Design a PC/Server for ROON and HQ Player

[jabbr]

1 hour ago, Miska said:

 

I mean 44.1k source to 44.1k x512 with poly-sinc-xtr, so we can compare it with the other figures for 2080...

 

Was the first one for that case?

Yes DSD64 source -> 10-11%

PCM44,PCM96 -> 44-52%

[Miska]

1 hour ago, jabbr said:

and I’m getting some stuttering with HD PCM sources ... I need to do some work to optimize unless @Miska is able to shed some light...

 

Can you tell the exact case/settings? Going from either 44.1k or 48k base rates to 44.1k base output rate should be fine.  Going from 44.1k-base to 48k-base DSD output rate ends up being crazy heavy. So 48k-base to 44.1k-base is much easier than vice versa. Within the same base rate it is also easy.

 

1 hour ago, jabbr said:

(note that I’m using generic 4.18.0 kernel with Ubuntu 18.04.02 — yes I’ll be maximally compliant to reduce variables

 

Ubuntu repos have also 4.18 lowlatency kernel, maybe you'd want to switch to that one for audio. It is more desktop/multimedia optimized one while generic is more for server cases.

 

[Miska]

1 minute ago, jabbr said:

Yes DSD64 source -> 10-11%

PCM44,PCM96 -> 44-52%

 

OK, so the Ti delivers pretty much the same performance ratio as the price ratio. About 1.5x price, about 2/3rds load. That is good since it means pretty exact bang for the buck.

 

[jabbr]

3 hours ago, Miska said:

 

OK, so the Ti delivers pretty much the same performance ratio as the price ratio. About 1.5x price, about 2/3rds load. That is good since it means pretty exact bang for the buck.

 

 

Wow! When I upconvert 24/96 or 24/192 -> DSD512 (22.6Mhz) as opposed to to 24.6Mhz the GPU utilization drops from 50% or so down to 25-30% and the hiccups stop ... and in fact much much better than 44.1kHz -> which uses 45-50% (?)

 

So folks, this is why I'm not giving listening impressions right off the bat...

[Jud]

4 hours ago, Miska said:

Ubuntu repos have also 4.18 lowlatency kernel, maybe you'd want to switch to that one for audio. It is more desktop/multimedia optimized one while generic is more for server cases.

 

I'm using the low latency kernel with the Lubuntu variant (which I think may be just a bit lighter weight, unless you're using Ubuntu Server with Embedded), and my experience has been positive.

[jabbr]

40 minutes ago, Jud said:

 

I'm using the low latency kernel with the Lubuntu variant (which I think may be just a bit lighter weight, unless you're using Ubuntu Server with Embedded), and my experience has been positive.

 

I use Ubuntu server with xorg— then I ssh in:

 

$ ssh -Y [email protected]

 

any GUI apps I occasionally need pop up on my laptop

 

The “fix” was to upsample not in the same family but to 22.6Mhz — I don’t understand why but @Miska said so  

 

Now optimize everything else so I’m comparing +\- Roon, +\- NAS etc etc and will put low latency kernel in the mix ... but first need to determine where the delays are because if it’s done wrong, things get worse ... the best I can tell, when the GPU is st 25-30% load, no hiccups and then slight at 45-50% ... at the same time CPU goes from 200% to 300% (I have 20 cores so this is minimal) so the key us likely keeping the GPU well fed  

[Miska]

44.1k xN to 22.5792M and 48k xN to  24.576M should end up in practically same load.

 

The notable differences are:

1) 44.1k xN to 24.576M - not going to work

2) 48k xN to 22.5792M - works fine, a bit heavier than the simple cases, but not so much

 

[jabbr]

2 hours ago, Miska said:

44.1k xN to 22.5792M and 48k xN to  24.576M should end up in practically same load.

 

They aren't. Not close.

 

44,88 -> 22.57 use 44-50% (GPU) and 300% CPU

44,88,96,192 -> 24.57 use 44-50%

 

96,192 -> 22.57 use 25% (sounds the best too)

 

Quote

 

The notable differences are:

1) 44.1k xN to 24.576M - not going to work

2) 48k xN to 22.5792M - works fine, a bit heavier than the simple cases, but not so much

 

 

2) seems to use both less CPU and GPU here... I am using 180-200% CPU for this

[Miska]

10 hours ago, jabbr said:

They aren't. Not close.

