The issue I face is using the same hardware booting 3 OSes. Windows and Ubuntu play nice, but macOS does not. (Well, Windows plays, it is very rarely a nice OS.)
Since Apple and nVidia got into there little thing, there are no new graphics drivers for macOS.
There is limited support with AMD, so AMD is kind of a prerequisite with anything above 10th intel chips as there are only iGPU drivers for 10th gen and below. Which ideally puts you at the z490 motherboard era.
So you can imagine the fuss using a z790 with 13th gen chip, no iGPU support, no nVidia support, and limited AMD (which is also os version specific depending on the card).
Itās hard to find a case/board thatāll accept 2 graphics cards, a 6600xt and a 3080ti, and a flashed thunderbolt card. So I have to swap the gpu.
I considered/still considering an eGPU enclosure but not really cost effective.
Hello, @jussi_laako . Iām wondering if you can help me offload just a bit more of the convolution to my graphics card, if such a thing is possible.
Setup:
HQP Desktop 4.21.1
i9-10900K processor
32GB RAM
EVGA RTX 2080Ti video card
Windows Server 2019
Audiolense XO convolution filters at 192K in pipeline matrix for 5.1.2 MCH/Atmos speaker array
Roon feeds HQP
HQP Settings:
8 channels
SDM (DSD) default output mode
PCM and SDM defaults all poly-sinc-ext2
ASDM7ECv2 modulator
384K limit for PCM
44.1x64 limit for SDM
My issue is dropouts every 10 seconds or so with 5.1 DSD64 material. 2 channel DSD64, FLAC 5.1 and lossless Atmos 5.1.2 (24/48) material plays fine.
Of course, with no offload to CUDA, the processor alone just canāt handle the convolution:
CPU load looks fine, but likely the GPU is having a bit too hard time.
Windows Resource Monitor displays GPU load incorrectly when CUDA is used. Please open terminal and run ānvidia-smi -l 5ā to see overall situation. On Windows, nvidia-smi is not as useful though, but gives an idea anyway.
OK, given that PCM sources and stereo DSD sources play fine, the next suspect is SDM Integrator and SDM Conversion settings in DSD Source Settings dialog. What do you have for these two?
Ok, @jussi_laako , I have tried that. Dropouts are now about 20-30 seconds apart, instead of 10-15 seconds apart. So, getting closer. These are short dropouts. Other suggestions? JCR
I will check again with greyed vs checked. Since this server is also my Roon core, I canāt dedicate it to just running HQP alone. What about increasing buffer size?
If I do seek more CPU, which will entail a motherboard change as well because of the different sockets for the newer processors, what would you suggest if I want to upscale to DSD128 with convolved DSD64 MCH files and have no dropouts? JCR
Ok, so forum folks, if you are successfully using Roon to feed HQP for DSD5.1 MCH files upscaled to DSD64 after convolution, Iām interested in your hardware and settings.
In any case in light of your comment, @jussi_laako , I will not make any hardware changes. But with 856 DSD5.0 and 5.1 MCH albums, I am hoping for your upcoming things to up and come soon! Thanks. JCR
Definitely wait as Jussi said, but @Chunhao_Lee 's system will do what you need - and then some
The last update I saw was he is outputting 12 channels of DSD256 (Merging Hapi Mk2 and Merging Anubis from memory) for Dolby Atmos contentā¦ But canāt recall his PC details.
If you wanted a system with headroom for 5.1 channels, copying his PC would do it
Hi @dabassgoesboomboom . I can do eight channels of convolved, lossless Dolby Atmos (5.1.2 at 24/48) upscaled to DSD64 and lossy Atmos to DSD128, the limit of my Okto DAC. As I understand, it is actually more processor intensive to downconvert 5.1 DSD64 to convolve at ā in my case ā PCM 192 ā and then reconvert back up to DSD, than to play Atmos files, which donāt require the DSD-to-PCM conversion for convolution before upscaling to DSD. @Chunhao_Lee definitely has a bigger, badder, newer processor than mine. JCR
HQPlayer can DSP DSD contents natively without transfer to PCM. A decent GPU with sufficient VRAM is required. In my experience 5.1DSD64 ā DRC pipelines ā 5.1DSD256 the 12GB RTX3080ti is on par. Just FYI.
@jussi_laako does the adaptive loudness control have a negative impact on sound quality because itās post-processing? vs say a fixed house curve peq in each channel earlier in the pipeline?
I am surprised that this feature does not get used more often or discussed given the seemingly obvious benefit with a standard 2 channel speaker system and digital volume control. Correct me if I am wrong but I recall that you donāt really like it and only added due to a request? If so, how come?
That would be fun, @Chunhao_Lee ! I canāt complain about 5.1.2 Atmos even at DSD64.
Ok, @jussi_laako , I can report that greyed vs checked CUDA does not make a difference. However, I changed the modulator to ASDM7EC (v1) from v2. The track I just listened to had but two dropouts over 4 minutes, so getting closer.
What is the practical sonic difference between v1 and v2 of the ASDM7EC modulator? JCR