AGP Graphics Card Support Proposed For Removal From Linux Radeon/NVIDIA Drivers

[DanglingPointer]

Where the fuck do you think Spain is?

Madre de Dios! i didn't know Spain was the last bastion of AGP!

[Jabberwocky]

TL;DR I don't think many individuals in Africa use Linux outside of (sponsored) educational roles. Regarding hardware: I can confirm ancient hardware (20 years or older) is still very useful and being used.

I know the discussion is about AGP GART, but here's my abstract view for similar topics. I am pro removal of support for older hardware. I have many old systems, but I don't expect anyone to maintain kernel code for those old systems. Userland needs to support older kernels (currently I don't know of any issues, but just saying), backwards compatibility and projects like llvmpipe are important and needs to keep being supported/improved.

I am from Africa, which part do you mean.

Off topic, but just to answer your question... I am from South Africa. I gave my old Pentium 4 laptop to my "gardener" last year. The laptop is 18 years old and has a ATI Mobility Radeon 7500C (AGP based). I was surprised to see similar laptops going for ~$100 USD in local classifieds. I am not sure if anyone pays that much for it, but it's refurbished and listed. Regarding the laptop everything besides the original battery still works. I had Windows XP and Ubuntu dual boot setup. His kid is a computer wizard so I thought he might enjoy that setup. I can confirm that very old hardware is still valuable for majority of people here.

Electricity is a lot more expensive in many African countries. You might lose few bucks when upgrading to RPi 3 but you'll save a lot when paying the next electricity bill. For example, according to some online statistics the average price was $0.49/kWh in Liberia and $0.15/kWh in the US. Years ago, if you had a 300W PC, in Germany the price for operating it 24/7 was around 300 euros per year. In Liberia the sum would probably be $1000. While your average salary is $350. So 3-4 months of work just to pay for the electricity. RPi 3 is about as fast as those AGP era computers and would only cost $40 to operate 24/7.

You're right about the costs, but very few people run their devices 24/7. Most people can't afford a raspberry pi 3 and simply want to check their email when their mobile phone is broken, outdated, or stolen.

Here in South Africa we don't have power all year long, but if there is power it's usually cheap (leaving local price because exchange rate is very high currently). In 2005 we paid R0.31/kWh ($0.0155 USD/kWh) and today we are paying just over R0.98/kWh ($0.049 USD/kWh), data comes from South Africa's Council For Scientific and Industrial Research (CSIR). Some areas in South Africa, like Soweto, has free electricity. Some people are running ant miners and buying new fancy cars and authorities turns a blind eye, that said it is a very small number of people that are abusing illegal power connections to that degree. We are probably the edge case relative to other places in Africa like you have said.

[tsuru]

Whew... this brings back memories of running a dual boot gentoo / ms windows on my old Tyan S2462 with GeForce 3 (nv20)

[caligula]

You're right about the costs, but very few people run their devices 24/7. Most people can't afford a raspberry pi 3 and simply want to check their email when their mobile phone is broken, outdated, or stolen.

Certainly true. But it's easy to scale down the numbers and still, spending 1/12 of your income on computer electricity is just way too much. I used to have couple of huge workstations with 10k-15k rpm SCSI RAIDs and 21" CRTs. Those workstations were loud and amazing sources of heat especially during the summer. These days I might not even need the AC thanks to better power efficiency.

[cb88]

As long as the maintenance costs are virtually zero why not just piss off about it... if someone wants to run on their old hardware let them.

Take NetBSD for instance its supports a vastly wider range of hardware than any other *nix... even has relatively modern driver implementations for 30+ year old hardware. Not because it hard to maintain because it isn't... but because it was designed to just work and not be a maintenance burden. Anyway when Linux finally kicks the bucket on such hardware I suspect alot of people will give NetBSD a spin on there.

Also as already mentioned this is just about disabling GART in the chipset... it probably wouldn't have much effect on acutal performance though it would be interesting to see benchmarks.

[starshipeleven]

As long as the maintenance costs are virtually zero

They aren't. Every time someone is changing common code they need to avoid breaking old stuff.

Take NetBSD ... it was designed to just work and not be a maintenance burden.

[citation needed]

Also as already mentioned this is just about disabling GART in the chipset... it probably wouldn't have much effect on acutal performance though it would be interesting to see benchmarks.

Who cares, this hardware isn't running anything where performance is relevant

[cb88]

They aren't. Every time someone is changing common code they need to avoid breaking old stuff.