 

44,88 -> 22.57 use 44-50% (GPU) and 300% CPU

44,88,96,192 -> 24.57 use 44-50%

 

Hmmh, to me these two look very similar?

 

10 hours ago, jabbr said:

2) seems to use both less CPU and GPU here... I am using 180-200% CPU for this

 

That is interesting pattern, I have to check it out too. Running on plain CPU 96k -> 22.5792M takes somewhat more than 96k -> 24.576M:

> time resample-wav /scratch/personal/jussi/audio-tests/tones-int32/96/Swept_int.wav /tmp/tmp.wav 22579200 --xtr
67.512u 0.526s 0:42.21 161.1%	0+0k 6064+1411216io 1pf+0w

> time resample-wav /scratch/personal/jussi/audio-tests/tones-int32/96/Swept_int.wav /tmp/tmp.wav 24576000 --xtr
43.904u 0.513s 0:38.66 114.8%	0+0k 8+1536016io 0pf+0w

 

[jabbr]

For reference with cuda="0"

------ 22.57 MHz ----

44.1k  -> 1200% cpu

88.2k -> 1200%

96k -> 1200%

192k -> 1300%

DSD64 -> 380%

------ 24.57 Mhz -----

44.1k -> 3900%

88.2k -> 3900%

96k -> 1345%

192k -> 1100%

DSD64 -> 450%

 

There may be an issues in the settings I applied with GPU so I will repeat

 

[jabbr]

96 and 192k both use 50% GPU

 

I reported this incorrectly yesterday. I had "auto rate family" checked and was setting 22.57 as max rate, but it was dropping down to 12Mhz. When I set the rate to 22.57 and remeasure (properly) it uses 50%.

[jabbr]

Ok so I'm going to say that the E5-2470 v2 is not the best CPU for Hqplayer because it has a base clock of 2.5 Ghz and HQPlaer likes a higher clock.

[Miska]

10 minutes ago, jabbr said:

Ok so I'm going to say that the E5-2470 v2 is not the best CPU for Hqplayer because it has a base clock of 2.5 Ghz and HQPlaer likes a higher clock.

 

As usual, it is combination of all things, what you want to what output. With, or without CUDA. And not just base frequency, but amount of cache, RAM speed, support for AVX2, AVX-512, FMA. Unfortunately so many variables that it is hard to predict how some particular combination will fare without testing. 3 GHz on Intel core is not same as 3 GHz on AMD core, etc. 2 GHz ARM core can barely do DSD64...

 

And some things are just magic black box of architecture generation. I don't have a clear answer why RTX 2080 is so much faster than GTX 1080 with HQPlayer poly-sinc-* work load. Nvidia doesn't shed much light on details. No ray-tracing or AI networks here. But at Volta step they magically improved something.

 

But so far, clock-for-clock, new architectures have been improving.

 

[improvedsound]

I built a Music Server based on:

Intel Xeon processor (Coffee Lake) with TDP 45W, clock 2.9 GHz, 6 cores 

ECC DDR4-2666  

Intel Optane SSD 800P Series 58GB M.2 with only Windows 10 LTSB,

HQplayer resident on RAM (Primo ramdisk)

Separate linear power supply for motherboard, processor, 8TB HDD

 

44.1> DSD 256 poly-sinc-xtr the CPU works at 25/38%
44.1> DSD 256 poly-sinc-2s the CPU works at 15/18%

 

My DAC: Merging Hapi 

[improvedsound]

here some pictures of my creature

 

[PieterP]

Diy chassis is cool anyway! Congrats and happy listening. Picture of the front is missing? Where did you get the heat pipes and copper plate?

