Re: x86-64-v2: oggenc, zstd (Stephan Kulow)
---------- Forwarded message ----------
From: Stephan Kulow <coolo@suse.de> To: factory@lists.opensuse.org Cc: Bcc: Date: Sat, 8 Oct 2022 08:13:11 +0200 Subject: Re: x86-64-v2: oggenc, zstd
Yes, systems mainly doing heavy work will be upgraded when faster hardware and budget is available. But the former high performance systems remain suitable for doing ordinary work for long time. Yes. But the question remains on how many of those you plan to run tumbleweed on.
Greetings, Stephan
Stephan:
As posted previously, I'm running TW and a number of Gecko Rolling, and Debian Sid, and Manjaro, along with Lubuntu developmental . . . on a '12 cMP using a Xeon i7 processor of that era . . . . The plan is to run it until a number of parts blow up. I'm also running Leap 15.5 on an '09 MBP using Core2Duo processor, perfectly competent for daily driver. F
My 2 cents - sorry to flog a dead horse We support obsolete CPU's (IBM mainframes, Power PC, ARM (raspberry PI)) - do we even have enough users of these platforms to justify that development? Why would we stop supporting working x86-64 machines that have enough cpu and disk to be useful. I think that any machine that supports SATA 3 drives should still be supported until SATA drives are no longer made. I still think that the answer is a repo for v1 and one for v4 builds - give those that want the new hardware a proper speedup and keep those that are fine as it is a choice. Leave the non-v2 to v4 machines a life. I hate to have to migrate all ny friends with old machines to versions that want to keep old hardware working like sparky linux or mint linux. Quit being like Microsoft and want people to buy new hardware for features 99% of users will never use. I keep having to tell Windows users that I cannot recover their pictures from dead hardware without the bitlocker password for the drive I removed. Few if any Windows 10 or 11 ever make the USB device with the bitlocker recovery key. I too, have virtual machines with most flavors of linux - Fedora, Sparky, Ubuntu, Manjaro, Mint, Kali, Pi OS as well as Windows 95, XP,, 10, 11 currently running and a few images that have been retired (NT 3.1, 4, 2000, 2003 server, 2006 server). On 10/8/22 09:20, Fritz Hudnut wrote:
---------- Forwarded message ----------
From: Stephan Kulow <coolo@suse.de> To: factory@lists.opensuse.org Cc: Bcc: Date: Sat, 8 Oct 2022 08:13:11 +0200 Subject: Re: x86-64-v2: oggenc, zstd
> Yes, systems mainly doing heavy work will be upgraded when faster > hardware and budget is available. But the former high performance > systems remain suitable for doing ordinary work for long time. Yes. But the question remains on how many of those you plan to run tumbleweed on.
Greetings, Stephan
Stephan:
As posted previously, I'm running TW and a number of Gecko Rolling, and Debian Sid, and Manjaro, along with Lubuntu developmental . . . on a '12 cMP using a Xeon i7 processor of that era . . . . The plan is to run it until a number of parts blow up. I'm also running Leap 15.5 on an '09 MBP using Core2Duo processor, perfectly competent for daily driver.
F
On Samstag, 8. Oktober 2022 16:54:13 CEST Larry Len Rainey wrote:
My 2 cents - sorry to flog a dead horse
We support obsolete CPU's (IBM mainframes, Power PC, ARM (raspberry PI)) - do we even have enough users of these platforms to justify that development?
ARMv6/v7 IMHO is not in a very good state, many packages are not even compiling any more. On the other hand, it does not matter much, there is no official support, and build failures does not affect other architectures. You might not aware of it, but IBM mainframes and PPC64 are current hardware. And IBM is a partner of SUSE, and there is likely a lot of money involved to keep these platforms supported.
Why would we stop supporting working x86-64 machines that have enough cpu and disk to be useful. I think that any machine that supports SATA 3 drives should still be supported until SATA drives are no longer made.
I still think that the answer is a repo for v1 and one for v4 builds - give those that want the new hardware a proper speedup and keep those that are fine as it is a choice.
Putting v2 and v3 in the same bin as x86_64 baseline (there is no such thing like v1 or v0) and only optimizing for v4 is kind of stupid. Most of the baseline hardware has been decommisioned already (look at what you get today when looking for refurbished hardware - 2nd to 7th generation Intel i3/5/7), and is more and more likely to just fail. v2 and v3 are common today, while v4 is probably almost as rare as baseline. All processors from AMD and Intel released in the last 11/10 years are level v2 or better. Most of the older processors support only DDR2 (a few support DDR3) and PCIe 2.0. Leap 16 is not released today, but likely sometime in 2023. Most hardware in 2023 will be v2 or better, and during the next ten years this will only go up.
Leave the non-v2 to v4 machines a life. I hate to have to migrate all ny friends with old machines to versions that want to keep old hardware working like sparky linux or mint linux.
These will be still supported with Leap 15.5. Nobody forces you to migrate to Leap 16. Refurbished hardware which supports v2 (including SSD, 8 GB of RAM, 4th Gen Intel i5 Quadcore, i.e. v3) is available today for well below 200€.
Quit being like Microsoft and want people to buy new hardware for features 99% of users will never use.
99% of users will use SSE3/SSE4*, as the software and hardware supports it. 99% of users have bought sufficient hardware during the last 10 years. Regards, Stefan
Stefan Brüns composed on 2022-10-09 00:42 (UTC+0200):
99% of users have bought sufficient hardware during the last 10 years.
Where did that number come from? Not all users buy. Many can only acquire whatever is available for no money. -- Evolution as taught in public schools is, like religion, based on faith, not based on science. Team OS/2 ** Reg. Linux User #211409 ** a11y rocks! Felix Miata
On Sonntag, 9. Oktober 2022 01:32:20 CEST Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 00:42 (UTC+0200):
99% of users have bought sufficient hardware during the last 10 years.
Where did that number come from?
From common sense. Actually, during the last 10 years you could not buy anything new which is not x86_64-v2. x86_64 has been available since late 2003, and x86_64-v2 was introduced early 2011 (Intel Sandybridge)/late 2011 (AMD Bulldozer). So the first 7 years of x86_64 are baseline, everything after (11 years) is x86_64-v2. So even assuming not a single unit has been decommisioned, and shipments per year are the same (actually, it went up from 180M units in 2004, 310M in 2013 to 345M in 2021) 60% of all x86_64 PCs ever built are v2 or better. Taking growing markets and decommissioning into account, 99% is a quite plausible number. (BTW: You question my 99%, but leave Larrys 99% claim without questioning it. Go figure ...)
Not all users buy. Many can only acquire whatever is available for no money.
Enterprises buy. Private persons buy. And this hardware ends up with persons who can not afford to buy sooner or later. Regards, Stefan
Stefan Brüns composed on 2022-10-09 02:42 (UTC+0200):
Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 00:42 (UTC+0200):
99% of users have bought sufficient hardware during the last 10 years.
Where did that number come from?
From common sense. Actually, during the last 10 years you could not buy anything new which is not x86_64-v2.
Why 10? In Linux user circles recommendations are universally to not buy "new", but to buy something in the at least 6-12 month old range. So really, 10 isn't 10, but less. 10 is arbitrary. Lots of people never buy major expense items new. I've never bought a "new" car, only "pre-owned". I've acquired far more PCs for free than using money, and only a small portion using money were "new" (as in less than a year since availability of their CPUs was introduced). Several have been older than a year when "new". Among my siblings' households, all 4 have at least 2 PCs running Linux, while 3 have almost no PCs less than 12 years old. The newest 7 are Core2Duo Dells, except for one gamer who built a new AMD less than 5 years ago.
x86_64 has been available since late 2003, and x86_64-v2 was introduced early 2011 (Intel Sandybridge)/late 2011 (AMD Bulldozer).
So the first 7 years of x86_64 are baseline, everything after (11 years) is x86_64-v2. So even assuming not a single unit has been decommisioned, and shipments per year are the same (actually, it went up from 180M units in 2004, 310M in 2013 to 345M in 2021) 60% of all x86_64 PCs ever built are v2 or better.
So, 40% are unaccounted for, either recommissioned, cannibalized (partial recommissioning), or landfill or recycler fodder.
Taking growing markets and decommissioning into account, 99% is a quite plausible number.
Reason cannot just ignore those not "buying". Acquisitions also come from those decommissioned by original users, not due to failure, but due to policy, warranty expiration, or less utopian reasons, of otherwise perfectly useful equipment that may have been little more than dust collectors most of their calendar lives, and having much otherwise useful potential.
(BTW: You question my 99%, but leave Larrys 99% claim without questioning it. Go figure ...)
That sentence had the M word in it, so no good reason to eval.
Not all users buy. Many can only acquire whatever is available for no money.
Enterprises buy. Private persons buy. And this hardware ends up with persons who can not afford to buy sooner or later.
Not a reason to arbitrarily increase the pool of the valueless among the indigent and landfills. It's bad for the planet to dig up all that ore and burn fuel to manufacture unneeded replacements that may be in /actual/ use no more than marginally better than working equipment they replace. -- Evolution as taught in public schools is, like religion, based on faith, not based on science. Team OS/2 ** Reg. Linux User #211409 ** a11y rocks! Felix Miata
On Sun, Oct 9, 2022 at 4:57 AM Felix Miata <mrmazda@earthlink.net> wrote:
Stefan Brüns composed on 2022-10-09 02:42 (UTC+0200):
Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 00:42 (UTC+0200):
99% of users have bought sufficient hardware during the last 10 years.
