Design a PC/Server for ROON and HQ Player

[jabbr]

Seems to be for software defined radios mostly. Did you check how much those cost? Last time I checked possibly capable enough Xilinx Virtex-series FPGA costed $1k+. So if you are in market for $100k DACs, then that could be the way to go.

 

SFDR figures from the first PDF are not very promising for audio...

Yah. I've been playing around with the "regular" zynq's (e.g. picozed/snickerdoodle/parallela) which are more reasonably priced. Aiming at hitting DSD1024 x 32 channels direct DSD output from a 1Gbe input (that's essentially saturating the 1GbE link).

 

In any case these have the more pedestrian 1Msps XADC so perhaps something to fool around with. Looking at: http://www.analog.com/media/en/technical-documentation/technical-articles/not-your-grandfathers-adc-rf-sampling-adcs-offer-advantages-in-systems-design.pdf

Raises question as to whether decimation would improve SFDR enough to be interesting ... just a thought but not something I'm actively looking at (the XADCs are just sitting there on chip unused for me at the moment)

 

Regarding the Zynq, depending on performance, might go with zero-copy between the TCP input buffer and the FIFO sent to the PL, so the real consideration for me is integration with networkaudiod. In any case this is where the SDR techniques have popped up (reading tutorials etc).

[MarcinGD]

I just bought 6700k to replace 6700. Will check how it will perform in streacom fc8. Waiting for shipment. Will try also to downclock to check if I can go lower with temp.

So I did place 6700k into FC8 case. And I'm very surprised because temperature is the same as it was on 6700.

[Miska]

So I did place 6700k into FC8 case. And I'm very surprised because temperature is the same as it was on 6700.

 

At least you save quite a bit of power consumption if you don't utilize the built-in GPU/VPU much while loading the CPU cores.

[Miska]

Today I made upgrade for my fanless HQPlayer Embedded machine, motherboard was changed to ASUS PRIME Z270M-PLUS and CPU to i5-7600T. This was installed in already existing Streacom Alpha 10 chassis.

 

Storage is Intel 600p series M.2 SSD and RAM is Kingston HyperX Fury DDR4-2133.

 

Only three heatpipes would fit with this motherboard, but since the CPU is only 35W TDP, it seems to be enough (because the chassis is rated for 65W TDP).

 

I can report that with this setup, it can achieve DSD512 with -2s filter variants at least when using auto rate family. CPU temperature averaging between 42 and 46 C and total load around 50%. But overall, it is intended to work to just DSD256, so it has some reserve left for convolution and such.

 

[tboooe]

@Miska what's the load when converting from DSD to DSD512?

 

Today I made upgrade for my fanless HQPlayer Embedded machine, motherboard was changed to ASUS PRIME Z270M-PLUS and CPU to i5-7600T. This was installed in already existing Streacom Alpha 10 chassis.

 

Storage is Intel 600p series M.2 SSD and RAM is Kingston HyperX Fury DDR4-2133.

 

Only three heatpipes would fit with this motherboard, but since the CPU is only 35W TDP, it seems to be enough (because the chassis is rated for 65W TDP).

 

I can report that with this setup, it can achieve DSD512 with -2s filter variants at least when using auto rate family. CPU temperature averaging between 42 and 46 C and total load around 50%. But overall, it is intended to work to just DSD256, so it has some reserve left for convolution and such.

 

[ATTACH=CONFIG]33691[/ATTACH]

[Miska]

@Miska what's the load when converting from DSD to DSD512?

 

Total load is average 50%... I don't have more precise info at hand.

[tboooe]

Ok I am just very curious because at least in my system converting from PCM to DSD uses a lot less cpu than going from DSD to DSD512.

 

PCM>>DSD256 35%

DSD<<DSD256 80%

 

Total load is average 50%... I don't have more precise info at hand.

[Miska]

Ok I am just very curious because at least in my system converting from PCM to DSD uses a lot less cpu than going from DSD to DSD512.

 

It depends a bit, I didn't try that case yet with that particular hardware. But so far (on other machines) 44.1k PCM -> DSD512 with -2s filter has been roughly the same load as DSD64 -> DSD512.

