Jump to content

Q. Upsampling Computing Power


Recommended Posts

Q. Is upsampling done with the computer processor, or with the Ram? Audirvana, Macbook Air 1.6ghz i5 with 8gb Ram will not go the distance to DSD512, it does DSD64, but distorts and fails to do DSD512.

Processor or Ram or both perhaps? Cheers

 

 

Link to comment
Share on other sites



14 hours ago, Way60 said:

Q. Is upsampling done with the computer processor, or with the Ram? Audirvana, Macbook Air 1.6ghz i5 with 8gb Ram will not go the distance to DSD512, it does DSD64, but distorts and fails to do DSD512.

Processor or Ram or both perhaps? Cheers

 

 

Mac only support upto DOP DSD 256, unless you add an endpoint e.g. UltraRendu or SOtM ... my experience .. RAM minimum is 16GB, and for DSD 512 you need a Mac computer with Benchmark above 4000 (https://browser.geekbench.com/mac-benchmarks)

  • Like 1
Link to comment
Share on other sites


I've got a Mac Mini 2012 with i7 processor and 16GB of RAM. I've allocated 8GB of RAM as a memory cache.

 

I use Audirvana and have experimented with up-sampling to DSD 128 (the highest my DAC can accept). The processor can keep up, but the fan starts to spin hard and the noise becomes intrusive.

 

If you want to experiment to find out if up-sampling to higher DSD formats makes any difference, I suggest you use local audio files you're familiar with. You can pre-convert a copy of the files to avoid the real time processing hit and should be able to do the comparison.

  • Like 2
Link to comment
Share on other sites

5 hours ago, davm said:

I've got a Mac Mini 2012 with i7 processor and 16GB of RAM. I've allocated 8GB of RAM as a memory cache.

 

I use Audirvana and have experimented with up-sampling to DSD 128 (the highest my DAC can accept). The processor can keep up, but the fan starts to spin hard and the noise becomes intrusive.

 

If you want to experiment to find out if up-sampling to higher DSD formats makes any difference, I suggest you use local audio files you're familiar with. You can pre-convert a copy of the files to avoid the real time processing hit and should be able to do the comparison.

Thanks davm, I have experimented,  [it plays upsampled to dsd64 no worries and does play dsd512 for a while] and I think it sounds good in my set up. How do I convert files before hand, I only use local files on my Mac ;  no streaming etc. Cheers

Link to comment
Share on other sites


1 hour ago, Way60 said:

Thanks davm, I have experimented,  [it plays upsampled to dsd64 no worries and does play dsd512 for a while] and I think it sounds good in my set up. How do I convert files before hand, I only use local files on my Mac ;  no streaming etc. Cheers

 

I convert audio files on my Mac using an app called XLD. It can convert to DSD format, although the limit on my Mac is DSD256.

 

Start XLD and use the Preferences dialog to choose the output format and the Option button to choose which DSD format for the conversion. Then just use drag and drop to drop the folder with the source files onto the XLD icon in the dock.

Edited by davm
spelling
Link to comment
Share on other sites

I've got an i9-11900K upsampling all music to DSD512 'on the fly' using HQPlayer.

 

Cost me about $1600 for the PC build.

 

It's got a big and quiet Noctua fan for cooling and a quiet Fractal Design Define R6 case.

 

Very capable and quiet but it's nowhere near any listening rooms anyway, so the very quiet Noctua fan noise is not a factor at all.

 

Listening areas connect to it over Cat 6 UTP cable.

 

Here is an example of CPU loading (8 physical cores, 8 virtual cores)

 

image.png.d35d58df9228b675f439c29a45b31f59.png

 

Edited by rand129678
Added pic
  • Like 1
Link to comment
Share on other sites

10 hours ago, rand129678 said:

I've got an i9-11900K upsampling all music to DSD512 'on the fly' using HQPlayer.

 

Cost me about $1600 for the PC build.

 

It's got a big and quiet Noctua fan for cooling and a quiet Fractal Design Define R6 case.

 

Very capable and quiet but it's nowhere near any listening rooms anyway, so the very quiet Noctua fan noise is not a factor at all.

 

Listening areas connect to it over Cat 6 UTP cable.

 

Here is an example of CPU loading (8 physical cores, 8 virtual cores)

 

image.png.d35d58df9228b675f439c29a45b31f59.png

 

It's interesting to me that all the cores seem to quite busy.

 

I've never taken the time to investigate if up-converting files like this can be spread across multiple cores, but I've always assumed the process would be single threaded. Is it possible the CPU loading is due to multiple tracks being converted simultaneously, rather than all those cores working on one conversion?