Where did that number come from?
From common sense. Actually, during the last 10 years you could not buy anything new which is not x86_64-v2.
Why 10? In Linux user circles recommendations are universally to not buy "new", but to buy something in the at least 6-12 month old range. So really, 10 isn't 10, but less.
10 is arbitrary. Lots of people never buy major expense items new. I've never bought a "new" car, only "pre-owned". I've acquired far more PCs for free than using money, and only a small portion using money were "new" (as in less than a year since availability of their CPUs was introduced). Several have been older than a year when "new". Among my siblings' households, all 4 have at least 2 PCs running Linux, while 3 have almost no PCs less than 12 years old. The newest 7 are Core2Duo Dells, except for one gamer who built a new AMD less than 5 years ago.
x86_64 has been available since late 2003, and x86_64-v2 was introduced early 2011 (Intel Sandybridge)/late 2011 (AMD Bulldozer).
So the first 7 years of x86_64 are baseline, everything after (11 years) is x86_64-v2. So even assuming not a single unit has been decommisioned, and shipments per year are the same (actually, it went up from 180M units in 2004, 310M in 2013 to 345M in 2021) 60% of all x86_64 PCs ever built are v2 or better.
So, 40% are unaccounted for, either recommissioned, cannibalized (partial recommissioning), or landfill or recycler fodder.
Taking growing markets and decommissioning into account, 99% is a quite plausible number.
Reason cannot just ignore those not "buying". Acquisitions also come from those decommissioned by original users, not due to failure, but due to policy, warranty expiration, or less utopian reasons, of otherwise perfectly useful equipment that may have been little more than dust collectors most of their calendar lives, and having much otherwise useful potential.
(BTW: You question my 99%, but leave Larrys 99% claim without questioning it. Go figure ...)
That sentence had the M word in it, so no good reason to eval.
Not all users buy. Many can only acquire whatever is available for no money.
Enterprises buy. Private persons buy. And this hardware ends up with persons who can not afford to buy sooner or later.
Not a reason to arbitrarily increase the pool of the valueless among the indigent and landfills. It's bad for the planet to dig up all that ore and burn fuel to manufacture unneeded replacements that may be in /actual/ use no more than marginally better than working equipment they replace.
If we're going to bring up the environment, the computers made more than ten years ago are vastly more energy inefficient and pull more energy than most computers you can buy in the last five years. Recycling those computers and harvesting their base materials to produce better computers would be good for the environment because it reduces the carbon footprint of that person using a computer. The reason why x86_64-v3 is unreasonable is that there were Intel CPUs made as recently as 2020 that can't satisfy that spec, nevermind x86_64-v4. Raising the baseline to x86_64-v2 also has the effect of raising the generational baseline to something that is generally useful for contemporary Linux system software. -- 真実はいつも一つ！/ Always, there's only one truth!
On Sun, 9 Oct 2022 08:32:04 +0200 Neal Gompa wrote:
If we're going to bring up the environment, the computers made more than ten years ago are vastly more energy inefficient and pull more energy than most computers you can buy in the last five years. Recycling those computers and harvesting their base materials to produce better computers would be good for the environment because it reduces the carbon footprint of that person using a computer. Maybe the systems which are obsoleted do not run 24/7 but only a few hours a week. Then the energy saved by a new system will never compensate the energy consumed for manufacturing the new system.
Raising the baseline to x86_64-v2 also has the effect of raising the generational baseline to something that is generally useful for contemporary Linux system software.
Actually it means obsoleting systems for the sake of ... obsoleting systems (because some day in the future they could be insufficient to run a then current Linux). In german c't magazine 2011 issue 12 there was a suggestion for a high performance PC with a Phenom II X6. Probably this system still outperforms everything built ever since with something called Celeron inside. And it will be obsoleted by requiring x86_64-v2. According to the data there is almost no performance gain by using x86_64-v2. I can not think of a good analogy, but it is similar to saying: we only support USB 3 superspeed keyboards from now on. Nobody will have any benefit because nobody can type so fast. But it sounds cool. Kind regards, Dieter
On Sun, Oct 09, 2022 at 10:27:53AM +0200, dieter wrote:
On Sun, 9 Oct 2022 08:32:04 +0200 Neal Gompa wrote:
If we're going to bring up the environment, the computers made more than ten years ago are vastly more energy inefficient and pull more energy than most computers you can buy in the last five years. Recycling those computers and harvesting their base materials to produce better computers would be good for the environment because it reduces the carbon footprint of that person using a computer. Maybe the systems which are obsoleted do not run 24/7 but only a few hours a week. Then the energy saved by a new system will never compensate the energy consumed for manufacturing the new system.
Raising the baseline to x86_64-v2 also has the effect of raising the generational baseline to something that is generally useful for contemporary Linux system software.
Actually it means obsoleting systems for the sake of ... obsoleting systems (because some day in the future they could be insufficient to run a then current Linux).
In german c't magazine 2011 issue 12 there was a suggestion for a high performance PC with a Phenom II X6. Probably this system still outperforms everything built ever since with something called Celeron inside. And it will be obsoleted by requiring x86_64-v2.
For some tasks, sure. It will be limited to amounts of memory and bus speeds that are not exactly great today. Even if you can mount a SSD in it which probably was not a reasonable option then it will not support NVMe. It likely supports USB3 with some questionable extra controller if at all, it does not look like chipset support existed. Support for routing PCIe externally with something like USB C or Thunderbolt probably existed but not all boards support it even today. And it does not support CXL which very few if any new systems support today. Some of these problems will make the system eventually impractical to use, probably in the near future. But it does support ECC which is not common in new consumer hardware so it is not exactly easy to replace, either. Some server Celerons do, though. Thanks Michal
On Sun, 9 Oct 2022 11:09:14 +0200 Michal Suchánek wrote:
In german c't magazine 2011 issue 12 there was a suggestion for a high performance PC with a Phenom II X6. Probably this system still outperforms everything built ever since with something called Celeron inside. And it will be obsoleted by requiring x86_64-v2.
For some tasks, sure. I would say enough tasks.
It will be limited to amounts of memory and bus speeds that are not exactly great today. Even if you can mount a SSD in it which probably was not a reasonable option then it will not support NVMe. It likely supports USB3 with some questionable extra controller if at all, it does not look like chipset support existed. Support for routing PCIe externally with something like USB C or Thunderbolt probably existed but not all boards support it even today. And it does not support CXL which very few if any new systems support today. yes it is correct, the system possibly does not support every use case or connectivity which exists today. But for many other it is perfectly usable. And I think the users should have the freedom to decide by themselves whether the hardware is still up to their needs or not.
It would be a completely different story if some hardware is not supported anymore. Something like "hardly somebody is using PCs with BIOS any more and nobody is willing to maintain BIOS support, therefore we have to stop offering it and have to require UEFI". But in this x86-64-v2 case it is really only about activating or changing compiler flags.
Some of these problems will make the system eventually impractical to use, probably in the near future. But it does support ECC which is not common in new consumer hardware so it is not exactly easy to replace, either. Some server Celerons do, though. yes I agree, someday the system will be insufficient or vital components will fail and it makes no sense to replace them. It is also not my intention to keep the support for pre-x86_64-v2 CPUs forever. I just want to say that now and possibly for some years more it is too early, especially when the gain is so low.
Kind regards, Dieter
On Sonntag, 9. Oktober 2022 10:27:53 CEST dieter wrote:
On Sun, 9 Oct 2022 08:32:04 +0200
Neal Gompa wrote:
If we're going to bring up the environment, the computers made more than ten years ago are vastly more energy inefficient and pull more energy than most computers you can buy in the last five years. Recycling those computers and harvesting their base materials to produce better computers would be good for the environment because it reduces the carbon footprint of that person using a computer.
Maybe the systems which are obsoleted do not run 24/7 but only a few hours a week. Then the energy saved by a new system will never compensate the energy consumed for manufacturing the new system.
There are plenty of decommisioned newer systems.
Raising the baseline to x86_64-v2 also has the effect of raising the generational baseline to something that is generally useful for contemporary Linux system software.
Actually it means obsoleting systems for the sake of ... obsoleting systems (because some day in the future they could be insufficient to run a then current Linux).
In german c't magazine 2011 issue 12 there was a suggestion for a high performance PC with a Phenom II X6. Probably this system still outperforms everything built ever since with something called Celeron inside. And it will be obsoleted by requiring x86_64-v2.
Yeah, actually a great example: The article had two systems, one AMD (Phenom II) and one Intel (i5-2500). The Intel one already is x86_64-v2. The AMD sytem has an idle power consumption of 53W, while the Intel system uses only (cough) 27W. In multicore benchmarks, the AMD is slightly faster, the Intel wins in singlecore benchmarks by a significant margin. An Intel i3-4350 is twice as fast in single/dual-core load, and about the same speed in 6+ core loads. A Pentium G5500 is (significantly) faster in all benchmarks. The "high end" system is beaten by the low end from 4 years ago. Current entry level systems use 8-20 Watt in idle (some even less). Thats a difference of at least 30 Watts, significantly more under load. The old system will spend more time in high load than a current one. Regards, Stefan
On 09.10.22 08:32, Neal Gompa wrote:
If we're going to bring up the environment, the computers made more than ten years ago are vastly more energy inefficient and pull more energy than most computers you can buy in the last five years.
