What’s so bad about cpu fans?

[Nenon]

2 hours ago, Miska said:

I think best would be to use objective data (measurements) to figure out where things are broken and not working as designed. So that it can be fixed for good

This is not the 'Objective-Fi' forum. This is the 'Music Servers' forum.

Nothing is broken. Music servers make a night and day difference in a good resolving system, even when the server is behind NAA. Those who think otherwise should stay away from this forum. We will talk about measurements here when our measurement tools get sophisticated enough to measure emotional impact.

 

@Miska Let us build state of the art server and enjoy them. It's our hobby that we enjoy and we are your clients at the end. Don't be a party pooper and take those discussions to the Objective-Fi forum if you want to comment further. Thank you.

 

1 hour ago, Miska said:

And to use full DSP potential, you need powerful CPU and GPU. And that practically means having cooling fans! Like i9-11900K + RTX 3090 combination, or Ryzen 9 5900X + RX 6900XT.

I completely agree with that statement! 

But that is if we need to do realtime upsampling/DSP. Many of us don't need that, and I don't remember the original poster mentioning anything about upsampling.

[sledwards]

17 minutes ago, Nenon said:

This is not the 'Objective-Fi' forum. This is the 'Music Servers' forum.

Nothing is broken. Music servers make a night and day difference in a good resolving system, even when the server is behind NAA. Those who think otherwise should stay away from this forum. We will talk about measurements here when our measurement tools get sophisticated enough to measure emotional impact.

Night and day difference? Certainly glad not in my high resolving system. Not up for a little spirited discussion, huh?

[RickyV]

11 minutes ago, sledwards said:

Night and day difference? Certainly glad not in my high resolving system. Not up for a little spirited discussion, huh?

That’s because a lot of people are sick of those never ending, leading to nowhere objective, subjective discussions. Both sides have their bias okey.

[Miska]

58 minutes ago, Nenon said:

But that is if we need to do realtime upsampling/DSP. Many of us don't need that, and I don't remember the original poster mentioning anything about upsampling.

 

Only reason for HQPlayer and NAA to exist in first place is upsampling and other DSP algorithms. To use math to enhance performance of already existing and new systems. And the development is objectively driven. This is the reason why I'm talking about this in first place.

 

If you don't need upsampling, things like CPU/GPU/case fans are non-issue in first place. You can run for example HQPlayer Embedded on RPi4 without upsampling with just bare heat sinks on CPU and RAM. Without any DSP processing it is non-issue in first place because you can run everything at less than 2W power CPU.

 

This is also point of NAA, that you can run all the DSP to (objectively) improve performance on a heavy ~1kW loud machine in the basement. And still the actual playback is performed by a low power CPU running even out of 4 AA batteries. Network keeping these two completely different devices completely separate.

 

 

When it comes to fans, they are much less of an issue than a spinning HDD. So before worrying about fans, you need to get rid of spinning HDD's. Although SSD's have their own issues too. This is where network comes to help, by isolating electrically noisy things.

 

[Rovo]

Miska,

one last short remark from my side. I started to notice the differences when upgrading the endpoint to the DAC with a high quality power supply (PH SR7T). You mentioned in another topic that the NAA endpoint you normally use profits from an upgrade of the power supply to a medical grade power supply. We have taken the quality of the power supplies much further and then we notice a difference. Seems we are on the same path with respect to hardware, only on different points of the path. Just to mention, no criticism on the software qualities of HQPlayer.

Kind regards,

Robert

 

[Foggie]

15 hours ago, Nenon said:

But that is if we need to do realtime upsampling/DSP. Many of us don't need that, and I don't remember the original poster mentioning anything about upsampling.

Good discussion and information in this thread.  Not to go too OT, but I'm curious about this comment's context.  I thought the primary purpose of HPQ has been pcm > dsd realtime upsampling (or pcm > hi rate pcm) and thus use of NAA?  Does your comment refer to not using HQP (or other upsampling software) as well as not using NAA and going DAC direct from server? 

 

This appears to be the two most popular "methods" => either DAC direct (from server) or a multiple computer setup (use of NAA or enpoints) to DAC.  In both scenarios good power and proper implementation is certainly important

[Rovo]

Sometime ago Nenon explained the alternatives to "deliver" the signal to the DAC. He will be able to point you to his post with the explanation. In short two main routes to follow are:

- no upsampling, using one server with two CPU's (one CPU doing the server task and the other CPU the endpoint task!), in this one server you really try to minimize "disturbances" like upsampling. So in this method you can still use HQPlayer (server task, but not advised to do upsampling!) and the NAA (endpoint task). The transport of the signal from HQPlayer to the HQPlayer NAA (so the communication of the server CPU to the endpoint CPU) is of a very high quality (credit to Miska). Just take a look at all the information on the Taiko Extreme.

