PC+SW+DSD direct VS Any DAC

[phofman]

Thanks. That's 0.135 * 14 = 189% of one core, i.e. almost two full powerful amd64 cores at 4GHz just for upsampling 2ch audio, IIUC.

Just for comparison with CPU load of pure PCM processing: Resampling 16(!) channels PCM 48kHz -> 384kHz at float32 precision (using camillaDSP) takes 82% of one 1.3GHz core of Radxa Pi S (tiny passive 4-core ARM SBC powered by its USB input, running cold). Resampling at float32 precision puts noise floor at -150dB (the spectrogram captures just some of the channels, the chain is: generator on PC USB host 48kHz/32bit/16ch -> Pi S USB device -> resampling to 384kHz -> I2S out @384/32/16ch -> I2S in @384/32/10ch -> Pi S USB device -> spectrogram on PC USB host 384/32/only 5ch due to UAC2 bandwidth limit)

Resampling 16ch at float64 goes only up to 192kHz on that weak CPU (because the resampling of all channels runs in one thread only), noise floor below -180dB:

 

[widemediaphotography]

Just as an example of DSD upsampling/filtering "efficiency" in HQPlayer

Someone trying to make HQPlayer work with DSD512 with built-in filters a few months ago:

View attachment 379772

Turns out Titan V 12GB may not be enough, according to Miska, may need even more GPU memory:

View attachment 379773


Lets see... Titan V 12GB GPU - $1500, idle power consumption 95W. Consumption under load - 345W

I tried, that setup is non compatible on V5 or my setup. Its impossible start the playback

 

[pkane]

I tried, that setup is non compatible on V5 or my setup. Its impossible start the playback

<s> You obviously need to upgrade your setup! You’re missing out on the best-sounding filter… according to some That $1500 GPU, a larger PC and an upgraded version of HQPlayer is all you need to achieve audio nirvana! </s>

 

[widemediaphotography]

<s> You obviously need to upgrade your setup! You’re missing out on the best-sounding filter… according to some That $1500 GPU, a larger PC and an upgraded version of HQPlayer is all you need to achieve audio nirvana! </s>

My Nvidia 3080 RTX card already has 16Gbyte DDR5 Ram. That setup is Simply no compatibile and I won't upgrade nothing

 

[ajar]

Rather than argue about your definition of efficiency, the real question is what do you gain with all this extra processing? Where’s the evidence that this is really needed for listening to audio?

Thank you! Stop the arguing about the definition of efficiency. The thread is about getting better results.