I'd question that, at least with the given timeframes. My (not hypothetical) "server" machine at home, running 24/7 is a Fujitsu Siemens Esprimo E730 machine. Small form factor professional desktop machine. Intel Core2 Duo E8500. 8GB RAM. 2 disks with rotating rust, 2TB and 4TB, spun down approx 95% of the time. 2 SSDs, 512GB and 2TB. On-board gigabit ethernet, pcie gigabit ethernet card, pci DVB-S2 card. Running vdr, openvpn, ssh, owncloud, some KVM vm occationally, NFS server, dhcp+dns+tftp server (dnsmasq). When idling (which is is probably about 90% of the time), it draws about 30-35 Watts. Given the hardware in there (2x GBit ethernet, DVB-S2 card), there is not a huge part of the energy consumption that can be accounted to CPU / chipset. The improvements with energy consumptions in the last 10 years have mostly been in the areas of "connected standby" (windows only AFAIK), faster suspend and resume etc, which all is a non-issue here. I have been involved with handling of and caring/developing for notebook computers running linux for 20 years now, and the great improvements to battery life have been the intel Centrino Platform (pentium M, the return to the old P3 architecture) and then again the intel core/core2 platforms (where, AFAICT also the "we switch peripherals into low power modes" stuff started to be actually of benefit), but the last 10 years have mostly improved the battery technology and not really reduced power usage in a significant way. It is a different issue if the CPU is actually doing stuff and running at full speed with all cores loaded fully, but this is in my experience a rather rare event (and on machines that are expected to run 100% CPU all the time usually even the basic power saving settings are disabled for alleged performance reasons).
Raising the baseline to x86_64-v2 also has the effect of raising the generational baseline to something that is generally useful for contemporary Linux system software.
I would go with that argument if it was true. Running a linux XFCE desktop on the 8GB Thinkpad x200s (core-2 duo ultra low power 1,8GHz?) is getting a tad laggy, but it's mostly the tab-switching in the browsers or stuff like that which I'd actually guess is due to the graphics stuff, but it is well usable still. "Linux system software" (which, in my book, is kernel, system services, most simple server daemons) runs perfectly well on such a system. If we were to "remove support for hardware that prevents systems being useful", then we would remove support for rotating-rust-drives. Adding a SSD to an even more than 10 year old system makes it from "unbearably bad" to "neatly usable again" in no time with only a small investment. Even my trusty Toughbook CF-51 (intel centrino, Banias 1,6GHz, 2GB RAM) became usable again after adding an ide-to-m2 adapter and a m2.sata drive, and it has a REAL SERIAL PORT!!! ;-) Have fun anyway. seife -- Stefan Seyfried "For a successful technology, reality must take precedence over public relations, for nature cannot be fooled." -- Richard Feynman
On Sonntag, 9. Oktober 2022 10:39:20 CEST Stefan Seyfried wrote:
On 09.10.22 08:32, Neal Gompa wrote:
If we're going to bring up the environment, the computers made more than ten years ago are vastly more energy inefficient and pull more energy than most computers you can buy in the last five years.
I'd question that, at least with the given timeframes.
My (not hypothetical) "server" machine at home, running 24/7 is a Fujitsu Siemens Esprimo E730 machine. Small form factor professional desktop machine. Intel Core2 Duo E8500. 8GB RAM. 2 disks with rotating rust, 2TB and 4TB, spun down approx 95% of the time. 2 SSDs, 512GB and 2TB. On-board gigabit ethernet, pcie gigabit ethernet card, pci DVB-S2 card.
Running vdr, openvpn, ssh, owncloud, some KVM vm occationally, NFS server, dhcp+dns+tftp server (dnsmasq).
When idling (which is is probably about 90% of the time), it draws about 30-35 Watts. Given the hardware in there (2x GBit ethernet, DVB-S2 card), there is not a huge part of the energy consumption that can be accounted to CPU / chipset.
When I measured my Haswell E3-1245 workstation some years ago, it measured at about 15W in idle. This is with a Supermicro workstation board (2x GBit, several PCIe + PCI) and at that time included 2x spinning rust, a SSD and a Soundblaster Audigy 2 (PCI). So probably about 15-20W to be saved. c't had a test of refurbished office PCs in issue 18/2022, 6th or 8th generation Intel i5, idle power between 8.3W and 11.9W, 111€-329€. The DVB-S2 PCI card can be replaced with a USB stick or PCIe card, and after 3 years you reach break even.
The improvements with energy consumptions in the last 10 years have mostly been in the areas of "connected standby" (windows only AFAIK), faster suspend and resume etc, which all is a non-issue here.
There have been improvements throughout the bench. For desktop PCs, one of the most important improvements have been power supplies and efficiency under low load.
I have been involved with handling of and caring/developing for notebook computers running linux for 20 years now, and the great improvements to battery life have been the intel Centrino Platform (pentium M, the return to the old P3 architecture) and then again the intel core/core2 platforms (where, AFAICT also the "we switch peripherals into low power modes" stuff started to be actually of benefit), but the last 10 years have mostly improved the battery technology and not really reduced power usage in a significant way.
20 years ago, systems had 512 MByte of RAM, Core2Duos typically 2 or sometimes 4 GByte. Now typical systems have 8 or 16 GByte of significantly faster RAM, and idle power nevertheless went down slightly. Energy ratings of batteries stayed the same (40-70Wh), only energy density went up. Many efficiency improvement were set off by larger (high res) screens, larger memory, higher peak performance. Current generation dedicated GPUs need 5W idle while driving a 4K display, a few years ago 15-20W were normal. A NUC with a Celeron N5105 needs 3W in idle, 30W max and has a higher performance than the "highend" Phenom II X6 mentioned elsewhere (53W idle). Yes, hardly any improvements the last few years.
It is a different issue if the CPU is actually doing stuff and running at full speed with all cores loaded fully, but this is in my experience a rather rare event (and on machines that are expected to run 100% CPU all the time usually even the basic power saving settings are disabled for alleged performance reasons).
Raising the baseline to x86_64-v2 also has the effect of raising the generational baseline to something that is generally useful for contemporary Linux system software.
I would go with that argument if it was true. Running a linux XFCE desktop on the 8GB Thinkpad x200s (core-2 duo ultra low power 1,8GHz?) is getting a tad laggy, but it's mostly the tab-switching in the browsers or stuff like that which I'd actually guess is due to the graphics stuff, but it is well usable still.
So, "getting laggy". For a system, which likely has been upgraded as far as possible (more RAM, apparently SSD). With some effort and some compromises you were able to squeeze out a little bit of extra lifetime. And still supported with Leap 15.x. Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory. Notable exception, but far from typical.
"Linux system software" (which, in my book, is kernel, system services, most simple server daemons) runs perfectly well on such a system.
If we were to "remove support for hardware that prevents systems being useful", then we would remove support for rotating-rust-drives. Adding a SSD to an even more than 10 year old system makes it from "unbearably bad" to "neatly usable again" in no time with only a small investment. Even my trusty Toughbook CF-51 (intel centrino, Banias 1,6GHz, 2GB RAM) became usable again after adding an ide-to-m2 adapter and a m2.sata drive, and it has a REAL SERIAL PORT!!! ;-)
A RPi3 or RPi4 probably runs cirles around it (or many other AArch64 SBCs), and also come with a real serial port, even several. Regards, Stefan
On 09. 10. 22, 20:14, Stefan Brüns wrote:
So, "getting laggy". For a system, which likely has been upgraded as far as possible (more RAM, apparently SSD). With some effort and some compromises you were able to squeeze out a little bit of extra lifetime. And still supported with Leap 15.x.
Leap 15.x is unusable due to deprecated (by upstream) python stack. Every other python upstream dropped support for 3.6. And that's only one example. Will be python updated to something maintained in 15.5? Barely. 15.x is simply too old to run on machines with current SW needs today, please stop suggesting that. People are using what they are using for a reason. TW still runs just fine on both old (as old as my i586 desktop (with an SSD) -- used very seldomly) and new machines. Again, there is little need to upgrade CPUs today. If you have enough RAM and an SSD, you are fine. (If you don't do computations, graphics, or stuff like that, which most people don't.)
Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory. Notable exception, but far from typical.
What? Both my core2duos have 8g/12g RAM and an ssd. People do upgrades of components. However, there is a little point in upgrading the motherboard/CPU. It's sometimes even impossible. For example due to connected old (and expensive) peripherals like through legacy PCI. The systems run just fine 24/7. Why would you people still want to abandon them for no good reason? That, I still don't understand. regards, -- js suse labs
On 09.10.22 20:14, Stefan Brüns wrote:
On Sonntag, 9. Oktober 2022 10:39:20 CEST Stefan Seyfried wrote:
When idling (which is is probably about 90% of the time), it draws about 30-35 Watts. Given the hardware in there (2x GBit ethernet, DVB-S2 card), there is not a huge part of the energy consumption that can be accounted to CPU / chipset.
When I measured my Haswell E3-1245 workstation some years ago, it measured at about 15W in idle. This is with a Supermicro workstation board (2x GBit, several PCIe + PCI) and at that time included 2x spinning rust, a SSD and a Soundblaster Audigy 2 (PCI). So probably about 15-20W to be saved.