- upsampling, separating the upsampling from the endpoint (the principle that I am following and that Miska decribes), in this one you really need to separate the PC's so the endpoint (e.g. HQPlayer NAA) does not get "disturbed"

 

Reason for me for following the upsampling path is my choice for the T+A DAC 8 DSD. This DAC performs best with DSD512. In the first method power supply is very important to both the CPU's, but this is one power supply because of the one motherboard. In the second method I noticed (and others) that not only the power supply to the NAA is very important, but also the quality of the power supply of the other PC's does matter.

[elan120]

7 hours ago, Rovo said:

In the second method I noticed (and others) that not only the power supply to the NAA is very important, but also the quality of the power supply of the other PC's does matter.

Very good information, thank you!

 

Just curious, what do you use as NAA that you find successful?  I have tried few NUCs in the past as well as connect upsampling server direct, in my system, going server direct show more positive result, but I am still thinking perhaps if a better NAA is used, the result would be different?

 

[Foggie]

7 hours ago, Rovo said:

Sometime ago Nenon explained the alternatives to "deliver" the signal to the DAC. He will be able to point you to his post with the explanation. In short two main routes to follow are:

- no upsampling, using one server with two CPU's (one CPU doing the server task and the other CPU the endpoint task!), in this one server you really try to minimize "disturbances" like upsampling. So in this method you can still use HQPlayer (server task, but not advised to do upsampling!) and the NAA (endpoint task). The transport of the signal from HQPlayer to the HQPlayer NAA (so the communication of the server CPU to the endpoint CPU) is of a very high quality (credit to Miska). Just take a look at all the information on the Taiko Extreme.

- upsampling, separating the upsampling from the endpoint (the principle that I am following and that Miska decribes), in this one you really need to separate the PC's so the endpoint (e.g. HQPlayer NAA) does not get "disturbed"

 

Reason for me for following the upsampling path is my choice for the T+A DAC 8 DSD. This DAC performs best with DSD512. In the first method power supply is very important to both the CPU's, but this is one power supply because of the one motherboard. In the second method I noticed (and others) that not only the power supply to the NAA is very important, but also the quality of the power supply of the other PC's does matter.

I hadn't read the info you cited from Nenon, interesting.  Especially running HQP on one processor and the NAA on the other processor all on same wkst?

 

Although the Taiko examples and references are all fine and well, 99% of the people aren't using or planning on spending $30k on a server - so context is important, especially as far $ is concerned.  I suppose one can use some of the same principles that have been gained with the likes of the Taiko implementation. 

 

Additionally, I don't know the percentage of headphone v.s. 2 channel users, but I would say they are polar opposites and one has to consider the info posted by a HP or speaker user.  Me, I can't stand the damn things (HP) 😀

 

As to the upsampling option you cited, I do separate all tasks with three computers (below), so hopefully I'm doing something correct and am similar to you as I upsample to DSD as well.

1. Small form factor HP ProDesk 600 G5 ( i7 9700) running Ubuntu server as roon core with one large capacity HD with music (in util room)
2. High end wkst (xeon processor, 64GB ECC mem, nvidia P4000 (in util room).  Stock 700w PS.
3. NAA / endpoint is optical rendu in audio rack (fiber run from network rack)

I would be interested to know if its possible or even remotely worth it to supplement the power supplies in #1+ #2 with high quality LPS, although my wkst #2 might be difficult.  Sorry to veer OT

 

 

 

 

[Rovo]

elan120, Foggie, 

 

some time ago I decided to use the T+A DAC 8 DSD as my DAC. Investigating on the internet this DAC is regarded as a very high quality DAC for its price and I had good experiences with the T+A DAC 8. Only drawback is that this DAC performs best with DSD512, so upsamling is required for streaming. Streaming is a requirement for me. Others sometimes choose for offline upsampling. See the PGGB thread. By the way I am a speaker user. So these a my basic choices. 

 

Also investigating on the internet, the path followed by Taiko seems to be currently the best. With my choices above I try to stay as close as possible to the path followed by Taiko and Nenon. I have choosen for a three PC setup as Foggie: 

- Server PC for Roon Server, i5-10600K, Windows 10 + AO 3.0

- Upsampling PC for HQPlayer Embedded, i9-10900K, Ubuntu Server 20.04 + Jussi's "magic"

- Endpoint PC for HQPlayer NAA, Xenon E-2236, Windows Server 2019 + AO 3.0 (limited to Windows because of the driver for DSD512 for the T+A DAC 8 DSD)

 

For the endpoint PC I try to utilize the information gathered from Romax, Ray-dude, Nenon and Taiko. High quality power supply, currently PH SR7T and next Nenon's V3 unregulated power supply. JCAT USB XE and NET XE cards and some more details,. So a high quality Endpoint with a more powerful CPU (described by Nenon to give a bigger fuller sound). In the past I used a mini-ITX motherboard with a <6W CPU, but I like this more powerful CPU better. Following this path I recently noticed that upgrading the power supply of the Server and Upsampling PC still mattered. To get on topic again, if better power supplies matter most likely all the comments on fans will matter too. 

