Mozilla Firefox 53.0 Released, Drops Old Linux CPU Support

[chithanh]

And here I am believing that software should be optimized for the hardware it runs on. ie : running gentoo with march=native... Can't change the CPU without preparations.
I haven't done any testing/benchmarking, I admit that, but optimizing for the lowest common denominator always seemed a bit silly to me (why shouldn't software use AVX instructions on my machine which supports it?)

Sometimes this is ok (e.g. if you play back video), but other times optimizing with special instructions can give undesirable results. For example floating point multiply-add operations (a+=b*c) can have slightly different results depending on whether you use multiplication and addition instructions, or a single fused mutiply-add (FMA) instruction.

This is probably also the reason behind hsivonen's comment for no plans to non-IEEE-compliant floating point implementations, as doing strict IEEE math on x87 is cumbersome and slow. And you don't want e.g. your facial recognition software written in Rust deliver different output depending on whether it runs on an SSE2 machine or not.

[rene]

And here I am believing that software should be optimized for the hardware it runs on. ie : running gentoo with march=native... Can't change the CPU without preparations.
I haven't done any testing/benchmarking, I admit that, but optimizing for the lowest common denominator always seemed a bit silly to me (why shouldn't software use AVX instructions on my machine which supports it?)

Thanks for that.
SSE2 is mandatory in AMD64, for low power hardware, there is Dillo

I run T2 (formerly ROCK Linux) always highly optimized long before there was march=native, ... however, this wasted too often a whole afternoon when I needed to boot / access something in emergency, recovery, migration, etc.

I can plug in my USB3 SSD in any of my systems from old Mac to latest Dell XPS13 and it boots without any issue. Of course I will next to never notice a performance difference, if I build with core2 or core-i7-avx2: https://www.youtube.com/watch?v=D3H8GrVm_XI

[caligula]

Slowest 2 core 10 W Goldmont Celeron J3355 shows 880 singlerank

<snip>

That would be very worthy upgrade over any Athlon XP or Pentium 3 on any point - performance, power consumpation and of course support Even integrated GPU in that might beat 90% of AGP cards Not to mention you can watch videos with that like a man, etc...

If you only want to watch videos, the mobile platforms would do a much better job. Some chinese cherry trail NUC platform or development board perhaps? Even the first gen Rpi model A runs FullHD H264 High Profile 5.1 videos just fine. Most people have realized this. The old computers are becoming obsolete. Even the empty PC case can easily weigh 20+ pounds without any electrical components, 30+ with everything installed inside. A NUC is 1-2 pounds. The choice is quite obvious for a regular traveller or if you need to move every now and then.

[s_j_newbury]

How much is that benchmark measuring memory bandwidth? Assuming both CPUs are used with Gentoo with march=native data-align=abi so the code is generated is properly scheduled and cache aligned I doubt there would be much difference for non memory bound benchmarks.

So the big difference would be power consumption, but most of the time, when not compiling (!) the CPU is halted anyway. If your desktop

[s_j_newbury]

If you only want to watch videos, the mobile platforms would do a much better job. Some chinese cherry trail NUC platform or development board perhaps? Even the first gen Rpi model A runs FullHD H264 High Profile 5.1 videos just fine. Most people have realized this. The old computers are becoming obsolete. Even the empty PC case can easily weigh 20+ pounds without any electrical components, 30+ with everything installed inside. A NUC is 1-2 pounds. The choice is quite obvious for a regular traveller or if you need to move every now and then.

Undeniably true of course, if you're buying a new machine. But if your current system meets your needs why would or should you?

[caligula]

Undeniably true of course, if you're buying a new machine. But if your current system meets your needs why would or should you?

For instance the power consumption can be a real problem in some countries. My old PC used to draw around 100W when idle and I was running it 24/7. In some countries that would cost a fortune, maybe a month's net salary or even more. A tablet or say Orange Pi with proper sleep/suspend modes would save 99% of that money. Also easier to carry around.

[starshipeleven]

Undeniably true of course, if you're buying a new machine. But if your current system meets your needs why would or should you?

Just so you (= any kind of people attached to obsolete hardware, not necessarily you) can stop whining on forums about programs that drop support for your obsolete crap that was still "meeting your needs", for starters.

Really, anything dropped here is obsolete, you can keep trying to keep it alive by using Gentoo and babysitting it like it was your first born or just let it go, fully knowing that it has long since outlasted its planned service life (was not supposed to last more than 5 years back then due to fast performance increases), and use all the time you spent babysitting it or waiting for the damn thing to load stuff to do something more useful.

And note that I'm not a fan of upgrade for the sake of it. I still have quite a bit of mini-boards with Geodes (from before they became AMD) running, but they are low-power stuff, in use as headless systems doing simple house automation jobs, running LEDE firmware.

I also still have a Phenom first gen from 2008 or something as a secondary PC that is still 100% fine, but Athlon XP 3200+.... wtf man.

[s_j_newbury]

I'm no fan of planned obsolescence, but the Athlon XP was only by way of example. I could have presented the argument the other way around, that writing non-portable code means the future developnent stays locked into the current architectural paradigm. Even depending on ISA extensions means it becomes impossible to drop that extension in the future for something else so the "standard" becomes ever more complex.

Personally, as far as CPUs go, I'm happy with my i7-3840QM in my laptop, and my undervolted Phenom II 965 in my media server is still as useful today as it was when I assembled it. I guess they're well on the way..

[s_j_newbury]

For instance the power consumption can be a real problem in some countries. My old PC used to draw around 100W when idle and I was running it 24/7. In some countries that would cost a fortune, maybe a month's net salary or even more. A tablet or say Orange Pi with proper sleep/suspend modes would save 99% of that money. Also easier to carry around.

A system that draws 100W idle is misconfigured or really poorly designed! CPU halt has been around a long time!

[starshipeleven]

I'm no fan of planned obsolescence, but the Athlon XP was only by way of example.

It's not planned obsolescence, it's normal obsolescence. Stuff gets obsolete as the world around it changes, it's not bad, it's normal. You know, all the hippie stuff about nature cycle and all that. It applies also to technology.

Planned obsolescence is designing stuff to break at specific times well before they should really need to be replaced or with idiotic limitations that will act like kill switches (proprietary undocumented connectors, must connect to an online server to work and so on). For example Pogoplugs and other derivative Kirkwood-based embedded devices. Many can be found on ebay for peanuts because they are completely useless now that the online service they were supposed to work with has been shut down.

Useless if you don't have a UART-USB dongle and can't install LEDE/Openwrt/Debian on them, that is, but most people can't.

I could have presented the argument the other way around, that writing non-portable code means the future developnent stays locked into the current architectural paradigm. Even depending on ISA extensions means it becomes impossible to drop that extension in the future for something else so the "standard" becomes ever more complex.

It works like this everywhere (GPUs, PLC in automation systems, microcontrollers inside 99% of the electronic devices) and always did.
Manufacturers have to expose new hardware functionality without breaking programs that work already, how can they do that without using this "extensions that eventually become standard" system?

Personally, as far as CPUs go, I'm happy with my i7-3840QM in my laptop, and my undervolted Phenom II 965 in my media server is still as useful today as it was when I assembled it. I guess they're well on the way..

Since the new devices aren't particularly more powerful than that, these aren't yet obsolete (and won't be for a long time if it keeps going like this). If Intel pulls some bunnies out of its hat and next 2-3 gens of CPUs gain like 50% or more performance per generation like back then in the Athlon XP times, they will become obsolete and the world will start to leave them behind.