Ubuntu 21.04 vs. Windows 10 Trade Blows On The AMD Ryzen 9 5900HX / ASUS ROG Strix G15

[oiaohm]

I had the same experience before, it's so damn slow and luggy! I couldn't understand how ppl can use it.
The fact is, that Windows 10 just can't work on HDD
But when I installed it on machine with SSD - everithing became as fast as on my lovely Arch.

Number 1 Windows 10 and 11 specification still says it support spinning harddrive. So its still valid for a OEM to make a Windows 10/11 machine without a SSD. So this is a case Microsoft need to either fix windows or at least update the specification sheet.

Next for a person like me who deals with databases and blender Linux comes out ahead even on SSD. When you did deeper this is a IO operations issue. Windows generally is using more IO operations than Linux doing the same things yes this is where the slow and luggy is coming from with spinning rust harddrives. Yes you have tones of ram and windows does not effectively use it for disc cache either. There is another side effect to increased IO operations with windows turns out with SSD to be generally higher SSD running temperatures with windows vs linux. Yes that SSD running temperatures difference between windows and Linux can be difference between a SSD overheating resulting stalling or not overheating so running perfectly fine.

So the difference between windows and Linux due to storage either is reduced to basically nothing if you have a fast SSD with high IO operations able to performed and you are not doing any high IO operation work. If you are doing high IO operation work you will notice overhead under windows that you are not getting as much out the SSD as you should because windows is using some. Worst case with SSD cooling is not ideal and you have stalling out under Windows but not under Linux again the IO difference. Yes and this IO difference appears with a spinning rust harddrive as totally bad performance because the higher price to perform IO with a harddrive vs a ssd. This is a case just because you can skip having a disc cache and use direct IO instead does not mean you should.

I am also expecting when we have direct IO from storage to video card to have high odds of stutter under windows than Linux due to this poor use of cache by windows resulting in lots of IO. Yes this excess IO also is something that makes running windows virtual machines worse.

Yes some areas windows 10 beats Ubuntu because of the different architecture level the binaries are built with. You also see Ubuntu lose to Clear Linux OS developed by Intel in the same areas. This is a trade off to allow installing on older hardware.

https://www.phoronix.com/scan.php?page=news_item&px=Glibc-2.33-Coming-HWCAPS

There is work to fix this problem with glibc 2.33 most distributions don't have this yet and will take a while to get it in place.

RedEyed the problem here Linux is a fairly safe bet. Linux is a lot more tolerant to being ripped out of one machine and put in another in case of issue as well yes this will get worse with Windows 11 TPM setup.

RedEyed something to be aware is I am still running computers with current version testing version of debian as desktops that Windows 10 will not install properly on because they are too old. Yes they have sata SSD.


Interest enough SSD not all of them are created equal and lot them when they age end up slower than spinning harddrive at times. Another fact is a spinning rust harddrive can maintain its performance right up to 95 percent full. SSD can slow down in performance at about 80% full. Interesting enough a SSD over 80% full slowed due internal fragmentation makes your old spinning rust hard-drives look fast.

Sorry RedEyed if you are needing to have a SSD to fix a performance problem you better be sure not to fill as much as you can fill a harddrive. This cache issue with windows is problematic for those with HDD and SSD. People with HDD notice the problem straight way. People with SSD its more minor things at first but once they have really filled their SSD having windows stall out while attempt delete files to get SSD back to the correct fill level is not fun. Linux of course does not stall out badly with a over filled SSD and its for the same reason Linux handles a HDD better.

SSD is a true double sided sword to fixing performance issue because you fix the issue in one place but then move it to another.

[RedEyed]

Number 1 Windows 10 and 11 specification still says it support spinning harddrive. So its still valid for a OEM to make a Windows 10/11 machine without a SSD. So this is a case Microsoft need to either fix windows or at least update the specification sheet.