 

That 50% average figure is for 44.1k -> DSD512 with poly-sinc-xtr-2s and ASDM7.

 

For the record, here are the temps after couple of hours playing to DSD256 (room is 21.5C):

acpitz-virtual-0
Adapter: Virtual device
temp1:        +27.8°C  (crit = +119.0°C)
temp2:        +29.8°C  (crit = +119.0°C)

coretemp-isa-0000
Adapter: ISA adapter
Physical id 0:  +50.0°C  (high = +74.0°C, crit = +80.0°C)
Core 0:         +49.0°C  (high = +74.0°C, crit = +80.0°C)
Core 1:         +50.0°C  (high = +74.0°C, crit = +80.0°C)
Core 2:         +50.0°C  (high = +74.0°C, crit = +80.0°C)
Core 3:         +50.0°C  (high = +74.0°C, crit = +80.0°C)

[tboooe]

It depends a bit, I didn't try that case yet with that particular hardware. But so far 44.1k PCM -> DSD512 with -2s filter has been roughly the same load as DSD64 -> DSD512.

 

That 50% average figure is for 44.1k -> DSD512 with poly-sinc-xtr-2s and ASDM7.

 

Really? Must be because my CPU is and old i7-3770s for DSD >> DSD512 to be more than twice cpu usage of PCM >> DSD512. I guess its time to upgrade CPU!

[jabbr]

It depends a bit, I didn't try that case yet with that particular hardware. But so far (on other machines) 44.1k PCM -> DSD512 with -2s filter has been roughly the same load as DSD64 -> DSD512.

...

 

That is my experience also. PCM 24/192 -> DSD512 not a problem

 

OTOH DSD256 -> DSD512 chokes my aging system.

[tboooe]

OTOH DSD256 -> DSD512 chokes my aging system.

Thats what I dont understand. I have an i7-3770s cpu. Yes its a few generations old but what about it makes it struggle with DSD>>DSD conversion versus the newer generation cpus?

[Miska]

Thats what I dont understand. I have an i7-3770s cpu. Yes its a few generations old but what about it makes it struggle with DSD>>DSD conversion versus the newer generation cpus?

 

CPU's keep improving a lot all the time, although the difference between Skylake and Kaby Lake is not that big. But in this case it is not necessarily the CPU alone, but part of it could be memory bandwidth the the L2/L3 cache implementation.

 

Although DSD->DSD conversion is not particularly heavy on cache or memory, compared to something like poly-sinc-xtr (non-2s) to DSD rates.

[Quadman]

I am researching to build a bigger badder Roon/HQP DSD512, maybe 1024 machine within a budget, not crazy money. My current set up is a pretty decent 512 machine, i7-6700K, 8 Gb 3000MHz ddr4 ram, geForce 1060 GPU. I can handle all filters in HQP at 512 tho sometimes when I hit play in roon HQP initializes for a long time and then roon crashes. Restarting Roon and HQP usually solves this and I am good. CPU utilization in Task manager and resource monitor is usually in the low 20's and never, other than initialization, over mid 30's. I got very intrigued when Miska posted a screen shot of his CPU utilization with his really big boy 6950x machine playing Dire Straits- Telegraph Road at 2:36 into the song, 2 channels, no convolution, with cuda offload, over in the HQP thread. Task manager showed his utilization at 11% and resource manager showed it at 14%. Here's the link to his screen shot http://www.computeraudiophile.com/f11-software/hq-player-20293/index328.html (post 8185). I played the same song, with the same filters except I played thru Roon which is not a CPU intensive program (.1 to .4% utilization). I varied the CPU speed to see the effect, if any, of CPU speed. In one my CPU was 3.08 GHz and the other was 4.45GHz. Here are screen shots of my utilization at the same point Miska took his (2:36)

 

The utilization between the 2 CPU speeds I consider insignificant and are essentially a push. Miska has stated you need a certain CPU speed to handle dsd512, but he has also stated elsewhere and right above this post that memory bandwidth and L2/L3 cache have a big effect on utilization. The bad boy 6950x processor has 4 memory lanes whereas the 6700 and 7700 has 2 memory lanes, the 6950x has 25MB of L3 cache whereas the 6700 and 7700 has 8MB of L3. Speed in the above screen shots favor the 6700K and memory speed I assume is also faster in the 6700 (3000 vs 2400). I am thinking that memory bandwidth and L3 cache has a bigger effect than we, so far, have given credit to.