OK, i did not really revisit this the last years. 15W total with 2 disks (usually 2 to 5W each, unless spun down, 2x gigabit links (1W each) and even more peripherals is really a nice number. I'm going to get an Esprimo P710 Core i3 3220 (AFAICT the last model year that still contains a PCI bus for my DVB-S2 card ;-) and do some measurements at that one. Should be of about the same age and perform similarly well (power consumption wise).
c't had a test of refurbished office PCs in issue 18/2022, 6th or 8th generation Intel i5, idle power between 8.3W and 11.9W, 111€-329€. The DVB-S2 PCI card can be replaced with a USB stick or PCIe card,
No it can't. The USB boxes are *all* crap and usually randomly fail to respond after a few weeks of uptime (I have quite a selection of these resting in some buckets...) The PCIe cards are either just usb chips with an PCIe usb controller (the cheap ones), or there are only externally maintained drivers (of questionable quality) available. Due to the current dire situation with upstream kernel DVB driver maintenance, anything not already in tree for quite some time will most likely not get into it because the current maintainer is apparently a bit "complicated". So it will be a 3rd gen i3 for now. I'll check the idle power consumption and then decide if there is significantly more that can be saved by dropping the DVB-S2 card any maybe replacing it with another network SAT>IP tuner (if I can still find one on ebay).
The improvements with energy consumptions in the last 10 years have mostly been in the areas of "connected standby" (windows only AFAIK), faster suspend and resume etc, which all is a non-issue here.
There have been improvements throughout the bench. For desktop PCs, one of the most important improvements have been power supplies and efficiency under low load.
OK, I was mostly looking at Laptop machines. I'll now check a (slightly ;) newer desktop
I have been involved with handling of and caring/developing for notebook computers running linux for 20 years now, and the great improvements to battery life have been the intel Centrino Platform (pentium M, the return to the old P3 architecture) and then again the intel core/core2 platforms (where, AFAICT also the "we switch peripherals into low power modes" stuff started to be actually of benefit), but the last 10 years have mostly improved the battery technology and not really reduced power usage in a significant way.
20 years ago, systems had 512 MByte of RAM, Core2Duos typically 2 or sometimes 4 GByte. Now typical systems have 8 or 16 GByte of significantly faster RAM, and idle power nevertheless went down slightly. Energy ratings of batteries stayed the same (40-70Wh), only energy density went up.
Many efficiency improvement were set off by larger (high res) screens, larger memory, higher peak performance.
That's possible. It's just that 10 year old Laptop machines are, with very small investments (ssd, memory) able to run current software totally fine for many applications. And even the battery life of a 12 year old X200s (which got a new battery a few years ago, of course) is still very usable (3+ hours, I have not run down that battery for quite some time so I don't know the exact time. Battery life of that core2 duo ultra-low-voltage machine is "orders of magnitude" better than my Toughbook CF51, 1.6GHz Banias machine which, with a new battery, can reach 2+ hours if I'm careful).
Current generation dedicated GPUs need 5W idle while driving a 4K display, a few years ago 15-20W were normal.
No idea, the last dedicated GPU I used was the ATI graphics in an hp nc6000 notebook almost 15 years ago (and when I, just for fun, tried to install tumbleweed on that machine a year ago I found out that this driver is really no longer working, even though it is still included, while the intel driver for the Toughbook's i815 graphics still chugs along nicely ;)
A NUC with a Celeron N5105 needs 3W in idle, 30W max and has a higher performance than the "highend" Phenom II X6 mentioned elsewhere (53W idle).
aren't the NUCs basically "notebook systems without a display" (components wise)? But yes, that's impressive.
Yes, hardly any improvements the last few years.
It will be interesting to see the difference between the old Core2 duo machine (with all the non-essential hardware removed) and the new core i3 machine (before adding the additional disks and peripherals). I'll report back how much of a difference that makes. I'll let me suprise how much the bare metal consumes and how much my additional peripherals draw.
I would go with that argument if it was true. Running a linux XFCE desktop on the 8GB Thinkpad x200s (core-2 duo ultra low power 1,8GHz?) is getting a tad laggy, but it's mostly the tab-switching in the browsers or stuff like that which I'd actually guess is due to the graphics stuff, but it is well usable still.
So, "getting laggy". For a system, which likely has been upgraded as far as possible (more RAM, apparently SSD). With some effort and some compromises you were able to squeeze out a little bit of extra lifetime. And still supported with Leap 15.x.
And windows 10. The "getting laggy" is that I start to notice the screen redraw of the browser when switching tabs, so it is maybe taking 30+ms when I do not notice this on my newer T430 machine. So it's not 'oh, I have to wait for a second after switching tabs'. (That's more the experience on the Toughbook, but right now nobody is really arguing for no longer supporting that machine ;-) And yes, almost all my old machines (and of everyone who asks me for Hardware advice) have been upgraded wrt. memory and switched from rotating rust to SSD, because that's really the cheapest way to make an old computer perfectly usable again for very little money. A 10 year old machine with 4+GB RAM and an SSD is still perfectly usable for most people nowadays, and the necessary investment is usually less than 50€.
Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory. Notable exception, but far from typical.
Most core2duos that are still in use today will have an SSD and 4+GB of memory. At least that is my experience and the advice I give to everyone who asks me wrt. computer hardware. "Should I buy the new shiny laptop from Mediamarkt? It has 2TB storage!!!!" "no, let's add some memory and a ssd to your old machine and it will do better for a tenth of the price.
If we were to "remove support for hardware that prevents systems being useful", then we would remove support for rotating-rust-drives. Adding a SSD to an even more than 10 year old system makes it from "unbearably bad" to "neatly usable again" in no time with only a small investment. Even my trusty Toughbook CF-51 (intel centrino, Banias 1,6GHz, 2GB RAM) became usable again after adding an ide-to-m2 adapter and a m2.sata drive, and it has a REAL SERIAL PORT!!! ;-)
A RPi3 or RPi4 probably runs cirles around it (or many other AArch64 SBCs), and also come with a real serial port, even several.
Of course. But the toughbook has the added benefit of being tough. And the raspberry pi's serial ports just die (togehter with the rest of the CPU) if I connect them to the wrong pins of the Motorcycle ECU I'm trying to program with it. And there's no Display and battery to see what I'm doing. And it can't run the software needed. So the toughbook is the right tool for *that* job ;-) Even if it's old and slow. And it doesn't complain if it get's dropped from the seat, again. -- Stefan Seyfried "For a successful technology, reality must take precedence over public relations, for nature cannot be fooled." -- Richard Feynman
Just a short final update on this off-topic subject.... (executive summary: you're right, newer machine consumes half the power) On 10.10.22 19:21, Stefan Seyfried wrote:
On 09.10.22 20:14, Stefan Brüns wrote:
On Sonntag, 9. Oktober 2022 10:39:20 CEST Stefan Seyfried wrote:
When idling (which is is probably about 90% of the time), it draws about 30-35 Watts. Given the hardware in there (2x GBit ethernet, DVB-S2 card), there is not a huge part of the energy consumption that can be accounted to CPU / chipset.
When I measured my Haswell E3-1245 workstation some years ago, it measured at about 15W in idle. This is with a Supermicro workstation board (2x GBit, several PCIe + PCI) and at that time included 2x spinning rust, a SSD and a Soundblaster Audigy 2 (PCI). So probably about 15-20W to be saved.
OK, i did not really revisit this the last years. 15W total with 2 disks (usually 2 to 5W each, unless spun down, 2x gigabit links (1W each) and even more peripherals is really a nice number. I'm going to get an Esprimo P710 Core i3 3220 (AFAICT the last model year that still contains a PCI bus for my DVB-S2 card ;-) and do some measurements at that one. Should be of about the same age and perform similarly well (power consumption wise).
It went on to be an Esprimo P757, Core i3-6100@3.7GHz, the PCI socket is provided by a PCIE-to-PCI adapter card with an ASMedia 1085 bridge chip. The old machine in fact did draw about 40 watts (I measured pretty much exactly 9.6kWh / 10days). The new machine draws about 20-21 watts when idling, and even with all CPU's loaded I can't get it much above 40 watts. The only difference in the used hardware is, that the new machine had an m2.sata ssd with 256GB included and I'm using that one as a systems volume instead of the 500GB Samsung that was in the other one (and which was not even half used), maybe this saves some milliwatts? And I left off the second PCIE-1GBit NIC, as it was unused at the time anyway (thanks to a VLAN enabled switch I acquired some time ago). I doubt that the idling, non-connected NIC makes for a big difference in power but it is a difference.
It will be interesting to see the difference between the old Core2 duo machine (with all the non-essential hardware removed) and the new core i3 machine (before adding the additional disks and peripherals). I'll report back how much of a difference that makes.
The bare core i3 machine (no additional disks, ssds, SAT-TV card, ...) did draw about 10W with windows 10, a little bit more with tumbleweed, but I did not do any optimisation, the SATA ALPM would probably have saved that additional watt or so). I'll try to get the old machine running without much things connected, but I'd guess it will be about the same 10 watts difference that the additional hardware, so it will be around 25-30 watts.
I'll let me suprise how much the bare metal consumes and how much my additional peripherals draw.