  • Like 1
Link to comment
Share on other sites

2 hours ago, davm said:

It's interesting to me that all the cores seem to quite busy.

 

I've never taken the time to investigate if up-converting files like this can be spread across multiple cores, but I've always assumed the process would be single threaded. Is it possible the CPU loading is due to multiple tracks being converted simultaneously, rather than all those cores working on one conversion?

 

Good question.

 

In the HQPlayer world (I can't talk about other software) the modulator has always required high single core speed.

 

So the long belief that HQPlayer needed the fastest single core CPU's has mostly been due to the modulator needing this speed. 

 

Stereo tracks need at least 2 physical cores with high single core speed.

 

I feed HQPlayer from Qobuz,  Apple Music Hi-Res and Deezer and Soundcloud and other digital sources. 

 

I have HQPlayer handling my convolution and DSP crossovers for my DIY DSP speakers (8-channels) applying it to any source I want.

 

All the below is kind of simplified to make it kind of easier to follow.

 

Say I'm playing a 24/96kHz track on Apple Music, I feed this into HQPlayer which then will convert this to ~80bit/96kHz because convolution for digital room correction and DSP crossovers is done by HQPlayer at the source rate before upsampling.

 

Then this is converted to ~80-bit/24MHz (DSD512). This upsampling task is what needs more cores and core speed is less important. Some of HQPlayer filters even benefit from offloading to the best GPU's on the market.  Especially for those upsampling DSD multichannel tracks (which I don't).

 

Once it is at ~80bit/24MHz (DSD512) then it is the modulator that converts it 1-bit and this is where you need at least 2 very fast CPU cores. This task needs faster speed than GPU's can offer.

 

But in the past 12 months HQPlayer's dev has done some really major optimisations for the latest Intel and AMD CPU's that have >8 physical cores that are also all fast now.

 

This was a game changer of an update because previously I would only see 2 cores nearly maxed out and the rest of the cores doing little and some of the things I was doing needed fast GPU help (CUDA offloading).

 

So now you can see the convolution, upsampling and modulator loading are really nicely distributed and no GPU help (CUDA offloading) is required for me now. It is all CPU. Which keeps the price of the machine reasonable.

 

There are still some HQPlayer features that need the best NVidia GPU on the market to help but I don't need those fortunately. 

 

 

Edited by rand129678
Typo
  • Like 2
Link to comment
Share on other sites

21 hours ago, davm said:

 

I convert audio files on my Mac using an app called XLD. It can convert to DSD format, although the limit on my Mac is DSD256.

 

Start XLD and use the Preferences dialog to choose the output format and the Option button to choose which DSD format for the conversion. Then just use drag and drop to drop the folder with the source files onto the XLD icon in the dock.

Cheers, I will look into it

Link to comment
Share on other sites

I've found my Synology NAS DS1621+ does just fine upsampling to native DSD512 for Roon. Problem is that DSD only seems an improve playback with certain DAC's. Overall the original recording and mastering quality seems far more important.

Link to comment
Share on other sites

37 minutes ago, MattyW said:

Problem is that DSD only seems an improve playback with certain DAC's.

Definitely.

 

I have a couple DACs that directly convert 1-bit SDM to analogue.

 

One example where deliberately upsampling PCM to DSD maybe doesn't make sense: if you feed DSD to Rob Watts Chord DACs, he has his FPGA very deliberately convert (decimate) DSD back to PCM anyway. 

 

Just one example but there are more.

 

Edited by rand129678
typo
Link to comment
Share on other sites

34 minutes ago, rand129678 said:

Definitely.

 

I have a couple DACs that directly convert 1-bit SDM to analogue.

 

One example where deliberately upsampling PCM to DSD maybe doesn't make sense: if you feed DSD to Rob Watts Chord DACs, he has his FPGA very deliberately convert (decimate) DSD back to PCM anyway. 

 

Just one example but there are more.

 

 

Also the Rockna Wavedream. It runs PCM internally up-sampled by 16x. The Denafrips and Musician DAC's also see no benefit from DSD. Upsampled content sounds identical to my ears..... AKM based DAC's on the other hand definitely do benefit as they run native DSD internally though the sonic benefits (more natural, liquid sound) are along the lines of how discrete R2R and multibit DAC's sound with all content anyway so kind of pointless.

 

Anyway, sorry for going a bit off topic there. I'll refrain from commenting further.

  • Like 1
Link to comment
Share on other sites


 Share

  • Recently Browsing   0 members

    • No registered users viewing this page.




×
×
  • Create New...
To Top