 

The new AMD Ryzens look impressive but they only support 2 channels of memory so they are out of consideration. I looked into the very reasonably priced 6800K (about $350) from intel (6 cores/12 threads,4 channels memory, 3.4GHz base frequency, 15MB L3 cache) then I studied some benchmarks on CPUboss and the 6800 does not fair that well against the 6700 and especially the 6950x. I looked further up the line at the 6850K (which differs most from the 6800 in base speed only, 3.6GHz vs 3.4 for the 6800) It fairs much better against the 6950x than the 6800 and beats the 6700 in all tests but single thread. It also retails for around $600 vs $1600 for the bad boy 6950x. So it looks like I will head down the more cores, more memory bandwidth more L3 cache route for my next build. Has anyone else used CPU's like the 6800, 6850, 6900 or 6950x for up sampling? Does my logic hold water?

[Miska]

I looked further up the line at the 6850K (which differs most from the 6800 in base speed only, 3.6GHz vs 3.4 for the 6800) It fairs much better against the 6950x than the 6800 and beats the 6700 in all tests but single thread.

 

I thought about 6850K myself for stereo playback case when running Roon on the same machine because it seems to be sort of sweet spot in terms of cores vs clock speed and should have good cache and RAM access performance. Should be fine without GPU offload too.

 

Here's load figure for same settings and same machine, but without CUDA offload (HQPlayer 3.15.1). CPU load jumping between 5 and 21%. But notice the core utilization is not so great in these cases, so some cores are idling and thus you get lower total load figure.

 

[Quadman]

Thank you Miska, even without cuda offloading the core usage (on what we can see) is much lower than my 6700K with cuda offloading. Memory lanes and L3 is obviously very important in up sampling to the highest rates.

[Miska]

First short report on my new AMD Ryzen server.

 

CPU: AMD Ryzen 7 1700X

Motherboard: ASUS PRIME X370-PRO

RAM: Kingston HyperX FURY DDR4-2666

OS: Ubuntu Server 16.04 LTS

 

With HQPlayer Embedded, upsampling from RedBook to DSD512 (22.5792M) using poly-sinc-short-mp and ASDM7 takes 16% of CPU time total. No CUDA offload.

 

Pretty nice figures for the price.

[Jud]

On a site like the one at the following link, what would the more significant specifications/benchmarks be for stereo playback, with and without convolution; and what if any important information is missing?

 

http://cpu.userbenchmark.com/Compare/AMD-Ryzen-7-1800X-vs-Intel-Core-i7-7700K/3916vs3647

[plissken]

On 3/4/2017 at 4:47 PM, Miska said:

Today I made upgrade for my fanless HQPlayer Embedded machine, motherboard was changed to ASUS PRIME Z270M-PLUS and CPU to i5-7600T. This was installed in already existing Streacom Alpha 10 chassis.

 

Storage is Intel 600p series M.2 SSD and RAM is Kingston HyperX Fury DDR4-2133.

 

Only three heatpipes would fit with this motherboard, but since the CPU is only 35W TDP, it seems to be enough (because the chassis is rated for 65W TDP).

 

I can report that with this setup, it can achieve DSD512 with -2s filter variants at least when using auto rate family. CPU temperature averaging between 42 and 46 C and total load around 50%. But overall, it is intended to work to just DSD256, so it has some reserve left for convolution and such.

 

 

Did you remove the anodization where those heatsinks are before applying them?

[Miska]

2 hours ago, plissken said:

 

Did you remove the anodization where those heatsinks are before applying them?

 

No, there's just plenty of regular white thermal paste between heat pipes and the heatsink (so that the pipe makes proper thermal contact all around, including the mounting block), and same between the PSU heatsink and the case heatsink. Thermal paste between the CPU and cooling block as well as cooling block and heatpipes is Noctua's gray stuff. So far the CPU core temp has not exceeded +60C.

 

For the earlier motherboard (Gigabyte) I would have needed Streacom's thermal riser and then all four heatpipes would have fit. Don't know how much it would have affected cooling capacity though.

