Mesa Considers Raising CPU Support Baseline

[skeevy420]

My impression is that -march=haswell has been seen as the "new" x86_64 target for a while now (AVX2, FMA etc.), while basic x86_64 (16 int + 16 float/simd registers w/SSE2) has been the natural baseline for, quelle surprise, x86_64 since it landed on the market 10+ years ago?

The only "new" thing here is the levels in between I think?

Happy to learn something new as always, so don't be afraid to correct my perspective if you feel I'm barking up the wrong tree. =)

Pretty much, but the levels are what give everyone something to standardize around. Before that all they had was generic or DIY.

I'd like to say that there's some BIOS/UEFI stuff to consider but I think those no-avx Intel's throw that out of whack; whatever the case, GRUB is adequate enough. But, yeah, seems like haswell/zen or sandy/bulldozer are the dividing lines depending on what distribution or person you're talking to.

kpedersen With the BSDs I don't think any of this matters as much since they can just rebuild themselves and they have more integrated bootloaders, but on the Linux side, ideally, a package manager should simply upgrade a person to their respective V-Level with no user interaction whatsoever. In a perfect world Legacy Linux would simply be the basis for Modern Linux where you're silently upgraded to whatever is the best you support whenever package management is done....and that'll be a lot easier if GRUB is forced on everyone since it has the broadest Linux support.

[angrypie]

I just wonder if there is really no way of emulating these newer instructions in software on older processors.

That would only slow things down even more.

Old machines won't be around forever: components die, specially motherboards, and even if your CPU survives it will be hard to find a replacement, and even if you find one, you cannot really know for how long it will work. Even that old Celeron A in a PoS computer will have to be replaced by a Raspberry Pi at some point.

[DanL]

No - stop taking the photo so literally / pedantically

I'll take it however I want it.

Quite clever, really. Was it really not obvious to you, or are you just trolling?

Of course it was obvious to me. Was it not obvious to you that I was being sarcastic?
And clever? LOL. Maybe blurry picture of Linus' middle finger was funny/clever the first 1,000 times I saw it. Now it's just an annoying waste of pixels, especially if it's posted in a vaguely related (at best) topic to what inspired blurry middle finger in the first place.

[bridgman]

One point mentioned in the RFC that I don't see being considered in these discussions is that the proposed change was intended to happen at the same time that older GPU drivers were moved off to a legacy branch, and that the new requirement would only apply to systems with GPUs still supported by current Mesa drivers.

From a quick skim through the thread only the system TemplarGR mentioned would obviously be affected (unless nouveau has "classic" driver code for older parts) and it wasn't clear if TemplarGR's post was serious or in jest.

[commodore256]

I feel Linux should run on hardware older than 20 years because patents last 20 years. You think technology moves fast? Imagine how much faster it would move if patents didn't exist. If I could support a contemporary hardware vendor that only uses patent expired technology, I would, but my performance baseline for general computing is like a workstation from 2013 with support for h265.

So 16 years of patents left to go with the h265 codecs, but by then h268 will be out and would probably revolutionize everything again and increasing the baseline.


Technology only moves fast enough for expired patents to be almost worthless.

[guara]

I don't believe raising the minimal hardware requirement to 2000-2005 era hardware is a problem for consumers, even those who, like me, keep their hardware until it becomes economically unviable to repair. Maybe it will become a problem for a company running their infrastructure on ageing i386 technology, but I understood that Mesa has only raised the requirement to anything that is amd64 compliant.

[CommunityMember]

Perhaps it is just time for Mesa to start by announcing a sunset date for formal x86 32-bit support (they will accept patches, but will not be testing there). That moves any support load to those who want to continue to run "ancient" hardware.

[Vistaus]

Simply stated, there are lots of industrial systems, that are build into large objects like buildings running things like heating/cooling, power distribution, ventilation etc. Those are not easily replaced, were meant to last decades or longer and use some sort of display for interaction. The software on those can be upgraded, but replacing the hardware is usually almost impossible unless you replace the entire building installation.

I'm 200% sure those systems aren't using mainline Mesa anyway and if they do, it's an old version.

[AmericanLocomotive]

I'm 200% sure those systems aren't using mainline Mesa anyway and if they do, it's an old version.

Exactly. There are so many people coming up with contrived scenarios here for justifying supporting ancient hardware, it's mind boggling. As someone who spent a lot of time working in industry, I can promise you that industrial systems are NOT running bleeding edge kernels or mesa. Industrial systems are typically built to a hardware/software spec, and will continue to use that same software stack until the day the machine dies. We had brand new CNC machines in 2019, using Siemens controls, that were still on 2.6 series Linux kernels.

The only time industrial equipment EVER gets updated is either to fix a critical bug that effects the operation of the machine, or to potentially fix a security flaw if it's a network connected piece of equipment. But even then, it will be a custom build still based on the original kernel by the OEM.

Oddly enough I remember when Linux was for enthusiasts with weird, wonderful and "ancient" hardware.

Since when did it change to only become about consumers and gamers who should probably be running Windows anyway?

Sure, they can regress the "baseline" all they want. Other projects that do happen to want to keep the old PPC MacBook or SGI machines will simply pick up the slack and fix the brokenness (probably just a build time flag anyway).

Linux did gain a lot of popularity for running well on *old* hardware. However, CPUs without SSE2 are not *old*, they are positively archaic at this point. The last mainstream (Intel, AMD) CPUs without SSE2 were released ~18-20 years ago.

That's like complaining in 2001 - in the era of Pentium 4s (with SSE2, mind you...) and Athlon XPs that you couldn't compile Linux 2.2.19 on your Commodore 64. That's the kind of time span we're talking about here. It's time to move on and start developing code and applications that actually target modern hardware.

[tildearrow]

No, the impact should be very positive for all but the very oldest systems.

I am done here. I won't report any more typos on Phoronix for 22 days.