Best CPU for hqplayer

[SwissBear]

On 11/29/2020 at 11:16 PM, Miska said:

If you are connecting DAC directly, one alternative is Z490 VISION D which has DAC-UP2 USB ports.

Is there an alternative which would be fanless or compatible with a fanless case ?

What are your thoughts about dedicated USB "Audiophile" cards (SOtM and the like). As they are very expensive, are they at least useful ?

[SwissBear]

On 2/10/2021 at 11:08 PM, Miska said:

Some load figures for the Ryzen 7 5800X with the new special build of HQPlayer 4 Desktop running on Ubuntu Desktop 20.04 LTS:

 

 

Client and terminal with ssh connection to the Linux machine running on Mac Mini M1 under macOS.

 

P.S. Part of this performance is thanks to the G.Skill DDR4-3400 CL16 RAM...

 

 

Hi Jussi,

 

Congrats for the perf of this build.

Would you have a typical config (mobo, memory, aso)  you would like to share with us ?

Thanks in advance :-)

[SwissBear]

I am considering building a PC (probably with dual boot Windows and Linux) for HQPlayer, to use as a server together with a NAA.

When looking at the performances displayed, it looks like may cores are idle. 

So is there a point to buy an i9 10900K vs i7 10700K ?

Would a Xeon with even less cores be an option ?

Also, is there a preference for Ryzen compared to Intel ? Which model then ?

Finally, what are the minimum specs for a graphic card (I am not playing games) just to offload some processing to the GPU) ?

On the French speaking forum, many users are advocating Windows Server 2019. Any interest to run HQPlayer ?

Thanks.

[SwissBear]

4 minutes ago, Miska said:

It depends on clock speeds, but likely you won't save money that way though.

 

Thanks Jussi.

 

It was more about lowering the power usage (80W vs 125W) in order to move to a LPSU if necessary.

 

Any idea about Intel vs Ryzen and Windows Server vs Windows 10 ?

[SwissBear]

56 minutes ago, Miska said:

Apart from Apple's M1, Core i5-7600T is lowest power (35W TDP) server I have capable of doing DSD256 with ASDM7EC:

 

 

Close to the limits, but still doing it reliably. In this case, cores running at 3.6 GHz.

 

This with the new AMD/Intel optimized build. OS is Ubuntu Server 20.04 LTS, with my custom kernel.

 

I think I'll want to test this with the new 10th gen T-series CPU as well.

 

 

Do you have any experience with the 10th generation E-series CPU (TDP 65W) ?

[SwissBear]

8 minutes ago, Miska said:

 

No, all my 10th Gen Core's are K-series...

 

Thank you. I will start with the K series, and see how much headroom I have given my expectations and if I can move down to the E series in order to operate fanless.

[SwissBear]

10 minutes ago, Miska said:

 

For me, fanless is not an issue for many reasons. One is that fans can be quiet enough. And second is that HQPlayer server can be placed outside of listening room where loud fans don't matter... So usually I just look at raw performance.

 

 

I'm on the same page, and very much enjoying the way you have implemented NAA for this purpose. Just thinking about other and complementary options. Thanks anyway. This will give me the opportunity to enjoy other implementations of your software. I'am already very enthusiastic about testing the Linux version. Pretty much less enthusiastic about the Windows flavour, but I will definitely give it a try :-)

[SwissBear]

I was surprised to notice that the new machine I assembled, based on an i7-10700, and running Windows or Linux Embedded was not able to overcome my Mac Mini M1 running HQPlayer. 

Same limitations in running DSD 256 with ASDM7EC modulator and poly-sinc-ext2 filter: limited in practice to Redbook and obliged to switch to other filters for higher res input formats (I am running a quite heavy 2 channels convolution though). 

Although the OS of the Mac Mini is far from being optimised, I have difficulties to see the upside of assembling an Intel machine for the sole usage of running HQPlayer.

Am I missing something ?

 

[SwissBear]

3 minutes ago, Zauurx said:

@SwissBear

 

In Bios, hyperthreading is disabled? (> only 8 cores) and the frequency blocked to the max (without going through the turbo)? like OC.

What is the motherboard ?
With 8 cores at a permanent frequency of 4.6 (4.7 turbo specification for this CPU), things should go smoothly.

 

Hyperthreading, as well as CPU frequency, are in Auto mode in the BIOS.

MB is Gigabyte Z490 UD. Memory is G.Skill Trident Z.

I'm not saying that things are not going smoothly. But not any smoother than the Mac Mini M1... 