Peripherals -> about 10 watts. So for my case of a 24/7 running machine it is certainly worth going from a 12 year old machine to a 5 year old, the cost of the machine will be paid via the energy bill in about 2 years. But that's of course no reason for dropping support for the old machine, because it will still be good for non-24/7-running jobs ;-) Have fun, -- Stefan Seyfried "For a successful technology, reality must take precedence over public relations, for nature cannot be fooled." -- Richard Feynman
Stefan Brüns composed on 2022-10-09 20:14 (UTC+0200):
Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory.
People with more than 8G RAM are jaded. 4-8G is all an average mortal user needs, depending on whether he even knows that multitasking means. Those who have never been exposed to the benefits of SSDs are in a state of ignorant bliss, and haven't had to RMA 20% of their SSD acquisitions within two years of placed in service. I have 14 Core2Duos, 4 with SSDs, variously with 2G (4), 4G (8) or 8G (2) RAM. All are adequate for their designated tasks. Given their tasks all require little time powered, there's zero possibility of recovering cost of replacement of any of them. I also have around 6 variously named 65nm 2 thread "Pentiums" also still suited to task. That's 20 64bit Intel with 2 or more threads and 65nm or less lithography, plus 2 AMDs of similar capability, age and lith all being prematurely disabled from openSUSE for political and/or marketing reasons, leaving me with 5 operational @v2 or more, of which one is not yet 6 years old and yet only v2, plus a late K8 v2 with 16G RAM, SSD, 1 year old PSU, but motherboard that will no longer POST. What it all means is more uptime in other distros, less uptime in openSUSE, less experience to share in openSUSE forums, fewer used computers with non-zero marketplace value (aka pricier used computers, and used computers trickier to shop for), and fewer new openSUSE users via recommendations or personal installations, all for *non-quantified* overall performance benefit and me-too what RHEL did. -- Evolution as taught in public schools is, like religion, based on faith, not based on science. Team OS/2 ** Reg. Linux User #211409 ** a11y rocks! Felix Miata
On 12/1/22 18:35, Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 20:14 (UTC+0200):
Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory.
People with more than 8G RAM are jaded. 4-8G is all an average mortal user needs, depending on whether he even knows that multitasking means. Those who have never been exposed to the benefits of SSDs are in a state of ignorant bliss, and haven't had to RMA 20% of their SSD acquisitions within two years of placed in service.
I use a light weight window manager and its pretty common for my browser on its own to be above 8GB, if you end up with an electron app or two which are pretty common these days then its pretty easy to be hitting 16 already. I don't think I could have a functioning desktop with under 32GB to not be occasionally hitting issues. Having said that my machines are currently comfortably sitting at 11Gb and 13Gb running mostly general desktop workloads atm. -- Simon Lees (Simotek) http://simotek.net Emergency Update Team keybase.io/simotek SUSE Linux Adelaide Australia, UTC+10:30 GPG Fingerprint: 5B87 DB9D 88DC F606 E489 CEC5 0922 C246 02F0 014B
Simon Lees composed on 2022-12-01 21:45 (UTC+1030):
Felix Miata wrote:
People with more than 8G RAM are jaded. 4-8G is all an average mortal user needs, ... I use a light weight window manager and its pretty common for my browser on its own to be above 8GB, if you end up with an electron app or two which are pretty common these days then its pretty easy to be hitting 16 already. I don't think I could have a functioning desktop with under 32GB to not be occasionally hitting issues. Having said that my machines are currently comfortably sitting at 11Gb and 13Gb running mostly general desktop workloads atm. ... Emergency Update Team keybase.io/simotek SUSE Linux
I find it hard to imagine any member of the SUSE Linux development team as a member of average mortal users. I normally have hundreds of browser tabs open among 4 or more different browsers and yet it's uncommon for total RAM consumption by apps to consume more than 1/3 of my 32GB of RAM. Most RAM consumption here the vast majority of the time is in cache. -- Evolution as taught in public schools is, like religion, based on faith, not based on science. Team OS/2 ** Reg. Linux User #211409 ** a11y rocks! Felix Miata
On 12/2/22 04:46, Felix Miata wrote:
Simon Lees composed on 2022-12-01 21:45 (UTC+1030):
Felix Miata wrote:
People with more than 8G RAM are jaded. 4-8G is all an average mortal user needs, ... I use a light weight window manager and its pretty common for my browser on its own to be above 8GB, if you end up with an electron app or two which are pretty common these days then its pretty easy to be hitting 16 already. I don't think I could have a functioning desktop with under 32GB to not be occasionally hitting issues. Having said that my machines are currently comfortably sitting at 11Gb and 13Gb running mostly general desktop workloads atm. ... Emergency Update Team keybase.io/simotek SUSE Linux
I find it hard to imagine any member of the SUSE Linux development team as a member of average mortal users.
Nah, I have had laptops that are really just running a browser, mail client, irc client, Discord and a Text editor and I'd farm package building off to another machine.
I normally have hundreds of browser tabs open among 4 or more different browsers and yet it's uncommon for total RAM consumption by apps to consume more than 1/3 of my 32GB of RAM. Most RAM consumption here the vast majority of the time is in cache.
I have tended to find its certain websites that either use javascript badly or do things like embedding media poorly as a background that cause 90% of the issues. I could launch a huge number of simple websites with 16GB and be fine, but there was certain websites I needed occasionally and it would only take 1-2 of them open to grind the whole machine to a halt on a system with 32GB of ram I don't tend to have the same issue (but I don't tend to have 10 of those pages open at once). -- Simon Lees (Simotek) http://simotek.net Emergency Update Team keybase.io/simotek SUSE Linux Adelaide Australia, UTC+10:30 GPG Fingerprint: 5B87 DB9D 88DC F606 E489 CEC5 0922 C246 02F0 014B
On Thu, Dec 01, 2022 at 09:45:43PM +1030, Simon Lees wrote:
I use a light weight window manager and its pretty common for my browser on its own to be above 8GB, if you end up with an electron app or two which are pretty common these days then its pretty easy to be hitting 16 already. I don't think I could have a functioning desktop with under 32GB to not be occasionally hitting issues. Having said that my machines are currently comfortably sitting at 11Gb and 13Gb running mostly general desktop workloads atm.
Having hit the OOM reaper this week, I wonder if your browser observation is perhaps an example of a generalisation of Parkinson's law? The browsers consume until stopped. In my case competition against a freshly launched 4GiB VM in a system with 16GiB of physical memory triggered the OOM reaper. Not sure why OOM killer targetted plasmashell with a score of 0 this week, instead of one/many of the 34 chrome(ium) instances with a score of 300, or another 8 chromium instances with a score of 200, eg: Nov 30 14:57:18 chunk kernel: [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name Nov 30 14:57:18 chunk kernel: [ 17992] 507 17992 4366216 2998934 26849280 35029 0 plasmashell Nov 30 14:57:18 chunk kernel: [ 2515] 507 2515 296308145 15562 1798144 10297 300 chrome Nov 30 14:57:18 chunk kernel: [ 18647] 507 18647 8566333 6256 1388544 3418 200 chrome The journal reminds me it did the same in August too. My bad. Daniel
On Fri, Dec 2, 2022 at 4:07 PM Daniel Morris <danielm@ecoscentric.com> wrote:
Not sure why OOM killer targetted plasmashell with a score of 0 this week, instead of one/many of the 34 chrome(ium) instances with a score of 300, or another 8 chromium instances with a score of 200, eg:
Nov 30 14:57:18 chunk kernel: [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name Nov 30 14:57:18 chunk kernel: [ 17992] 507 17992 4366216 2998934 26849280 35029 0 plasmashell Nov 30 14:57:18 chunk kernel: [ 2515] 507 2515 296308145 15562 1798144 10297 300 chrome Nov 30 14:57:18 chunk kernel: [ 18647] 507 18647 8566333 6256 1388544 3418 200 chrome
Well, plasmashell consumes much more real memory so from this listing it is logical. oom_score_adj is not "OOM score", it is a hint.
On Fri, Dec 02, 2022 at 04:15:15PM +0300, Andrei Borzenkov wrote:
On Fri, Dec 2, 2022 at 4:07 PM Daniel Morris <danielm@ecoscentric.com> wrote:
Not sure why OOM killer targetted plasmashell with a score of 0 this week, instead of one/many of the 34 chrome(ium) instances with a score of 300, or another 8 chromium instances with a score of 200, eg:
Nov 30 14:57:18 chunk kernel: [ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name Nov 30 14:57:18 chunk kernel: [ 17992] 507 17992 4366216 2998934 26849280 35029 0 plasmashell Nov 30 14:57:18 chunk kernel: [ 2515] 507 2515 296308145 15562 1798144 10297 300 chrome Nov 30 14:57:18 chunk kernel: [ 18647] 507 18647 8566333 6256 1388544 3418 200 chrome
Well, plasmashell consumes much more real memory so from this listing it is logical. oom_score_adj is not "OOM score", it is a hint.
Aha. Thanks for the explanation. I'd over-guestimated a sum of the real memory for all the chromes, which turns out to be less than a quarter of plasmashell. But in any case I suppose the oom reaper deals with individual processes and not the sum of many suspect ones named the same. Daniel
On 12/2/22 23:36, Daniel Morris wrote:
On Thu, Dec 01, 2022 at 09:45:43PM +1030, Simon Lees wrote:
I use a light weight window manager and its pretty common for my browser on its own to be above 8GB, if you end up with an electron app or two which are pretty common these days then its pretty easy to be hitting 16 already. I don't think I could have a functioning desktop with under 32GB to not be occasionally hitting issues. Having said that my machines are currently comfortably sitting at 11Gb and 13Gb running mostly general desktop workloads atm.
