Best CPU for hqplayer

[Carousel]

I have my new case with integrated water cooler with pre-routed tubes, a Cooler Master NR200P MAX. I am just waiting on possible Black Friday sales. I am pretty sure that a 13700K would be nice enough. Just want to see if the price of the 13900K will get any better.

 

I am still not sure from reading through all your expert posts if the i9 brings anything vs. an i7 for HQPE anyway. Comments appreciated.

 

I am also watching some GPUs. Though testing with my son’s 9700K CPU and 2070 GPU did not seem to bring notable benefit by adding the GPU. If there is a good Black Friday deal on a 4060 GPU though, he might get one and then I can inherit the 2070 GPU.

 

Some of the MoBo’s I am looking at… Z790 and B760 ITX boards… are limited on their DDR5 speeds. But shouldn’t 6000MHz already be pretty fast?!

 

Another RAM question comes up, is there any real benefit from using 64GB vs 32GB for HQPE? I am still thinking of using Ubuntu Server with HQPE installed. Would that ever use more than 32GB? I might just get one 32GB stick and leave open getting a second one later.

[Carousel]

This is kinda an important area for me to clarify, since I do both stereo and 7.1 playback: Does having more than 8 cores matter? And is the answer different depending on whether an Intel CPU is involved or an AMD CPU is involved? I read comments here and there in the forums about stereo processing with Intel CPUs using only the two first cores, not more. But these pictures for an AMD 7950X seem to show fairly even spreading of stereo workload across its 12 cores and/or 24 threads. Which, if that is what can happen, makes me think that there would be a better chance of not needing a GPU as well. And then I would want to make sure I know how to make this happen, what settings and supplemental software if any etc.

[Carousel]

Thank you Miska. Thank you all for feedback and examples above. I went back and did more testing on my AMD mini-PC and the borrowed Intel machine with a 9700K and a 2070 GPU. I made better use of Windows and htop resource monitors too. Although the GPU was not needed for DSD256 or DSD128 downscaling to PCM 24-176 (multiple of 44), the GPU did make downscaling to 24-192 (multiple of 48) possible with DSD128. I do not really need that for my use cases. But it meant that I could find a "breaking" point where it made the difference with this particular equipment. 

 

I also saw indeed that multiple cores were in use with both AMD and Intel CPUs when processing stereo. So that is solved for me. Thank you for the explanation above as well so I can understand why. 

 

Don't know yet what I will buy next weekend. AMD CPUs are usually less expensive. But the motherboards I am looking at are cheaper for Intel. So prices would net for me today. I have lots of shopping options loaded as favourites though. So I am ready for next weekend.

 

Regarding RAM. From my testing today, 64GB probably is not needed. And I understand that 2x16 is probably better than 1x32. But it would be easier to upgrade to 64GB later... say for using for Plex server transcoding... or just for when the prices drop. 

[Carousel]

The answer depends a bit on your budget and what level of HQPlayer complexity you want to process. If you just use gauss filters and do not go beyond DSD256, then lots of machines will do.

 

But to keep it simple, when configuring a machine, make sure to select Nvidia GPU as a requirement, as CUDA acceleration helps to get beyond DSD256. That will probably return you at least i7 machines. Then buy up to your budget.

[Carousel]

Just built and started my new HQP5E server running Ubuntu Server. Have a 14700K CPU, 4060Ti GPU, 64GB DDR5 and a 2.5 gig ethernet. Still need to get my new embedded license file still though. I only just wrote for this with the new fingerprint. Am in Trial mode for now.

 

Two questions:

 

Does CUDA acceleration work in Trial mode? So far the new machine is not succeeding at downscaling DSD256 to PCM 24-192 without halting. PCM 24-176 works. But it worked on the old machine too. In hqplayerd.xml, there is the setting: cuda=”1”.

 

Anything else I can do in the meantime to tune the system? Should I disable hyperthreading in the BIOS for example??? Crank up the heat on the physical cores???