Next for a person like me who deals with databases and blender Linux comes out ahead even on SSD. When you did deeper this is a IO operations issue. Windows generally is using more IO operations than Linux doing the same things yes this is where the slow and luggy is coming from with spinning rust harddrives. Yes you have tones of ram and windows does not effectively use it for disc cache either. There is another side effect to increased IO operations with windows turns out with SSD to be generally higher SSD running temperatures with windows vs linux. Yes that SSD running temperatures difference between windows and Linux can be difference between a SSD overheating resulting stalling or not overheating so running perfectly fine.

So the difference between windows and Linux due to storage either is reduced to basically nothing if you have a fast SSD with high IO operations able to performed and you are not doing any high IO operation work. If you are doing high IO operation work you will notice overhead under windows that you are not getting as much out the SSD as you should because windows is using some. Worst case with SSD cooling is not ideal and you have stalling out under Windows but not under Linux again the IO difference. Yes and this IO difference appears with a spinning rust harddrive as totally bad performance because the higher price to perform IO with a harddrive vs a ssd. This is a case just because you can skip having a disc cache and use direct IO instead does not mean you should.

I am also expecting when we have direct IO from storage to video card to have high odds of stutter under windows than Linux due to this poor use of cache by windows resulting in lots of IO. Yes this excess IO also is something that makes running windows virtual machines worse.

Yes some areas windows 10 beats Ubuntu because of the different architecture level the binaries are built with. You also see Ubuntu lose to Clear Linux OS developed by Intel in the same areas. This is a trade off to allow installing on older hardware.
https://www.phoronix.com/scan.php?pa...-Coming-HWCAPS
There is work to fix this problem with glibc 2.33 most distributions don't have this yet and will take a while to get it in place.

RedEyed the problem here Linux is a fairly safe bet. Linux is a lot more tolerant to being ripped out of one machine and put in another in case of issue as well yes this will get worse with Windows 11 TPM setup.

RedEyed something to be aware is I am still running computers with current version testing version of debian as desktops that Windows 10 will not install properly on because they are too old. Yes they have sata SSD.


Interest enough SSD not all of them are created equal and lot them when they age end up slower than spinning harddrive at times. Another fact is a spinning rust harddrive can maintain its performance right up to 95 percent full. SSD can slow down in performance at about 80% full. Interesting enough a SSD over 80% full slowed due internal fragmentation makes your old spinning rust hard-drives look fast.

Sorry RedEyed if you are needing to have a SSD to fix a performance problem you better be sure not to fill as much as you can fill a harddrive. This cache issue with windows is problematic for those with HDD and SSD. People with HDD notice the problem straight way. People with SSD its more minor things at first but once they have really filled their SSD having windows stall out while attempt delete files to get SSD back to the correct fill level is not fun. Linux of course does not stall out badly with a over filled SSD and its for the same reason Linux handles a HDD better.

SSD is a true double sided sword to fixing performance issue because you fix the issue in one place but then move it to another.

I have nothing to add to your conclusions. I agree with everithing, linux rly has much better performance and I know it because I'm developing crossplatform AI software.

Regarding old hardware: my GF had linux (guess who installed it?) on ancient laptop (it's about 13 years old), installing windows 10 was not straightforward (old BIOS could read only FAT32, but Windows iso is > 4GB) , but afterwards, everything worked, even sleep and hybernation, and time on battery increased from 2 hours on linux to 4 hours on windows. (yes, I upgraded that laptop before: added sata SSD and RAM, because it was unusable even on debian xfce)

After beeing a linuxoid more than 10 years and propagating/moving to linux all my friends/relatives
One day, I lost interest in fixing linux software bugs, filling in issues, answering on quiestions on linux forums/stackoverflow, waiting for nvidia wayland support, waiting for bluetooth HSP/HFP support and many other things that should work in 21 centuary for ordinary users on PC/laptop, so I just gave up.

After installing Windows, I started doing things, that I wanted to do. On linux, first I had to configure/fix something, and then do things.
I decided just to pay money and have fewer issues and much better UX instead of waiting/contributing into linux desktop.

[RedEyed]

That's fair, and I think that's how it should be. People should use what's working for them. Linux and Ubuntu work best for me right now, that's not to say that won't change in the future, but I definitely won't be moving away from Linux. May use MacOS on the side since it's based off of Unix, but if that was the case, why not just use Linux? That's why I'm personally interested in the new Apple chips gaining Linux support. That's a win for everyone.