 

What is the correct or best way, I do not know. But following the path described by Nenon and others I notice with every step an increase in sound quality of my system. That is what matters. Some times with our current knowledge it can not be explained why this or that improves the sound quality. The willingness of people to cooperate on this forum, experiment, exchange information and staying open minded, can bring us all advantages. It some times can be difficult to accept critism (constructive critism!!!), but from own experience the end result will be positive. 

[elan120]

16 minutes ago, Rovo said:

elan120, Foggie, 

 

some time ago I decided to use the T+A DAC 8 DSD as my DAC. Investigating on the internet this DAC is regarded as a very high quality DAC for its price and I had good experiences with the T+A DAC 8. Only drawback is that this DAC performs best with DSD512, so upsamling is required for streaming. Streaming is a requirement for me. Others sometimes choose for offline upsampling. See the PGGB thread. By the way I am a speaker user. So these a my basic choices. 

 

Also investigating on the internet, the path followed by Taiko seems to be currently the best. With my choices above I try to stay as close as possible to the path followed by Taiko and Nenon. I have choosen for a three PC setup as Foggie: 

- Server PC for Roon Server, i5-10600K, Windows 10 + AO 3.0

- Upsampling PC for HQPlayer Embedded, i9-10900K, Ubuntu Server 20.04 + Jussi's "magic"

- Endpoint PC for HQPlayer NAA, Xenon E-2236, Windows Server 2019 + AO 3.0 (limited to Windows because of the driver for DSD512 for the T+A DAC 8 DSD)

 

For the endpoint PC I try to utilize the information gathered from Romax, Ray-dude, Nenon and Taiko. High quality power supply, currently PH SR7T and next Nenon's V3 unregulated power supply. JCAT USB XE and NET XE cards and some more details,. So a high quality Endpoint with a more powerful CPU (described by Nenon to give a bigger fuller sound). In the past I used a mini-ITX motherboard with a <6W CPU, but I like this more powerful CPU better. Following this path I recently noticed that upgrading the power supply of the Server and Upsampling PC still mattered. To get on topic again, if better power supplies matter most likely all the comments on fans will matter too. 

 

What is the correct or best way, I do not know. But following the path described by Nenon and others I notice with every step an increase in sound quality of my system. That is what matters. Some times with our current knowledge it can not be explained why this or that improves the sound quality. The willingness of people to cooperate on this forum, experiment, exchange information and staying open minded, can bring us all advantages. It some times can be difficult to accept critism (constructive critism!!!), but from own experience the end result will be positive. 

Thank you again for the detailed information, @Rovo,

 

I am in similar path, but trying to find a good setup for the endpoint PC at the moment.  I also find more powerful CPU seems to bring better SQ, and I have been looking to build another endpoint with that in mind.

 

In your experience, do you think something powerful like i9-11900K would be a good candidate at all?  Or XEON processor is a better choice in endpoint application?  Is it possible to know which motherboard you are using with your endpoint?

 

I plan to start acquiring components in the next few weeks, but CPU and motherboard selections are still undecided.

 

Another test I am also considering is trying out different power supplies for the two XE cards in my server, and potentially the endpoint, and I am curious if you have done similar test in the past?

 

TIA for your inputs.  Perhaps it is better to bring these discussions to PM so I won't derail OP's original question.

[semente]

Would it be difficult to run the fan with its own independent external power supply?

[Rovo]

elan120, 

 

my choice for the CPU was based on advice of similar CPU's of Marcin at the JCAT-forum. The motherboard I am using is a Supermicro X11SCL-F. I must say that I like the stability of a server motherboard. Installation software is most of the time easier than on a commercial motherboard. The Xenon E-2236 doen not have a graphical part, this can also improve the sound quality. Also the use of AO 3.0 pushed me in the direction of a server motherboard and Windows Server. Is it better in sound quality than a i9-11900K? I really do not know. What I have learned is that it is important to pay attention to all the details (no weakest link). A good implementation will always beat a bad implementation even when using better components. 

 

I have tested a SOTM SPS500 power supply, a HDPlex power supply and the current PH SR4 on the JCAT USB XE card. The SOTM sound was smooth, the HDPlex was more dynamic, but harsh and the PH SR4 combined the best of the two. 

 

If you want to discuss more details you can alwys PM me. 

[Miska]

For NAA, if you like high rate PCM (1.5M) or DSD1024 output, best option is some Intel Atom SoC based device. Powered from a suitable PSU or a battery.