Having hit the OOM reaper this week, I wonder if your browser observation is perhaps an example of a generalisation of Parkinson's law? The browsers consume until stopped.
In my case competition against a freshly launched 4GiB VM in a system with 16GiB of physical memory triggered the OOM reaper.
Not sure why OOM killer targetted plasmashell with a score of 0 this week, instead of one/many of the 34 chrome(ium) instances with a score of 300, or another 8 chromium instances with a score of 200, eg:
On my last machine before I upgraded from 16 to 32 GB of ram I was launching Firefox in a separate cgroup to limit it to 6GB of ram which generally made sure it was dieing first. -- Simon Lees (Simotek) http://simotek.net Emergency Update Team keybase.io/simotek SUSE Linux Adelaide Australia, UTC+10:30 GPG Fingerprint: 5B87 DB9D 88DC F606 E489 CEC5 0922 C246 02F0 014B
Simon Lees wrote:
On 12/1/22 18:35, Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 20:14 (UTC+0200): Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory. People with more than 8G RAM are jaded. 4-8G is all an average mortal user needs, depending on whether he even knows that multitasking means. Those who have never been exposed to the benefits of SSDs are in a state of ignorant bliss, and haven't had to RMA 20% of their SSD acquisitions within two years of placed in service. I use a light weight window manager and its pretty common for my browser on its own to be above 8GB, if you end up with an electron app or two which are pretty common these days then its pretty easy to be hitting 16 already. I don't think I could have a functioning desktop with under 32GB to not be occasionally hitting issues. Having said that my machines are currently comfortably sitting at 11Gb and 13Gb running mostly general desktop workloads atm.
That's not saying much. Webbrowsers use as much ram as available to them, so if you have more ram they'll use more of it. Regards, Lukas Straub
Am Donnerstag, 1. Dezember 2022, 09:05:37 CET schrieb Felix Miata:
Stefan Brüns composed on 2022-10-09 20:14 (UTC+0200):
Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory.
People with more than 8G RAM are jaded.
I'm planning my next desktop system, and I'm considering 64GB, possibly more. As a sidenote that side of that system is actually more like a (large) hypervisor. Planning to be able to run complete openshift clusters in VMs on it. Cheers MH -- Mathias Homann Mathias.Homann@openSUSE.org OBS: lemmy04 Jabber (XMPP): lemmy@tuxonline.tech Matrix: @mathias:eregion.de IRC: [Lemmy] on liberachat and ircnet (bouncer active) keybase: https://keybase.io/lemmy gpg key fingerprint: 8029 2240 F4DD 7776 E7D2 C042 6B8E 029E 13F2 C102
On Thu, Dec 1 2022 at 03:05:37 -0500, Felix Miata <mrmazda@earthlink.net> wrote:
Stefan Brüns composed on 2022-10-09 20:14 (UTC+0200):
Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory.
People with more than 8G RAM are jaded. 4-8G is all an average mortal user needs, depending on whether he even knows that multitasking means. Those who have never been exposed to the benefits of SSDs are in a state of ignorant bliss, and haven't had to RMA 20% of their SSD acquisitions within two years of placed in service.
I will generally agree, 8GB is still enough for most, sadly I will have to perform the forums migration in a week, which would have been impossible to do with just 8GB (I have tried before). It really depends on what you need to do on your machine. LCP [Jake] https://lcp.world/
On 2022-12-01 09:05, Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 20:14 (UTC+0200):
Most Core2Duos will not have an SSD, nor 8 GByte or even 4 GByte of memory.
People with more than 8G RAM are jaded. 4-8G is all an average mortal user needs, depending on whether he even knows that multitasking means. Those who have never been exposed to the benefits of SSDs are in a state of ignorant bliss, and haven't had to RMA 20% of their SSD acquisitions within two years of placed in service.
I have two core2duos machines, both with SSD. One has 4 gigs (a laptop), the other 8 (a desktop). I also have a Celeron laptop on SSD and 4 gigs. All are in use. -- Cheers / Saludos, Carlos E. R. (from 15.4 x86_64 at Telcontar)
Neal Gompa composed on 2022-10-09 02:32 (UTC-0400):
If we're going to bring up the environment, the computers made more than ten years ago are vastly more energy inefficient and pull more energy than most computers you can buy in the last five years. Recycling those computers and harvesting their base materials to
You really think recycling of materials happens to more than a small fraction of production? I haven't seen any numbers to actually support that theory. It's capacity to produce profit has proven poor so far.
produce better computers would be good for the environment because it reduces the carbon footprint of that person using a computer.
I've seen that argument and am not convinced "carbon footprint" is more than politics. Without carbon their couldn't be life. It's everywhere. We can't materially change the amount of it. All we can do is try to minimize converting it from neutral or positive to detrimental states, such as filling landfills, and without producing collateral damage as this junking of the Core2Duo-class population would constitute. Neither am I convinced of the "vastness" of the as-installed and used difference. I go by my consumption meter, listed by newest to oldest, 0 with discrete graphics cards (except for iMac with integrated 24" 1920x1200 & Radeon HD 2600): Consumption rate/instant|CPU model |Intro | Cores/Threads | Watts | RAM | storage 0.7041KWh/24h, 31W idle: v4 i5-11400 Q1'21 6 cores12 thrds 65WTDP 16GDDR4 120G NVME 0.7281KWh/24h, 29W idle; v3 i3-7100T Q1'17 2 cores 4 thrds 35WTDP 16GDDR4 120G NVME 2X1T HDD 0.4905KWh/24h, 21W idle; v2 G4600 Q1'17 2 cores 4 thrds 51WTDP 16GDDR4 120G NVME 1.0415KWh/24h, 37W idle; v3 i3-4150T Q2'14 2 cores 4 thrds 35WTDP 32GDDR3 120G SSD 2X1T HDD 1.1163KWh/24h, 46W idle; v2A10-7850K Q1'14 4 cores 4 thrds 95WTDP 8GDDR3 250G SSD 0.5581KWh/24h, 25W idle; v2 G3220 Q3'13 2 cores 2 thrds 53WTDP 16GDDR3 250G SSD 0.7082KWh/24h, 33W idle; v1 E7600 Q2'09 2 cores 2 thrds 65WTDP 4GDDR2 250G SSD 1.1345KWh/24h, 48W idle; v1 E8400 Q1'08 2 cores 2 thrds 45WTDP 8GDDR3 750G HDD 1.0490KWh/24h, 46W idle; v1 E8400 Q1'08 2 cores 2 thrds 45WTDP 4GDDR3 750G HDD 0.9752KWh/24h,114W idle; v1 T7700 Q2'07 2 cores 2 thrds 35WTDP 4GDDR2 1T HDD * iMac 1.1314KWh/24h, 48W idle; v1 E4400 Q2'07 2 cores 2 thrds 65WTDP 2GDDR2 160G HDD 1.3409KWh/24h, 57W idle; v1 E6700 Q3'06 2 cores 2 thrds 65WTDP 4GDDR2 1TB HDD I have a twin to the E8400/4GDDR3 above, but with Radeon HD 6450 discrete GPU from Q1'11 that idles @38W, less than with the IGP in use. :p Minimum uptime used to calculate KWh/24h was 120 minutes; my meter only displays 2 decimal places. What I see above looks like much less than a halving of power demands over a 16 year period, a savings somewhere between nil (i5-11400 .7041 vs. E7600 .7082) and 47% (2 core/35W i3-7100T .7281 vs worst Core2Duo E6700 1.3709 for 46.89% savings). Newest vs worst, .7041 vs. 1.3409 for equal TDP produced 48.64% savings. Best of newest vs best of oldest for equal TDP (.7281/.9752) is 74.66% for 25.34% savings. Disregarding TDP, best newer .4905 (which isn't even v3) vs best Core2Duo .7082 is 69.26% for mere 30.74% saving over 8 years of advances. Servers run 24/7. The rest are part time. Most businesses don't run theirs 24/7, or even 9-5/5. Homes I have to guess average under 24 hours per week, depending on users per PC and whether home learning is their only reason to be present. It certainly doesn't look like doubling of capability or halving of consumption every 18 months. It looks much closer to one tenth that; nothing "vast", once actual usage is weighed against the other numbers. -- Evolution as taught in public schools is, like religion, based on faith, not based on science. Team OS/2 ** Reg. Linux User #211409 ** a11y rocks! Felix Miata
On Sun, Oct 09, 2022 at 02:42:47AM +0200, Stefan Brüns wrote:
On Sonntag, 9. Oktober 2022 01:32:20 CEST Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 00:42 (UTC+0200):
99% of users have bought sufficient hardware during the last 10 years.
Where did that number come from?
From common sense. Actually, during the last 10 years you could not buy anything new which is not x86_64-v2.