[SwissBear]

12 hours ago, Miska said:

First short report on my new AMD Ryzen server.

 

CPU: AMD Ryzen 7 1700X

Motherboard: ASUS PRIME X370-PRO

RAM: Kingston HyperX FURY DDR4-2666

OS: Ubuntu Server 16.04 LTS

 

With HQPlayer Embedded, upsampling from RedBook to DSD512 (22.5792M) using poly-sinc-short-mp and ASDM7 takes 16% of CPU time total. No CUDA offload.

 

Pretty nice figures for the price.

 

Just discovered this thread where I found answers to a lot of questions which remained unanswered on other threads. Thanks to all participants

 

@Miska
Very interesting ! Thanks for sharing.

For a stereo application with convolution (4 correction impulses) and upsampling up to DSD 512 from PCM as well as DSD up to 128, would you suggest to go for this 1700X, the more powerful 1800X,  or the smaller 1700 with only 65w of TDP which would allow fanless enclosure ?

Thanks in advance.

[Miska]

16 minutes ago, SwissBear said:

For a stereo application with convolution (4 correction impulses) and upsampling up to DSD 512 from PCM as well as DSD up to 128, would you suggest to go for this 1700X, the more powerful 1800X,  or the smaller 1700 with only 65w of TDP which would allow fanless enclosure ?

 

As usual, there's never too much computing power. I took the 1700X for test machine because it was well available unlike the 1800X and cheaper.

 

Your use case is pretty heavy, so I would get as much computing power as possible and/or be prepared for GPU too...

 

And the 1700X actually seems to produce quite a bit of heat so it really needs proper cooling.

[OlivierFH]

On 22/03/2017 at 11:05 AM, Miska said:

First short report on my new AMD Ryzen server.

 

CPU: AMD Ryzen 7 1700X

Motherboard: ASUS PRIME X370-PRO

RAM: Kingston HyperX FURY DDR4-2666

OS: Ubuntu Server 16.04 LTS

 

With HQPlayer Embedded, upsampling from RedBook to DSD512 (22.5792M) using poly-sinc-short-mp and ASDM7 takes 16% of CPU time total. No CUDA offload.

 

Pretty nice figures for the price.

Hi Miska,

thanks for all ;-)

Have you done your Ryzen test with or without SMT actived ? As on Windows 10, we (or I) don't know if it is really well managed by the OS on Linux too. 

I saw there are tips with the number of bank of RAM and speed : 

 

Moreover and concerning the thermal management: AMD seams to tell that you have to decrease the thermal indication by 20°C to have the real one... Strange, isn't  ?

[Miska]

7 minutes ago, OlivierFH said:

Hi Miska,

thanks for all ;-)

Have you done your Ryzen test with or without SMT actived ? As on Windows 10, we (or I) don't know if it is really well managed by the OS on Linux too. 

I saw there are tips with the number of bank of RAM and speed : 

 

Moreover and concerning the thermal management: AMD seams to tell that you have to decrease the thermal indication by 20°C to have the real one... Strange, isn't  ?

 

I have SMT enabled, seems to be working and I expect the kernel to catch up fast on any missing things. Probably the SMT is not so different from HyperThreading which is well supported by Linux. Linux also has very good support for local and non-local NUMA etc which was critical in the past with dual-socket Opterons and nowadays with Intel CPUs too, not applicable in this case though. The Ubuntu 4.8.0 lowlatency-hwe-16.04 kernel seems to be best working option at the moment, possibly they have worked with AMD to include any critical patches.

 

I have dual channel 2 DIMM configuration with the DDR4-2666 (or 2667 when correctly rounded).

 

[Miska]

By the way, if someone needs, I built latest 4.9 kernel with patches for both Ubuntu 16.04 LTS and Debian Stretch. Someone was asking a while ago...

 

[shadowlight]

29 minutes ago, Miska said:

By the way, if someone needs, I built latest 4.9 kernel with patches for both Ubuntu 16.04 LTS and Debian Stretch. Someone was asking a while ago...

 

Thx Miska, that would me asking about patched kernel for Stretch.  I will have to wait couple of weeks before testing, since my dac is on the way from upgrade.