Linux distribution, I'll either run an older version or find one that is compatible. I didn't complain I couldn't run Windows XP on a 486, even though brand-new 486-equipped laptops were being sold just about into the Pentium-2 era.

You also conveniently forget the the only fanless laptops available today are powered by Atoms, Celerons and Pentiums. So a user who has a preference for fanless laptops gets unilaterally kicked out by various distribution minatainers just because they want to pursue instruction elitism by focusing on x64-v3?

Instruction elitism? Give me a break. It's about actually driving performance and efficiency forward, since the optimized baseline x86-64 code leaves anywhere from 5-20% performance on the table. That's an entire processor generation.

Also, there are a HUGE amount of fanless laptops sold with Intel Core-series CPUs. https://www.ultrabookreview.com/6520...ss-ultrabooks/ . In fact, the vast majority of ultra-portables and ultrabooks all have Intel Core-series CPUs. Really the only ones running Atoms, Pentiums or Celerons are the low-end budget full-size machines.

I standby the notion that if you purchase low-end hardware, you need to be prepared to deal with low-end hardware issues.

[pininety]

Even v2 is long in the tooth. I know, I did a v2 to v3 change recently. Intel Westmere, basically the v2 cutoff line, to AMD Zen 2, well into v3. I was one of those "My 10 year old system is good enough, I'll tell ya what" people; then I upgraded and 4K switched from seconds per frame to 30+ frames per second, same HDD, same OS install, same GPU. Was being the key word. I'm still gaming on 1080p. 1080p60 maxed out looks and plays better than 4K with compromised settings to reach between 30 and 60.

IMHO, v2 is only for people who are happy or don't need any better than 1080p 30-60 fps. No matter what GPU you pair it with, that's about as good as you'll get because you're extremely CPU limited. Can't even play games like Death Stranding due to the dreaded AVX requirement. It freaking sucked owning that game and not being able to play it because my Westmere wasn't up to par.

Well, not everyone is gaming with their machines. I have an 9 year old T530 with an i7-3820QM. That does virtually everything I want and I barely notice in every day use a difference with my Ryzen 3600 desktop maschine which is on paper 50% faster. That CPU does only have AVX1 however and hence would not be usable with v3.

[hsci]

I have here a X220 - it does actually everything I want.

I'm questioning if numerous specific CPU features with flags and required adaption of machine code is what we want? Actually the execution should become faster and be reliable. Hopefully RISC-V will follow here a better approach.

[skeevy420]

Well, not everyone is gaming with their machines. I have an 9 year old T530 with an i7-3820QM. That does virtually everything I want and I barely notice in every day use a difference with my Ryzen 3600 desktop maschine which is on paper 50% faster. That CPU does only have AVX1 however and hence would not be usable with v3.

I agree. It's just that in my experiences v1/v2, regardless of gaming or not, are really only good for up to 1080p content and desktops aside from image manipulation or freesync/gsync 2K gaming for v2. A person can use those architectures for over 1080p, but they're not really suited for it and the CPU lag becomes noticeable.

Here's a Core2Quad with an RTX 2080. Long story short, GPU doesn't even run at full power with high settings because v1 C2Q can't keep up....on 1080p. Even on 1080p with one of the best GPUs around a v1 will struggle.

While I'm using gaming in my examples, because that and compiling things are the most intensive things my system does, gaming can translate over to how well productivity and office apps will run and respond and how well multi-monitor setups will work. If one full screen program with one monitor gives a system a hiccup then it can be assumed that GIMP on one monitor with LIbreOffice and Firefox with 27 tabs on the other monitor probably will too. Need to draw a rectangle? Let me freeze up that YouTube tutorial you're following while I render that.

I have the Pro APU version of your CPU -- a 4650g. Had to buy mine off the grey market. What's funny, and sad, is one month ago I paid around 50-100 less for a 4650G on the grey market than I'd pay for a 3600X on the legit market as of yesterday's prices

[skeevy420]

Yes, that is correct. It's their OS. They can arbitrarily decide whatever they want. Nothing is stopping you from just taking the source and recompiling without the AVX flags

When I buy a brand new GeForce 1010, I don't cry and complain I can't run Half-Life Alyx in VR on it. It is what it is. It's low end hardware. Doesn't matter that it's brand new. Users buying low-end CPUs should be doing so fully aware that they are making sacrifices. If I can't run a certain Linux distribution, I'll either run an older version or find one that is compatible. I didn't complain I couldn't run Windows XP on a 486, even though brand-new 486-equipped laptops were being sold just about into the Pentium-2 era.