 

For lower rates, you have various options with low power ARM CPUs. Powered from a suitable PSU or a battery.

 

Try to go for lowest power device that fits the bill. This also makes the device small and easy to have passive cooling.

 

Primarily, as much as possible, try to use Linux-based NAAs and avoid big bloated massive OS like Windows.

 

Smallest device I've been running NAA on is 400 MHz ARM9 CPU with 64 MB of RAM.

 

 

Most optimized out combo is:

• HQPlayer OS running as HQPlayer Embedded
• NAA OS running as NAA

 

[kelvinwsy]

I hv the 11900k as my new HQplayer Server and intend to run my previous 10700k as the NAA ...Presently I use the SOTM SM200 Ultra Neo as the endpoint .. also a quad core Atom minipc that was my previous NAA. I find that the ability of the VRM to deliver big current / big power delivers a very dynamic rich SQ!

 

Yes it is possible to power fans externally! I hv all my fans on an externally fan controller and separate PSU. Even the AIO Pump can be powered this way but I lose Pump RPM And cooling water flowrate ... so I leave this connected to the Mobo . but the pump is only Voltage regulated so no PWM fluctuations like the 4 pin fans.

 

Your mobo BIOS must allow bypassing of CPU Fans .. otherwise u need a special fan PCB work around solution . PM Me if u wish to discuss

[kelvinwsy]

I mean flowrate and rpm readout from the bios....

[Rovo]

kelvinwsy, 

I would be interested to know how you controlled your (CPU) fans separate from the motherboard. Would you like to share some details (special fan PCB)?

 

Kind regards, 

Robert

 

[Miska]

9 hours ago, kelvinwsy said:

I find that the ability of the VRM to deliver big current / big power delivers a very dynamic rich SQ!

 

Remember that all those VRMs on motherboard are switching regulators, essentially SMPS. So even if you use LPS to power your machine, you have a lot of big high current SMPS running there. For big CPUs we are talking about 1.2V 100A kind of stuff.

 

[kelvinwsy]

Yes of course ! I prefer to hv that big current rather not. I can hear the improvemt in SQ from Gigabyte Z490 Aorus Ultra - 55amps VRM to the Gigabyte Z590 Aorus Ultra - 90 amps.. Anyway I will run the Z490 NAA vs tge Sotm Sm200 Ultra Neo and report back

[kelvinwsy]

Btw I dont use Lpsu on the CPU and the ATX ! Tried the HDPlex 400w LPSU (old verdion) - it was wimpy sounding! U need big peak current delivery PSU for 11900k EXC/DSD256 upsampling on your SW😅! I use a Corsair RM1000X 

[Miska]

For regular machines I use two PSU models.

 

For bigger machines with big GPU, I use Seasonic Prime TX 1000W and for smaller machines without GPU the fanless Seasonic Prime Fanless TX.

 

Note that the also the regular Prime TX has hybrid cooling system and turns off the fan at lower loads and then as necessary gradually spins it up. So those are in most cases essentially silent.

 

Since most of my machines run 24/7, I like their 150000 hour MTBF and 12 years warranty.

 

[kelvinwsy]

8 hours ago, Rovo said:

kelvinwsy, 

I would be interested to know how you controlled your (CPU) fans separate from the motherboard. Would you like to share some details (special fan PCB)?

 

Kind regards, 

Robert

 

Rovo 

I sent u a Private Mail so as not to clog up this thread

[kelvinwsy]

Actually 3 PMs🤨

[MarcelNL]

M y experience with a PC not doing any OS upsampling, or anything other than pulling bits from a disk or the net and pushing them into a NOS DAC is that CPU processing power rules sound quality. My speaker system probably classifies as (highly subjective, as who is the judge) 'ultra revealing'.

 

Coming from an i5 NUC that on it's own had more than enough CPU power to accomplish the task at hand on a linear PSU I started with a Ryzen 3, then a Rysen 5, followed by a Ryzen 7 and now on a Ryzen 9 and each and every increment brought more and better sound.

Dynamics, micro detail, slam, low end extension and control, 'colors' within overtones, etc , they all improve!

 

Motherboards also have quite an impact, the motherboards capable of more 'direct' power improve sound quality.

I'm now using the Corsair AX1200i (which IMO is a Seasonic with a different badge, or vice versa), which is for all intents and purposes fanless.  

[elan120]

19 minutes ago, MarcelNL said:

Coming from an i5 NUC that on it's own had more than enough CPU power to accomplish the task at hand on a linear PSU I started with a Ryzen 3, then a Rysen 5, followed by a Ryzen 7 and now on a Ryzen 9 and each and every increment brought more and better sound.

Dynamics, micro detail, slam, low end extension and control, 'colors' within overtones, etc , they all improve!

Just curious, is this an endpoint or a server connected directly to DAC?