[Carousel]

OK. Thank you . That gave me a few things to work on. I am pretty sure that I managed to install or update and to activate to <nvidia-driver-535-server>. Though whenever I enter "nvidia-smi -L" I get an error message: "NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running". At one point I was asked to enter a password to use for secure boot. But when I reboot the PC, there is no request for a password. 

 

I was able to confirm this info about the nvidia.ko at least: 

filename:       /usr/lib/modules/5.15.0-89-lowlatency/updates/dkms/nvidia.ko

firmware:       nvidia/535.129.03/gsp_tu10x.bin

firmware:       nvidia/535.129.03/gsp_ga10x.bin

alias:          char-major-195-*

version:        535.129.03

supported:      external

license:        NVIDIA

srcversion:     3A52EBD98F6E1FB4A4E944F

alias:          pci:v000010DEd*sv*sd*bc06sc80i00*

alias:          pci:v000010DEd*sv*sd*bc03sc02i00*

alias:          pci:v000010DEd*sv*sd*bc03sc00i00*

depends:        drm

retpoline:      Y

name:           nvidia

vermagic:       5.15.0-89-lowlatency SMP preempt mod_unload modversions

sig_id:         PKCS#7

signer:         hqpserver Secure Boot Module Signature key

 

ANY MORE HINTS?!

 

PS: I looked at the following, but have not tried the "purge" options yet: NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running - Graphics / Linux / Linux - NVIDIA Developer Forums

[Carousel]

@el70 Just did some testing with a 14700K and an 4060TI 8GB GPU. No problems with sinc-medium from 24-192 to DSD256. But sinc long glitches. The GPU screenshot shows 100% utilization with sinc long. But 50-70% with sinc medium. Interestingly there is more CPU usage showing on htop with sinc medium. Maybe because it is not glitching the GPU?!? What lesson learned??? Try a 12GB GPU??? But so few of the 8188MiB seem to be used??? Experts please chime in!

 

 

 

 

 

 

 

 

[Carousel]

The anecdotal results from testing combinations of upscaling or downscaling and HQPlayer settings is that a 7950 (my neighbour has) will out do a 14700 (I have). You already have a lot of computational power there. Is there anything that you want to run that does not run for you? There are some combinations that cannot be solved with more power. For example, if a filter like closed-form fundamentally cannot convert between rate families. 

[Carousel]

I was wondering if I would be allowed to post GPU questions on the CPU thread! If the boss can, then it must be OK! 

 

Unfortunately the prices for the present 4070 Ti GPUs have not yet been slashed for closeout sales. I bought an 8GB 4060 Ti during Black Friday which I can return until 30th January as part of Christmas sales. The CPU is a 14700K. I do not go beyond DSD256. But I do up to eight channels of this (picture below). Basically everything plays fine with poly-sinc-gauss and the 4060 Ti. Also fine is when I downscale DSD256 stereo to PCM 24-192 for room correction softwares. The 4060 Ti won't do sinc-long. But sinc-LI does work in stereo. But then I keep bumping into 44K vs 48K no-playback-at-all issues and I am not always sure if a more powerful GPU would help at all. Whatever I am trying just might not work with the combinations I selected.

 

Still, the clock is ticking for me to return the 4060 Ti on the basis of insufficient performance. Looking at specs on 4070 GPUs and 4070 Ti GPUs, it looks like each step up from the 4060 Ti is pretty substantial 

+ 4060 Ti : 8GB and 4352 CUDA Cores

+ 4070 : 12GB and 5888 CUDA Cores -- and the price step up is fine

+ 4070 Ti : 12GB and 7680 CUDA Cores -- but the price doubles vs the 4060 Ti

 

Does anyone have a view / experience here? Is the step up to the 4070 enough to make a difference? Or is it just a waste of effort and really only the 4070 Ti would have a chance with filters like sinc-long going to or from DSD256?