Instruction elitism? Give me a break. It's about actually driving performance and efficiency forward, since the optimized baseline x86-64 code leaves anywhere from 5-20% performance on the table. That's an entire processor generation.

Also, there are a HUGE amount of fanless laptops sold with Intel Core-series CPUs. https://www.ultrabookreview.com/6520...ss-ultrabooks/ . In fact, the vast majority of ultra-portables and ultrabooks all have Intel Core-series CPUs. Really the only ones running Atoms, Pentiums or Celerons are the low-end budget full-size machines.

I standby the notion that if you purchase low-end hardware, you need to be prepared to deal with low-end hardware issues.

Amen to that.

[werfu]

All this debate for an additionnal repos. For those of us who lived through the first Linux years, having different architectures releases was quite normal. I remember seeing builds of the same distro for 386, 586, 686, IA64, PPC and Sparc. All of these had to be built manually. But let's face it, most of the package building is now automated. What's the repo size increase, a couple bucks more per month of hosting? IMHO this is a non issue. Just tune the installer to select the right architectures for your system and you're done. Leave an option to manually select the architecture if you'd like. Also, why not support a kind of downward-compatibility? Unlike i686 and AMD64, a package compiled for AMD64-v1 could use a shared lib for AMD64-v3, as the ABI compatibility should stay. Just have repos that superseeds others.

As for booting, that's an all other subject. Some distro had installer that suported both LILO and Grub. I don't see why having support for both Grub and systemd-boot would be such a problem. It's only a matter of running the right hook upon new kernel installation. Yes more maintaining. But if the users requires it, why not?

[Guest]

While Alan didn't go into it, v3, from what I can tell, is also UEFI-only. That, to me, is a bigger game changer than new CPU instructions. Being free of GRUB opens up a lot of possibilities and will allow development to move a lot faster. Think of how many things in the file system and encryption world alone are hampered by legacy systems needing GRUB. Being able to standardize around what is essentially "dump stuff into a ramdisk on an fat32 EFI" opens up a lot of things.

UEFI doesn't exclude GRUB. Besides, Arch doesn't force a bootloader. The live ISO uses systemd-boot IIRC.

[zwastik]

10 year old PCs are Nehalem based, and are honestly getting pretty pokey these days, especially the mobile parts. My i7-2630qm laptop (which was a solid mid-pack quad-core mobile Sandy Bridge part in 2011) isn't the fastest on a lot of modern websites or even HD video playback, and it has an SSD. Even 2009 era pre-Nehalem Core CPUs are barely tolerable for general web browsing, with the lowest end Celeron chromebook chips offering similar or better performance.

It's like complaining about software no longer supporting your 800 MHz Pentium 3 in 2010 when people are running 4 and 6 Core Core "i" series chips, or whining that you can't install Windows XP on your 486 in 2001 when everyone else is running 1GHz Athlons....

Technology marches on. It's about time our software starts being targeted and optimized for CPUs that are 7 years old, instead of 17.

I have a mobile Intel core i5-2410M, 8GB ram and a 512 GB SSD. I can browse the web, watch Netflix, youtube or FullHD video files just fine. Notice this is only a dualcore cpu. I do run sway in wayland so I dont have those big (and bloated?) DE, maybe that could be the difference.

10%-20% are NOT "modest" gains. They are the equivalent of a generation upgrade or two.

For some very specific programs, mostly some scientific software. Most people won't see the difference.

[AmericanLocomotive]

I have a mobile Intel core i5-2410M, 8GB ram and a 512 GB SSD. I can browse the web, watch Netflix, youtube or FullHD video files just fine. Notice this is only a dualcore cpu. I do run sway in wayland so I dont have those big (and bloated?) DE, maybe that could be the difference.

Video files are fine played with VLC are fine, 1080p30 YouTube is fine, 1080p60 youtube is iffy, and 4K just doesn't work at all - at least using Firefox.

For some very specific programs, mostly some scientific software. Most people won't see the difference.

No, that's 10-20% average gain across all workloads.