[citation needed]

Who cares, this hardware isn't running anything where performance is relevant

Oh poo... citation needed is just a cop out by people turning a blind eye to the facts. Its a well known fact that Linux has severe code churn that causes more regressions than benefits. It's also a well known fact that Linux itself slows down by a measurable percent each release Linus himself has even commented on it.

If breaking common code causes maintenance burden ... THAT IS WHAT CAUSED THE PROBLEM, not the existence of code that if left unchanged would continue to work for decades. *stop breaking common code*

People that bother to run new software on old hardware keep us grounded in our relative perspective of how much bloat we are actually dealing with.... and besides retrochalleges are basically a computer sport at this

Consider for instance a 50Mhz microsparc I have... is perfectly usable as a desktop for a vast array of tasks, and can evey run relatively heavyweight C++ software like KDE 1.1.2... however in the interveing years we have increased the bloat in that software 1000x .... you literally need a computer 1000x faster than that to run anything remotely usable of comparable features today. Sure in many cases the higher performance is an enabler... but it also is an enabler of bloat, which is why I run KDE 1.1.2 on some of my newer computers it does almost everything i need and doesn't waste too many resources to do it.

[starshipeleven]

citation needed is just a cop out

Ok, you don't like it when I try to be nice, I'll just call it blatant bullshit then.

Its a well known fact that Linux has severe code churn that causes more regressions than benefits.

blatant bullshit

It's also a well known fact that Linux itself slows down by a measurable percent each release

blatant bullshit

*stop breaking common code*

The only way to not do that is for it to stop being common code, because someone must work on it to improve the other hardware. Are you aware of how software development works?

Here we are talking of common code inside the radeon driver and inside the noveau driver. Because each driver supports cards on PCI, AGP and PCIe.

Consider for instance a 50Mhz microsparc I have... is perfectly usable as a desktop for a vast array of tasks, and can evey run relatively heavyweight C++ software like KDE 1.1.2... however in the interveing years we have increased the bloat in that software 1000x .... you literally need a computer 1000x faster than that to run anything remotely usable of comparable features today.

Not everyone can be a hardcore C programmer that optimizes all to the extreme, nowadays there are much more programmers than back then, but they aren't all as good.

But you can't buy 1000x slower PCs nowadays anyway, so does that even matter?

It's a tool, it does the same job and costs a fraction of what the 50Mhz microsparc system costed (with inflation adjustment). That's all that matters in the end.

which is why I run KDE 1.1.2 on some of my newer computers it does almost everything i need and doesn't waste too many resources to do it.

Does that even matter? Can you show any metric that makes using ancient unsupported software on modern hardware worthwhile?

Power usage, cost, and performance of payload applications changes?

Because if it doesn't then why you go the extra effort to do it. Just install a distro and get going, do stuff.

[oiaohm]

This is also not as black and white as look for the last AGP cards. This was noted back in 2008 that all AGP cards at that time were in fact AGP to PCIe bridge chip units. Those cards did not use the AGP host memory system well a lot of the time if you read through the quirks that are removed from this patch some of the cards were being pushed back into PCI mode already and no one was noticing the difference..

https://lists.freedesktop.org/archives/dri-devel/2020-May/265540.html

As what is noted here is powerpc systems with AGP has been set never used the cards in AGP mode for years so powerpc used all AGP cards in PCI mode. This was AGP just being a really faster PCI slot this turned out to be the more dependable way.

https://en.wikipedia.org/wiki/Pegasos

Yes really old powerpc systems with AGP graphics slots exist. The PCI mode by AGP slot is not that bad performing. Yes the AGP cards that would not work in a powerpc system is basically the first generation of AGP cards all other AGP cards work. Any AGP card that would worked in the legacy powerpc systems will basically be close to unaffected by this change.

This I think is pure storm in teacup. The AGP feature being removed did not exactly work well that really cannot be documented providing performance advantage that creating a lot of code to maintain and test is being removed.

When AGP cards were way more plentiful the powerpc platforms disabled this mode because maintaining and testing a list of quirks was not worth their time as it was not giving any extra performance. So there over a decade of documented stuff that this feature did not work well.

Developer does need to work on his comment comments to be a little more clear. " We are making AGP on x86 behave exactly like AGP on powerpc because there is no performance advantage maintaining stacks of unique quirks and behaviours for x86 AGP unique stuff" would be another way of writing what this patch-set does.

[CochainComplex]

Just remembering the days when having AGP was desireable.