More my experience was well. I have almost all the Windows in a VM that I go to from time to time, all of them have some use for me believe it or not. Plus I'm a sick son of a gun, and I like that stuff. But Linux is where the heart is, always

I used to use Windows 10 in VM to build binaries. Also, to save my SSD space, I placed VM on my HDD...
Frankly speaking, when I had to run winows in VM, I had increased blood pressure and I was very very angry, every time I start Windows VM on my HDD.
Because it was so damn slow, and during this, I heard my HDD, it was so loud that I was thinking it can be damaged.
I still hate windows for this, that it can't work on HDD, while any other linux distribution has no problem with this.

[RedEyed]

What on earth is going on with zstd and (to a lesser extent) 7-zip? Why would that be so much faster on Windows? After all isn't compression mostly CPU bound (in which case the differences should be minor)? And if it was IO bound shouldn't it favour Linux?

This makes me suspect there is a bug or missing feature with regards to power management, frequency scaling or something of that nature.

I don't know which version of zstd was used, but before v1.5.0 multithreading build option was not by default for linux
Release Zstandard v1.5.0 · facebook/zstd (github.com)

Now the dynamic library (typically libzstd.so.1 on Linux) supports multi-threaded compression by default.

[oiaohm]

Regarding old hardware: my GF had linux (guess who installed it?) on ancient laptop (it's about 13 years old), installing windows 10 was not straightforward (old BIOS could read only FAT32, but Windows iso is > 4GB) , but afterwards, everything worked, even sleep and hybernation, and time on battery increased from 2 hours on linux to 4 hours on windows. (yes, I upgraded that laptop before: added sata SSD and RAM, because it was unusable even on debian xfce)

After beeing a linuxoid more than 10 years and propagating/moving to linux all my friends/relatives
One day, I lost interest in fixing linux software bugs, filling in issues, answering on quiestions on linux forums/stackoverflow, waiting for nvidia wayland support, waiting for bluetooth HSP/HFP support and many other things that should work in 21 centuary for ordinary users on PC/laptop, so I just gave up.

Most of my machines are desktops. I gave up on having Nvidia graphics a few years back and that make my life so many times better under Linux. Either Intel IGPU or amd dGPU and everything been better.

Linux on laptops random laptops has been a nightmare. I have seen a laptop that is design for Linux do the exact other way when a person installed windows. In fact worse went from 8 hours under Linux to 1 hour under windows. Its absolutely critical in a laptop that power management in fact works or you are in for a very bad time. Yes it can be stupid with like CPU or GPU can be downclocked yet its pull more power than when it running at it rated clock speed. Yes there are laptops design purely for Linux out there. There are also laptops designed purely for windows and there are limit group that are design for both.

The reality is fighting with hardware that is not design for Linux does result in filling in a lot more issues. This does not make Windows the better choice over all. Just means if you have laptop design for windows with power management only for windows putting Linux on it will give you bad time. Nvidia graphics with Linux more often than not gives you a bad time. Intel and AMD for graphics behaviour is better with Linux. Horrible part here is AMD graphics drivers under windows are absolutely horrible quite often. This is a big problem the selection of parts to built a computer ideal for Linux absolutely don't match the selection of parts to make a computer ideal for Windows there is a overlap but its not as big as most would presume.

https://maurom.com/blog/2021/03/20/support-for-hsp-hfp-headset-profiles-on-ubuntu-and-debian/

bluetooth HSP/HFP support<< you need to take with a serous grain of salt. The bluetooth standard for HSP/HFP is absolutely horrible 8khz audio that makes land line sound good. The reality is Linux has support what is in the bluetooth standard for HSP/HFP for a long time. The add on codecs the stuff that make HSP/HFP sound good are not part of the bluetooth standard so not part of the free patent grant of bluetooth.