[SwissBear]

31 minutes ago, Zauurx said:

 

Apple M1 is a machine developed globally: CPU, motherboard, bios, OS to obtain a result (and if possible the best).
In the case of a DIY machine, you (or me) are the builder.
So the intel CPU will give a certain result depending on the motherboard, memory and BIOS settings chosen by you.
If not, it is the "auto" mode wheel (russian wheel) to have a PC that works but does not give the maximum.
This is what I have already said elsewhere with my i5-8400 CPU (stuck at 4ghz without turbo. Gigabyte Z370N), it plays DSD 256 with HAF convolution even with 96hz - 64 bit (Roon volume).

 

 

Thanks for sharing and for your didactic efforts. I appreciate.

I notice that you are using poly-sinc-short-lp-2s as a filter.

Have you succeeded in the same conditions (192k source) with poly-sinc-ext2 ?

Thanks

[SwissBear]

2 hours ago, Jean Paul D said:

I'm supposed to get a M1 tomorrow and I'll start trialing HQP on it.

Do you have setting recommendations ?

Are there things to do to the OS knowing that I will also use the M1 for videos (via Safari, via IINA) and maybe to process photos too but certainly NOT for messaging chatting emails etc

Do you have an experience of using Hijack+Loopback to send all audio to HQP ? alternative solution?

Do you use Screen share ?

 

Hi Jean-Paul,

 

There is unfortunately not much we can do in order to optimise the OS. The last versions of Mac OS render these optimisations cumbersome and not practical. I just removed all the services I felt were not necessary (automated upgrade aso...).

I have no experience using Hijack and Loopback. I simply use this machine to execute HQPLayer and that's it. It's headless and I access it via screen sharing, using VNC Viewer on my Mac (as I find it more reliable than the standard Mac OS tool).

Hope this help :-)

[SwissBear]

9 minutes ago, ArcticSapien said:

That's a big-ass PSU!

I guess some HQP filters will max out the i9-12900 and run hot. Will likely need a Liquid Cooler. I'm also considering under-volting it slightly to control the temperature.

 

I invite you to be careful with liquid coolers as many of them are not supported under Linux if this is you goal. I was surprised to discover that the one I bought (Gigabyte Waterforce 280) needed a program running under Windows to adjust fans according to needs. This makes usage touchy IMHO. 

[SwissBear]

On 1/13/2022 at 8:37 AM, Bertel said:

Interesting - I have exactly the same or at least very similar memory modules, but can only get to 22.3 GB/sec…

 

Do I assume correctly that the mainboard (Gigabyte Z690 Aorus Elite DDR4) is resposible for this?

Hi,

Seems I have the same cache bandwidth limitations as you have with my Gigabyte Z690 UD DDR4 mobo.

Have you found a solution apart from replacing the motherboard ?

 

[SwissBear]

Hi Jussi,

 

Tried to buy i9-13900T to build a new server (for HQPe) and heard that it's not going to happen anytime soon.

What is you return of experience on the Ryzen 9 7900 processor ? Is it worth trying for DSD 512 upsampling ?

Do you care for ECC memory at all or do you rely on not overclocking to not care for ECC ?

Thank you.

[SwissBear]

5 hours ago, Mops911 said:

I tried mine for sure and it seems (I could only run to nulloutput, but the timer gained 10sec over 1min) that DSD256 with mq runs at less than 20w...

 

;-)

Not sure I understand what you are trying to convey…

[SwissBear]

i9-13900T is able to squeeze DSD512 with ASDM7EC-super and poly-sinc-gauss-halfband.

Quite hot though (with passive cooling). After one hour, CPU temp is volatile but oscillating btwn 70 and 100 C.

Not sure I am willing to go this way for long. 

Will see

[SwissBear]

10 hours ago, Miska said:

 

Mini-ITX in Fractal Design Ridge case. 750W Seasonic SPX PSU. This combo can run pretty decent GPU as well (space and cooling for a big 3-slot card), if necessary. Alternatively it can be water cooled.

Thanks for your answer. My expectations for passive cooling might have been too ambitious...

Could you please comment on the opportunity to have a GPU in complement to this type of CPU ?

Also I noticed that my Fedora distribution was not able to boot (I tried several bios versions, several "legacy" settings) which led me to use yours. Any clue where this could come from ? Using your distro is very nice but prevents me from using the machine for other purposes when not listening to music... 

 

10 hours ago, Miska said:

I can do more testing after Munich.

 

Wish you successful exhibition there 😉 

[SwissBear]

5 minutes ago, Miska said:

 

As long as it doesn't limit performance too much due to thermal throttling, does it matter what temps it runs?