 

 

 

[Carousel]

@Miska Is it technically possible to upscale a PCM 24-48 source file to DSD256 using sinc filters like sinc-M or sinc-Mx? My set-up can do a 16-44 source file, but not a 24-48 source file. If it should be doable, then that could be an argument for my upgrading from an 8GB 4060 Ti to a 12GB 4070 Ti. If it is not doable, then there is no point. Note: I was using ASDM7ECv3. But DSD5 and 7 did not work either. My CPU is a 14700K. Thanks.

[Carousel]

Aha! Thank you both. I knew that the 2^x filters like closed-form were a problem here. And I had looked in the manual. Forty-eight is an integer. But, yeah, not an integer ratio multiple of 44.1! I am with you now. With your help I am getting closer. The 14700K plus the 4060 Ti seems to be working for me in stereo, for what I want to do with my stereo DAC. I am just moving on to my surround sound DAC now. The exaSound S88 is ESS9038PRO based. Just trying now, ticking 48K DAC in HQP does not seem to make it into one though! Using poly-sinc filters for 8-channels is not a problem into DSD256 though. I guess I can stick with what I have. A sad day for an electronics shopping addict. 

[Carousel]

Seems to be a no-go for 48K on the S88. Glad to to take any exaSound expert pointers though. Maybe I just maxed some other parameter in the mix...

 

Back on 44.1 DSD only, I just found the surround limit of my 14700K + 8GB 4060 Ti system. Upscaling 5-channel 24-96 to DSD256 using sinc-medium is bouncing around 95% of GPU utilization with occasional glitching. Not sure this justifies trading up to a 4070 Ti though...

[Carousel]

@blackhawk579  It depends a bit on which filters you want to use and just how much you want it to get beyond DSD256 to DSD512+. With that CPU, you should be able to do A LOT. With a 14700K I do not need a GPU to do what I want with the poly-sinc-gauss or ext filters up to DSD256. For sinc-long, a GPU is needed though. And with that PC case, it looks like adding a GPU later would not package with great cooling if at all. 

 

A key benefit of HQPlayer is to put the server machine elsewhere, so silent PCs are not so needed. One does need an NAA endpoint at the destination though. But it can be optimized as an endpoint too.

 

Myself, I did not want a big PC case. I looked at the silent PC designs. But wanted to protect for using a GPU. I opted in the end for a Cooler Master NR200P MAX with the PSU and the cooler already built in. I actually do have it in my listening room... behind a cabinet... and have not heard it.

[Carousel]

The NAA end point should be on before the HQPlayer server is started. So if you leave your Red on, you can turn the server on and off as you will. 

[Carousel]

@blackhawk579 As the other say, how high you might want to upscale is a matter of preference. I tried up to DSD512 for stereo, but fell back to DSD256 happy. Part of this is that my 8-channel DAC only does up to DSD256 for surround and I find it simpler to keep all DSD output at DSD256 (and all PCM output when not upscaling at 24-96, for which various dSP systems are optimized and not higher). 

[Carousel]

9 hours ago, blackhawk579 said:

ASUS RTX 4060 Ti DUAL OC Black 8GB

 

I returned my 8GB 4060 Ti last week. Am hoping to get an original 4070 Ti on sale or as a warehouse return deal as the new "Super" ones are coming out.  Either that or I will get a new 4070 Super.

 

The 4060 Ti can handle just about anything except the long sinc filters. It is not so much the GB that were too few. So I am not upgrading to a 16GB version. It was the "Volatile GPU-Utilization" from the Nvidia SMI report that went through the roof. Not sure, maybe that is driven more by CUDA cores, clock speeds, bus sizes. Expert comments welcome.

 

To be honest, though, the 4060 Ti was fine for my needs. I do not really need the shinier toy. Nevertheless...

 

[Carousel]

Here is something similar but with a 14700K and no GPU. It is a water-cooled CPU. HQP Embedded is 5.4.0.