There are a lot of cases where the issue with Linux is that the house is not in order in that area. People do have the problem with some brand of bluetooth headsets not working with some brands of phones or if you have different drivers installed in windows results in one head set not working and the other one is (yes this can be really annoying where A headset works with A drivers installed but not with B and B works with B drivers installed but not with A so the result is you cannot use both at once). This is because the bluetooth HSP/HFP is a disaster zone of vendor fragmentation with case of not forward/backward compatible codecs and these codecs have not been added to the standard for bluetooth. Please note bluetooth HSP/HFP is not the only area like this.

Yes it the 21 century and we still have not worked out how use a standard process properly and make properly compatible hardware.

[dnebdal]

If so then this is not a fair comparison and it should be done again.

It's completely fair, it's just not testing what you think it is.

Using the software as shipped, or even compiled from source with the default flags, is a test of the out of the box experience. How many users bother researching the best compiler flags for their specific hardware before using a compressor? If reasonably common software performs badly on a laptop/distro/OS without tuning, then it's simultaneously true that
a) you're not seeing the best performance it's possible to get out of the thing being tested,
b) this should probably be fixed upstream, and
c) for most users, "not running common things well out of the box" is a drawback even if there are ways to fix it.

Testing without installing ROCm and recompiling zstd for openCL support is much more representative of the experience of running Ubuntu on this laptop than doing so would be.

[RedEyed]

Most of my machines are desktops. I gave up on having Nvidia graphics a few years back and that make my life so many times better under Linux. Either Intel IGPU or amd dGPU and everything been better.

Linux on laptops random laptops has been a nightmare. I have seen a laptop that is design for Linux do the exact other way when a person installed windows. In fact worse went from 8 hours under Linux to 1 hour under windows. Its absolutely critical in a laptop that power management in fact works or you are in for a very bad time. Yes it can be stupid with like CPU or GPU can be downclocked yet its pull more power than when it running at it rated clock speed. Yes there are laptops design purely for Linux out there. There are also laptops designed purely for windows and there are limit group that are design for both.

The reality is fighting with hardware that is not design for Linux does result in filling in a lot more issues. This does not make Windows the better choice over all. Just means if you have laptop design for windows with power management only for windows putting Linux on it will give you bad time. Nvidia graphics with Linux more often than not gives you a bad time. Intel and AMD for graphics behaviour is better with Linux. Horrible part here is AMD graphics drivers under windows are absolutely horrible quite often. This is a big problem the selection of parts to built a computer ideal for Linux absolutely don't match the selection of parts to make a computer ideal for Windows there is a overlap but its not as big as most would presume.

https://maurom.com/blog/2021/03/20/support-for-hsp-hfp-headset-profiles-on-ubuntu-and-debian/

bluetooth HSP/HFP support<< you need to take with a serous grain of salt. The bluetooth standard for HSP/HFP is absolutely horrible 8khz audio that makes land line sound good. The reality is Linux has support what is in the bluetooth standard for HSP/HFP for a long time. The add on codecs the stuff that make HSP/HFP sound good are not part of the bluetooth standard so not part of the free patent grant of bluetooth.

There are a lot of cases where the issue with Linux is that the house is not in order in that area. People do have the problem with some brand of bluetooth headsets not working with some brands of phones or if you have different drivers installed in windows results in one head set not working and the other one is (yes this can be really annoying where A headset works with A drivers installed but not with B and B works with B drivers installed but not with A so the result is you cannot use both at once). This is because the bluetooth HSP/HFP is a disaster zone of vendor fragmentation with case of not forward/backward compatible codecs and these codecs have not been added to the standard for bluetooth. Please note bluetooth HSP/HFP is not the only area like this.

Yes it the 21 century and we still have not worked out how use a standard process properly and make properly compatible hardware.

All my and others bluetooth headsets (I have experience with Sennheiser and Sony) work on Android phone w/o additional software, same on Windows, I don't need to install anyting to be able to use my BT headset with mic.

What you say, about drivers: I didn't have such situations despite my short period of using win10 (about 4 month), sure, I don't reject that such problems could exist.
In my opinion, the rule is simple: more users -> more budget -> more effort to keep/raise userbase .