[sig8]

I just finished building a HQPe server with i9-9900k, and a Gigabyte Z390 Aorus Pro motherboard. I am using Nocuta NH-D15 fan assembly for cpu cooling. It runs at about 35-40C doing DSD 64>512, and the temperature of the cpu does not rise much even if I disconnect the fans from the mobo, because the heatsink for NH-D15 is huge and that itself provides lots of cooling. Initial testing is with Closed Form-M and Closed Form 16M; I think these are the heaviest filters. There are no other fans because my assembly is an open case assembly using an HDPlex H5 passive case to hold things together.

[improvedsound]

On 3/29/2019 at 10:22 PM, PieterP said:

Diy chassis is cool anyway! Congrats and happy listening. Picture of the front is missing? Where did you get the heat pipes and copper plate?

 I bought the heat pipes in a German web store.  I'm sorry, but I can't  find it, no more, on the internet

 

here are some pictures about the front.

 

 

[asdf1000]

On 3/31/2019 at 5:46 AM, sig8 said:

I just finished building a HQPe server with i9-9900k

 

Hi, great build.

 

Can this CPU do PCM44.1 to DSD512x48 with poly-sinc-xtr (non 2s) fine?

 

If your DAC can take DSD512x48 (24.576 MHz) sample rate.

[improvedsound]

42 minutes ago, improvedsound said:

 

 

 

[sig8]

18 hours ago, Em2016 said:

 

Hi, great build.

 

Can this CPU do PCM44.1 to DSD512x48 with poly-sinc-xtr (non 2s) fine?

 

If your DAC can take DSD512x48 (24.576 MHz) sample rate.

My Holo Spring 3 does not do DSD 512x48, but this CPU easily does 44.1PCM>DSD512, with poly-sinc-ext.

[asdf1000]

5 minutes ago, sig8 said:

My Holo Spring 3 does not do DSD 512x48, but this CPU easily does 44.1PCM>DSD512, with poly-sinc-ext.

 

Noted.

 

I believe ext always runs in -2s mode though? @Miska

[Miska]

6 hours ago, Em2016 said:

I believe ext always runs in -2s mode though? @Miska

 

ext is single stage. ext2, for SDM outputs it is two stage with twice higher intermediate rate than -2s versions. Same as with -2s, if the ratio is small enough, then it is single stage. For example 192k source to 44.1x64 or 44.1x128 is single stage. For PCM, ext2 is always single stage. The first stage is always there, but the second stage may be dropped out (opposite to what DAC chips do).

 

ext and ext2 are totally different filters.

[asdf1000]

23 minutes ago, Miska said:

ext and ext2 are totally different filters.

 

Thanks, good thing I tagged you to double check. I don't like spreading fake news  😁

[twluke]

Hi there,

 

I recently built an i7-9700K system using a Z390 Phantom Gaming-ITX/ac MoBo, 16GB DRAM from Corsair, an SFX PSU (SX600-G from SilverStone), a Noctua cooler (NH-U9S) and an M1 case from SFFLAB for the HQP on a linux kernel (Ubuntu 18.04), focusing on playing upsampled DSD512 sources via Roon.

 

The results were quite good and there was no untoward rising of CPU temperature when playing DSD512 with a poly-sinc filter, usually within the ranges of 50 to 65°C as shown in the first picture below. Even with a poly-sinc-xtr filter, the system was well working without any stutters or signal drops in playing DSD512, permitting a rise of CPU temperature up to the ranges of 70 to 80°C around as shown in the second picture. I think this good result is considerably owing to the excellent cooling power of the Noctua cooler. It is almost quiet or nearly silent with a poly-sinc filter. Even with the xtr filter, the fan is very quiet and you can focus on listening to music without being distracted. The M1 case from SFFLAB was very compact but can deal with the size of RTX2080x GPUs. The interior of the case was well designed, allowing smooth air flow from the CPU fan for heat dispersion.

 

Finally what should be stressed here is the current system does not include any GPU, i.e., no cuda offload, yet capable of playing DSD512 with the xtr filter without any problems. Previously I reported a positive result of RTX2080 GPU for dealing with the xtr filter with a 8700K system but the current system gave me another positive result. Well, that said, I'm now back to the poly-sinc filter for listening.

 

Regards,