 

sinc-long is non-apodizing though. Is that not a worry? The apodizing score is listed as 735. The source is a FLAC 16-44 file on the sda1 SSD drive on the same Ubuntu Server machine running HQPlayer Embedded 5.4.0

 

 

 

 

[Carousel]

I do like sinc-L and that it seems easier for my kit to use than sinc-long. The only problem I had with it though is that whenever a 48-based track came up... instead of a 44.1 multiple... playback to DSD256 stopped for me as sinc-L needs an integer multiple for the destination playback samplerate.

 

Am curious though, as queried earlier today, that @AudioDoctor is using the non-apodizing sinc-long for 1x upscaling. Based on the HQP manual guide, I avoiding using non-apodizing filters for playback of lower sample-rate material (i.e., for most playback).

[Carousel]

@AudioDoctor @Miska Reads like I should replace my Duke Ellington & John Coltrane. I probably have, at the newest, a 2007 CD in a box in the cellar. Looks like I only updated the album art in 2017 to a bigger picture than I had originally. Checking Discogs, I could look for a 2018 Stereo & Mono release, or a 2020 or a 2023. Don't know if you would have recommendations from there: Duke Ellington & John Coltrane - Duke Ellington & John Coltrane | Releases | Discogs. I might just try a couple Qobuz downloads and see how they measure. Maybe I start with the 24-192!

 

As for my Babs & Duke, that is a 2016 CD rip... done in 2017. The apodizing count after 3 minutes of Stoona is only 7. And it sounds pretty good with sinc-long. So I take it, this one is OK as it is even with a non-apodizing filter. Right?

 

[Carousel]

@AudioDoctor The RGM Serenade to Sweden CD was one of my Dad's last CD buys in 2017 before he passed. So it is indeed a special one for me. 

[Carousel]

PS: just saw the Duke Ellington & John Coltrane link you posted as well above. Good price for DSD256!

[Carousel]

Well, my progeny made the mistake of letting their new 4070 Ti be delivered to my hands. It is not a Super, but they got a very good runout deal. Now I can testify that the 4070 Ti will run sinc-long that my 4060 Ti would not. This is with a 14700K. I know you DSD1024 stereo folks won’t be impressed here. But I am happy with the below DSD512 stereo results with sinc-long and ASDM7ECv3. Trying eight-channel sinc-long to DSD256 did not play though. The eight-channel charts below are for sinc-medium to DSD256 with ASDM7ECv3. All again with a 14700K.

 

Based on this, I would expect similar results with a 4070 Super.

 

Question for the experts: To be able to manage sinc-long in eight channels to DSD256, would more GPU help? Or would I need more CPU?

 

Looks like more CPU to me. But maybe there are more ways to offload to the GPU???

 

Sinc-Long ASDM7ECv3 16-44 Stereo to DSD512 Nvidia 4070 Ti

 

Sinc-Long ASDM7ECv3 16-44 Stereo to DSD512 14700K htop

 

Sinc-Medium ASDM7ECv3 24-48 8-Channel to DSD256 Nvidia 4070 Ti

 

Sinc-Medium ASDM7ECv3 24-48 8-Channel to DSD256 14700K htop

 

[Carousel]

Beauty!!! But whenever I brought home a math exam result of 100%, my mother always asked why I did not get bonus points for 105%! ;-)

[Carousel]

@Louie I can confirm that a 4060Ti cannot handle, for example, sinc-long. And it is NOT the GPU RAM that runs out. I borrowed a 4070 Ti with less GPU RAM (12 vs 16) and it worked. The new 4070 Super is basically a 4070 Ti and has a great price. I am not sure that the original 4070 would do. If you have more budget, but not silly amounts of budget, then a 4070 Ti Super probably would be great. Is more like an original 4080. I am saving up for one!

[Carousel]

The 4070 Super is in between at 220W. But, yes, is 12GB not 16GB. Make sure with the experts on this forum if the extra RAM is needed for what you want to do. And/or make sure you can return the 4060 Ti if it does not suffice for you. I returned mine for my use case.