[oiaohm]

All my and others bluetooth headsets (I have experience with Sennheiser and Sony) work on Android phone w/o additional software, same on Windows, I don't need to install anyting to be able to use my BT headset with mic.

https://forums.tomsguide.com/threads...-my-pc.425395/ This is a works for me answer who does not understand the problem. Since you windows machine is setup for Sennheiser/Sony If you try to pair a samsung headset it not going to work. Yes if paired a samsung headset first the sony and sennheiser are not going to work. I could give other brand combinations that result in screwed. Not all android phones accept sennheiser and Sony headsets either. Yes windows automatically installing drivers for you does hide that you have possible locked self in.

I have a raspberry pi 4 setup to emulate the 60 codec combinations for bluetooth headsets. Basically have the little pi pretend to be a stack of different headsets this is cheaper and more compact than having all them to test with. Fun point is you can cause windows 10 to instant crash by connecting bluetooth headsets in the right order. It worse its a boot loop. Linux may not support bluetooth HSP/HFP well but at least you cannot end up in a never end boot loop that requires a reinstall to fix.

There is also a bigger problem with the bluetooth standard in some of your cheaper phones are bluetooth 5 only. That right don't accept bluetooth 4 devices because bluetooth 5 does not mandate backwards compatibility support. Sennheiser don't in fact make any bluetooth 5 headsets. I would be guessing you sony are not 5 either.

What you say, about drivers: I didn't have such situations despite my short period of using win10 (about 4 month), sure, I don't reject that such problems could exist.
In my opinion, the rule is simple: more users -> more budget -> more effort to keep/raise userbase .

https://www.youtube.com/watch?v=CPZkfYWp9uo

What Gamigo does to MMO explains why "more users -> more budget -> more effort to keep/raise userbase" is not always true. You have missed greed as a critical feature.

There is a simple one with bluetooth headsets. I sell you a phone I want to sell you the accessories I make. If I can make a minor alteration in by bluetooth drivers that cause my competitors headsets not to work well and sometimes not all this can result in me being able to see more accessories to the people who bought my phone right. This is a game of vendor locking leading to profit. This basically the Gamigo pay to win store model expect its your bluetooth headset.

The reality is "bluetooth HSP/HFP" headsets is really a dumpster fire of problems that comes really clear once you have testing setup up and running checking compatibility. It should not be if everyone had added their features to the standard instead of doing their own stupidity out of a objective of greed.

I can tell you that one car vendor that was considering allowing the entertainment system to support bluetooth headset after seeing the mess decided not to. Yes this was to allow the people in car to listen to music to be less distracting to the driver. I really do wish pulseaudio, pipewire and bluez the best of luck making a solution that deals with this mess correctly.

The fact that Bluetooth has a standard and it is this far stuffed up just show how down right crappy companies can be.

[Dr. Righteous]

After being a linux user, linux developer and windows hater more than 10 years I decided to give windows 10 a try.
Nowadays (for me) it works blazingly well (mb awful windows experience in the past was due to slow hdd, low RAM and awful C++ development tools)
I switched all my work and user workflow to Windows. I don't want to say this, but I have no issue with windows 10, I don't need to constantly administrate my linux desktops, I just do what I need w/o configuring OS.
Currently I'm a Python guy, so I didn't notice any difference when switched from linux to windows in terms of working experience.
Sure, I love linux and use it for servers, but as for user - I chose windows 10, everything that I need already works very well, everybody support windows and I can finally buy a lot of good software.
What I really miss in windows is autocompletion in the terminal.

Yeah, I've really not touched Windows since the Win7 days. Every once in a while I'll have to accomplish a "windoze only" something so I sit down at my son win10 gaming station. I find myself saying. "hmm, thats cool" "wow, thats neat". But trade it for my up 24/7 linux workstation. Na.

[smitty3268]

I don't know which version of zstd was used, but before v1.5.0 multithreading build option was not by default for linux
Release Zstandard v1.5.0 · facebook/zstd (github.com)

Ding Ding Ding!

I assume Michael tested the Ubuntu default, which is 1.4.8, and he almost certainly downloaded the current 1.5.0 version for the Windows test.

Not only does that change that multithreading default you found, it has massive performance improvements to compression levels 5-12. The notes show it's easy to get > 100% gains.