False. The HP microserves that do not qualify as v2 people mentioned here were intorduced 2012 and discontinued 2015, and during that time (and some after) you could buy them new. At least get the facts straight. Sure, if you knew that people will arbitrarily decide to not support hardware without SSE 4.2 you could easily get pre-owned hardware that qalifies probably even 15 years ago. But you would not know back then, sse4.2 is just one of dozens of obscure CPU features that only specialized software used - until now. And you could not know that there won't be sse4.3 and sse4.4, either. Thanks Michal
On Sonntag, 9. Oktober 2022 10:31:59 CEST Michal Suchánek wrote:
On Sun, Oct 09, 2022 at 02:42:47AM +0200, Stefan Brüns wrote:
On Sonntag, 9. Oktober 2022 01:32:20 CEST Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 00:42 (UTC+0200):
99% of users have bought sufficient hardware during the last 10 years.
Where did that number come from?
From common sense. Actually, during the last 10 years you could not buy anything new which is not x86_64-v2.
False.
The HP microserves that do not qualify as v2 people mentioned here were intorduced 2012 and discontinued 2015, and during that time (and some after) you could buy them new.
At least get the facts straight.
The HP(E) microserver N36L/N40L was introduced 2011, the N54L (2012) was only a minor refresh. HPE microserver Gen8 was introduced in 03/2014, the N54 was available till 09/2015. Why would you buy an N54 when you could get a Gen8 for the same or even lower price? After 11/2014, the N54 price went up significantly. So, one example of somewhat special hardware. We all know server(-like) hardware has a longer time-to-market, and is kept alive for an extended timeframe. On the other hand, x86_64-v2 has been available from AMD since 2011, and Intel since 2009. But as said several times, you can keep running these with Leap 15.x for quite some time.
Sure, if you knew that people will arbitrarily decide to not support hardware without SSE 4.2 you could easily get pre-owned hardware that qalifies probably even 15 years ago. But you would not know back then, sse4.2 is just one of dozens of obscure CPU features that only specialized software used - until now.
AMD K10 and Intel Core2 are the only two CPU families which provide more than x86_64 baseline, but do not qualify as x86_64-v2. The only common extension of these two is SSE3; but SSSE3, the various SSE4{a,.1,.2} extensions can only be found in one or the other. Most notably, neither Core2 nor K10 support CMPXCHG16, which is required for proper 64bit support without workarounds. So after all, the distinction between baseline and v2 does not seem so arbitrary too me. Regards, Stefan
On Sun, 9 Oct 2022 17:32:26 +0200 Stefan Brüns wrote:
Most notably, neither Core2 nor K10 support CMPXCHG16, which is required for proper 64bit support without workarounds.
The cpu flag cx16 is available in Phenom II and Core2. What is the difference to CMPXCHG16? Kind regards, Dieter
On Sun, Oct 09, 2022 at 05:32:26PM +0200, Stefan Brüns wrote:
On Sonntag, 9. Oktober 2022 10:31:59 CEST Michal Suchánek wrote:
On Sun, Oct 09, 2022 at 02:42:47AM +0200, Stefan Brüns wrote:
But as said several times, you can keep running these with Leap 15.x for quite some time.
Can you? It's very unclear what the future of Leap is, in fact.
Sure, if you knew that people will arbitrarily decide to not support hardware without SSE 4.2 you could easily get pre-owned hardware that qalifies probably even 15 years ago. But you would not know back then, sse4.2 is just one of dozens of obscure CPU features that only specialized software used - until now.
AMD K10 and Intel Core2 are the only two CPU families which provide more than x86_64 baseline, but do not qualify as x86_64-v2. The only common extension of these two is SSE3; but SSSE3, the various SSE4{a,.1,.2} extensions can only be found in one or the other. Most notably, neither Core2 nor K10 support CMPXCHG16, which is required for proper 64bit support without workarounds.
They actually do support CMPXCHG16, both core2 and K10, even some late K8. What is problematic is that the feature is misreported on some early revisions of core2. I am not really all that worried about the SSE extensions - benchmarks presented so far do not show great gains from opportunistic vectorization of general purpose code. Thanks Michal
On 09/10/2022 18.59, Michal Suchánek wrote:
It's very unclear what the future of Leap is, in fact.
https://en.opensuse.org/openSUSE:Roadmap says
Wed, May 31, 2023 Gold Master build is done
o Leap is GOLD
and usually it is supported 18 months from then or 6 month after the next version is out. That would mean there is support for Leap until at least end of 2024. However, I heard people discuss about longer maintenance of 15.5 as well. And we had the openSUSE Evergreen community support in past years as well. Ciao Bernhard M.
On Sun, 9 Oct 2022 17:32:26 +0200 Stefan Brüns wrote:
Most notably, neither Core2 nor K10 support CMPXCHG16, which is required for proper 64bit support without workarounds.
As already mentioned the cpu flag cx16 is available in Phenom II and Core2. Additional remarks: A) the workaround (if needed) is done by the compiler, B) according to the data it has practically no negative impact on the performance. Maybe the feature is already detected and used at runtime without requiring specific compiler flags. Kind regards, Dieter
Am 09.10.22 um 02:42 schrieb Stefan Brüns:
On Sonntag, 9. Oktober 2022 01:32:20 CEST Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 00:42 (UTC+0200):
99% of users have bought sufficient hardware during the last 10 years.
Where did that number come from?
From common sense.
Ah, yes. Finally some hard data.
anything new which is not x86_64-v2.
x86_64 has been available since late 2003, and x86_64-v2 was introduced early 2011 (Intel Sandybridge)/late 2011 (AMD Bulldozer).
So the first 7 years of x86_64 are baseline, everything after (11 years) is x86_64-v2. So even assuming not a single unit has been decommisioned, and shipments per year are the same (actually, it went up from 180M units in 2004, 310M in 2013 to 345M in 2021) 60% of all x86_64 PCs ever built are v2 or better. Taking growing markets and decommissioning into account, 99% is a quite plausible number.
(BTW: You question my 99%, but leave Larrys 99% claim without questioning it. Go figure ...)
Not all users buy. Many can only acquire whatever is available for no money.
Enterprises buy. Private persons buy. And this hardware ends up with persons who can not afford to buy sooner or later.
Regards,
Stefan
-- Thomas Zimmermann Graphics Driver Developer SUSE Software Solutions Germany GmbH Maxfeldstr. 5, 90409 Nürnberg, Germany (HRB 36809, AG Nürnberg) Geschäftsführer: Ivo Totev
On Fri, Dec 2, 2022 at 1:08 PM Thomas Zimmermann <tzimmermann@suse.de> wrote:
Am 09.10.22 um 02:42 schrieb Stefan Brüns:
On Sonntag, 9. Oktober 2022 01:32:20 CEST Felix Miata wrote:
Stefan Brüns composed on 2022-10-09 00:42 (UTC+0200):
Where did that number come from?
From common sense.
Ah, yes. Finally some hard data.
Thatnks for the coffee spill.. :-D While I think changing the baseline is ok as the hardware moves on.. I must also say that the data suggest performance improvement is marginal at the best..or a statistical error at worst..
Stefan Brüns composed on 2022-10-08 19:42 (UTC-0400):
Felix Miata wrote:
Not all users buy. Many can only acquire whatever is available for no money.
Enterprises buy. Private persons buy. And this hardware ends up with persons who can not afford to buy sooner or later.
And now with an arbitrary support lifetime cutoff, once they get them they'll be useful to the poor from 0-4 years. I'm not sure very many private persons give theirs up while still functional, unless to family members. Reasons to replace don't seem to include performance to much degree any more. I have 3 PCs newer than the 8 year old one I run 24/7 and am typing this from. Once you have NVME, there isn't a whole lot of improvement, if any, left to notice checking email, shopping, and wasting time on social media. Most people don't invest in hardware replacement just to save a little energy, particularly those who use A/C little to none and are happy to have the heat less efficient ones emit. -- Evolution as taught in public schools is, like religion, based on faith, not based on science. Team OS/2 ** Reg. Linux User #211409 ** a11y rocks! Felix Miata
On 09.10.22 00:42, Stefan Brüns wrote:
And IBM is a partner of SUSE, and there is likely a lot of money involved to keep these platforms supported.
So are intel and AMD, and I can just guess what reasons lead to the specific selection of features for "x86_64 archtitecture level v2" that -- so conveniently -- finally makes those old machines that do not have sse4.2 obsolete ;-) Remember, that architecture level is not something that was designed in from the beginning, it was invented after the fact by SUSE, RedHat, intel and AMD. IF there were pure technical reasons for the specific feature set, there certainly would be convincing benchmark results, wouldn't there? -- Stefan Seyfried "For a successful technology, reality must take precedence over public relations, for nature cannot be fooled." -- Richard Feynman
On Samstag, 8. Oktober 2022 16:20:58 CEST Fritz Hudnut wrote:
---------- Forwarded message ----------
From: Stephan Kulow <coolo@suse.de> To: factory@lists.opensuse.org Cc: Bcc: Date: Sat, 8 Oct 2022 08:13:11 +0200 Subject: Re: x86-64-v2: oggenc, zstd
Yes, systems mainly doing heavy work will be upgraded when faster hardware and budget is available. But the former high performance systems remain suitable for doing ordinary work for long time.
Yes. But the question remains on how many of those you plan to run tumbleweed on.
Greetings, Stephan
Stephan: As posted previously, I'm running TW and a number of Gecko Rolling, and Debian Sid, and Manjaro, along with Lubuntu developmental . . . on a '12 cMP using a Xeon i7 processor of that era
All Core i3/5/7 are at least x86_64-v2, so you can upgrade that to Leap 16.0 or whatever it will be called. Assuming a release in 2023 Leap 16.x will be supported until 2033, i.e. more than 20 years after the hardware has been released.