Do you mean that the processor is protected against overheating and that there is no risk of "frying" it ?

5 minutes ago, Miska said:

Which way do you mean? I'm contemplating whether I would use the card slot for the RME AIO card for input purposes, or use it for example for some ASUS ROG STRIX series GPU (very quiet).

My intention was to understand what usage could a GPU have in complement to such a processor. I understand that your goal was to reach DSD512, which the CPU can achieve without GPU. Any other goal that adding a GPU could achieve ?

5 minutes ago, Miska said:

OK, that is strange. Fedora 37? Any errors or anything why? If it starts booting the kernel, remember that you can see the kernel boot log instead of splash if you hit ESC during the splash.

Thank you. I didn't know. Will look into that.

[SwissBear]

1 hour ago, chipvn said:

You could try to lower the default PL2 in bios down to about 75-85W (or lower), limit the max power cores clock down to about 4GHz and enable only about 2-4 efficient cores to see if it work better.

My passive cool 12700K can do Sync-MGa with ASDM7EC-super at DSD512 and the temp never exceed 65 degree C.

I only run 8 P-cores at max 3.6GHz / 2 E-cores at max 2.7GHz and 50W PL1 / 59W PL2.

Thanks for the tweaks. On my mb, CPU frequency adjustments are available only with "K" CPUs. But limiting the PL2 down to 65W seems rather efficient at reducing the core temp without altering the DSD512 capability. Thanks again.

[SwissBear]

2 hours ago, chipvn said:

Since you can limit the PL2, if you can play a little bit with Vcore (lower the CPU voltage), you can also significantly reduce the CPU temp. Default Vcore tends to be at the agressive side to ensure CPU can run at it highest speed within the given power setting.

Most of us don't need to run the CPU that agressive.

On my machine, what seems critical is the limitation of PL1 and PL2. The values you indicated work for me. Thank you.

[SwissBear]

2 hours ago, SwissBear said:

On my machine, what seems critical is the limitation of PL1 and PL2. The values you indicated work for me. Thank you.

I finally dropped my case and returned to "auto" parameters in the BIOS. This setting as well as the recent kernel I am using seem to make the very best out of this processor. When heat is excessive, load seems to be reallocated to a different core.

Reducing the number of E-cores did not seem to lower the temperature drastically. Reducing the max power resulted in stuttering occurring prematurely and this was very unpleasant to my ears and my speakers. Let's see how this modern processor handles things.

[SwissBear]

2 hours ago, Mops911 said:

Sorry , if I missed this, if I use sinc-MG with a file 96khz or more (but not 44khz) the track just does not start. Complex gauss modulator start but stutter, MG does not start at all. why?

Seems the sinc-mg is an integer filter only. If you have a 96k content, and a DAC which is not capable of DSD 48x, this will not work.

[SwissBear]

Hi everyone,

 

I'm contemplating upgrading my main machine which is a i9-12900-K with a GPU.
What would be the minimum graphic card sizing to process :

• PCM files at every sampling frequency
• DSD 64 and DSD 128

using ASDM7EC-super 512fs+ and demanding filters for PCM, with the constraint to process 4 impulses matrix convolution and stably output to DSD 512.
Would an RTX 4070 with 12 GB be sufficient or do I need to go to the RTX 4090 with 24 GB ?

Thanks in advance.

[SwissBear]

15 hours ago, SwissBear said:

Hi everyone,

 

I'm contemplating upgrading my main machine which is a i9-12900-K with a GPU.
What would be the minimum graphic card sizing to process :

• PCM files at every sampling frequency
• DSD 64 and DSD 128

using ASDM7EC-super 512fs+ and demanding filters for PCM, with the constraint to process 4 impulses matrix convolution and stably output to DSD 512.
Would an RTX 4070 with 12 GB be sufficient or do I need to go to the RTX 4090 with 24 GB ?

Thanks in advance.

@Miska Any recommendation on your side ? RTX 4070 or 4090 ? Thanks.

[SwissBear]

31 minutes ago, Miska said:

I used DDR5 6400 CL32 memory on my latest i9-13900T build and it seems to have really nice performance despite being low TDP.

 

Have you been able to produce DSD512 from all DSD formats with this system ? If I remember well, you have installed a small GPU on this system too. Haven't you ?

 

As you were referring to Apple silicon too, it will be interesting to have a data point for the new M2 Ultra. Given the price of the RTX 4090 GPUs, this Apple silicon could prove an interesting choice.