. . . . The plan is to run it until a number of parts blow up. I'm also running Leap 15.5 on an '09 MBP using Core2Duo processor, perfectly competent for daily driver.
F
You will be able to keep running this with Leap 15.5, which will likely be supported until 2028, you can (try to) keep this hardware running for 19 years. Regards, Stefan
Am Sa., 8. Okt. 2022 um 23:34 Uhr schrieb Stefan Brüns <stefan.bruens@rwth-aachen.de>:
You will be able to keep running this with Leap 15.5, which will likely be supported until 2028, you can (try to) keep this hardware running for 19
This is new to me. Where is this documented? I can't find it on https://en.opensuse.org/Lifetime Best Martin
On Saturday 2022-10-08 23:58, Martin Schröder wrote:
Am Sa., 8. Okt. 2022 um 23:34 Uhr schrieb Stefan Brüns <stefan.bruens@rwth-aachen.de>:
You will be able to keep running this with Leap 15.5, which will likely be supported until 2028, you can (try to) keep this hardware running for 19
This is new to me. Where is this documented? I can't find it on https://en.opensuse.org/Lifetime
Derives from https://www.suse.com/lifecycle/ I guess. SUSE Linux Enterprise Server 15 31 Jul 2028
Am So., 9. Okt. 2022 um 00:09 Uhr schrieb Jan Engelhardt <jengelh@inai.de>:
Derives from https://www.suse.com/lifecycle/ I guess.
SUSE Linux Enterprise Server 15 31 Jul 2028
Hrm. Did we do this for 42.3? Best Martin
On 09.10.2022 01:09, Jan Engelhardt wrote:
On Saturday 2022-10-08 23:58, Martin Schröder wrote:
Am Sa., 8. Okt. 2022 um 23:34 Uhr schrieb Stefan Brüns <stefan.bruens@rwth-aachen.de>:
You will be able to keep running this with Leap 15.5, which will likely be supported until 2028, you can (try to) keep this hardware running for 19
This is new to me. Where is this documented? I can't find it on https://en.opensuse.org/Lifetime
Derives from https://www.suse.com/lifecycle/ I guess.
SUSE Linux Enterprise Server 15 31 Jul 2028
I miss end of support for SLE 15 SP5 (which will be the end of support for Leap 15.5). And given that so far there was service pack each year, 2028 may well be SLE 15 SP9.
On 10/9/22 08:39, Jan Engelhardt wrote:
On Saturday 2022-10-08 23:58, Martin Schröder wrote:
Am Sa., 8. Okt. 2022 um 23:34 Uhr schrieb Stefan Brüns <stefan.bruens@rwth-aachen.de>:
You will be able to keep running this with Leap 15.5, which will likely be supported until 2028, you can (try to) keep this hardware running for 19
This is new to me. Where is this documented? I can't find it on https://en.opensuse.org/Lifetime
Derives from https://www.suse.com/lifecycle/ I guess.
SUSE Linux Enterprise Server 15 31 Jul 2028
There will be service packs for SLES 15 after SP5 so you can't really read anything into this other then given SP4 support will end 6 months after SP5, its probably reasonable to presume unless special magic happens that Leap 15.5 will be EOL when SLES 15 SP5 is which will likely be 6 months after the SLES 15 SP6 release. This is due to the fact that many security fixes come directly from SLE and those will stop at that point. -- Simon Lees (Simotek) http://simotek.net Emergency Update Team keybase.io/simotek SUSE Linux Adelaide Australia, UTC+10:30 GPG Fingerprint: 5B87 DB9D 88DC F606 E489 CEC5 0922 C246 02F0 014B
Stefan Brüns composed on 2022-10-08 23:33 (UTC+0200):
You will be able to keep running this with Leap 15.5, which will likely be supported until 2028, you can (try to) keep this hardware running for 19 years.
Time is a lousy measure. This v2 thing so far has proven entirely arbitrary, with no clear measure of loss or gain. It would condemn masses of poor people without means to buy new hardware to insecurity, creating yet more detrimental class division. Don't forget also, some users can only afford what they can get for free, probably most users in many countries. My 25 year old car only has 102,000 miles on it. But for its rust I would expect it to outlast me. I should be able to to expect the same for my internet access, record-keeping and written communication device, which isn't affected by flood-caused excess oxidation or exposure to public roads. v2 would not enamor many people to openSUSE, current users or prospective. It would chase away users and recommendations, and retard community growth. That's fine for SLE if that's its plan for increasing profitability, but not a good plan for a FOSS project that needs contributing users. IOW, v2 == bad karma. -- Evolution as taught in public schools is, like religion, based on faith, not based on science. Team OS/2 ** Reg. Linux User #211409 ** a11y rocks! Felix Miata
been working on computers since 1979, one of the hardest issues is hardware becomming unsuported , even though hardware is given a industry ideal expected life by manufacturers this is not the real world figures there guesses based on macines sold that are expected to replace what was previously sold , in most cases the actual number of machienes over the 10 years of age still in use is not a known figure and if it was passed on to someone that still uses it or failed and became e waste , anyway a pdp11/03 was installed as a city traffic light controler back in the 70's dec , it was not untill 2021 that it was replaced by a new system (at a cost of 15 milion) almost 50 years of in service active use , ok so not this issue , but you need to look at why it was not replaced earlyer hardware compatabilaty as it used custom i/o cards as well as custom software the same issues affect pc's though and no one has tuched on this in this descusion and this is verry important firstly you may have an old 586 machine still running , but why should you need to well here are the reasons 1/.isa card slots (ie for some obscure control monitoring card) process control or old programming hardware 2/. it still has a floppy port (hay i got lots of data still on 3.5" disks or even 5.25 or 8" disks with a cable adaptor 3/. obsolete i/o port standards (i know for example using a usb to parallel printer support fails when using parallel printer port stepper drivers because of usb lag (you need a proper inbult port for it )) 4/. it just works and does the job it was commisioned to do now its fair to say a 586 is not ideal these days , but if it fails you need to replace it , and heres the issue if it fails in 2 or more years time , then i would find that most v2 systems dont have the older i/o , many v0 v1 systems dont have isa for example and and almost none of the v2 dropping v0 v1 support is going to cut out old data recovery since a lot of v2,3&4 systems dont offer a floppy port that excludes the abilaty to recover old data from not only floppies but some of the old tape backup drives making those that need to looking for machines pre v0,1 older and harder to find not to mention the whole issue of power consumption same for those that need to have the abilaty to use isa i/o cards or even ide to recover files from a old hdd although it seems minor as those users may also have newer machines on hand , if they cant get a os to work then there screwed it may be fair to say they could use an historic archive for the os like say https://drive.google.com/drive/folders/1lR6JUARnlTSEh8GicWI3Ptll8aLJ-K4X?usp... (linux for machines circa 1996) but in this day and age it would not be advisable to use on a actave internet connected machine as its not patched and this is the other issue users v0 v1 will face , if support is dropped , securaty updates , those users would be left with no option but to fall back onto some archive that is nolonger maintained , because upgrading the hardware is not an option and in todays internet connected age securaty and patches are a big issue especialy since it seems to be the tend to drop leased line /rf links and connect them to the internet instead so they cant be secure or air gaped ! the main issue of dropping v0/v1 support is not just the hole it leaves but the fact its gong to make v0/v1 64bit hardware unsuported forcing the v0/1 users to revert to 32 bit archives as it seems the intent to drop it without having a fallback v0/1 legacy distro at the same time im all for optimisation support for v3,v4 and whatever next , but it cant be done by dropping support for those that depend on v0&1 support in the same way we still have 586 and 32 bit support and there is still a need for that also but as i pointed out its not all ways the cpu type that is important but the i/o the motherboard offers its fair to say ram requirement may be an issue on old machines but where hardware i/o is the issue a v0 or 1 system is more prefrable to a 586 system if the mobo will support it that v2,3and 4 systems nolonger support and where custom systems have been put in like machine monitoring and control they can only be replaced by those machines that still support the "special" expantion hardware yes there are still companies that pay server hall fees for there vax11/780 (running for what i dont know but prob because they have bespoke software on it maybe in fortran or cobol) also in terms of future sails (for business use ) keeping support is also good marketing for those that build new systems for such tasks knowing there going to be supported for a long time not just the 10 year expected pc life span , but control systems that may be expected to have a 50 year plus lifespan , scrapping a desktop pc because it wont support the latest windows is not as big a "e waste" issue as having to scrap the whole 30 million dollar autimated pick and place warehouse and i/o cards and equipment (a few hundered thousand tons of "e waste")that it was controling just because its nolonger supported and cant continue as its "unsafe to have it have it unpatched"
participants (23)
-
Andrei Borzenkov -
Bernhard M. Wiedemann -
Carlos E. R. -
Cristian Rodríguez -
Daniel Morris -
David Powell -
dieter -
Felix Miata -
Fritz Hudnut -
Jacob Michalskie -
Jan Engelhardt -
Jiri Slaby -
Larry Len Rainey -
Lukas Straub -
Martin Schröder -
Mathias Homann -
Michal Suchánek -
Neal Gompa -
Robert Webb -
Simon Lees -
Stefan Brüns -
Stefan Seyfried -
Thomas Zimmermann