HQPlayer4 EC modulator tips and techniques

[Miska]

1 hour ago, 4est said:

Thanks for being specific. If I might ask, do you have specific RAM that you would recommend for use in the Gigabyte Aorus and Designare MBs that have been discussed? I recall seeing you recommend CAS and speed, but I cannot find it.

 

I've been using HyperX (Kingston) Predator RAM. Offers low latencies and good speed. Doesn't cost that much, RAM is cheap again.

 

The amount of RAM you have should be at least the amount OS + HQPlayer takes, plus the largest file you'd ever listen. If you use Roon, it can also consume quite a bit. My largest file is around 8GB (5.0 channel DSD256).

 

[Luca72c]

23 hours ago, Miska said:

 

I would go for the 9900KS as long as stocks last...

 

 

Thank you Miska.

And what if you should want to take a stereo pcm file and convert it to dsd256 using EC modulators but also convolution crossover/req to obtain 6 (or 8 😎) dsd256 channels? 

9900ks would still be the preferable cpu or a higher core count cpu like 3950x would be better (also using cuda gpu if needed)?

[Miska]

29 minutes ago, Luca72c said:

And what if you should want to take a stereo pcm file and convert it to dsd256 using EC modulators but also convolution crossover/req to obtain 6 (or 8 😎) dsd256 channels? 

9900ks would still be the preferable cpu or a higher core count cpu like 3950x would be better (also using cuda gpu if needed)?

 

We don't know yet if 3950X can run EC modulators at DSD256...

 

If you need more channels, you would likely need to add some suitable GPU to deal with the filters so that all cores are left to run the modulators.

 

I have not tried that with 8 channels yet, because my only 9900K machine is my iMac which has (useless) AMD GPU...

 

[Luca72c]

1 hour ago, Miska said:

 

We don't know yet if 3950X can run EC modulators at DSD256...

 

 

Wasn't somebody able to run them using 3900x? I thought 3950x should do the task more easily... 

The problem could be: will it be able to do that on each core pair? 

[Miska]

2 minutes ago, Luca72c said:

Wasn't somebody able to run them using 3900x? I thought 3950x should do the task more easily... 

The problem is: will it be able to do that on each core pair? 

 

Yes, sounded like on Linux it was working better.

 

More cores there are, lower the base clocks get. This is one of the critical aspects. 3900X is 3.8 GHz and 3950X is 3.5 GHz. And AMD doesn't sound like it is able to run all core boost. So the unknown part is how it manages when all cores are under high load. With just two cores under high load it is simpler and it can run high boost clocks.

 

We have seen it before that from the specs it is hard to estimate how the performance will be under real usecase.

 

[Luca72c]

20 minutes ago, Miska said:

 

More cores there are, lower the base clocks get. This is one of the critical aspects. 3900X is 3.8 GHz and 3950X is 3.5 GHz. And AMD doesn't sound like it is able to run all core boost. So the unknown part is how it manages when all cores are under high load. With just two cores under high load it is simpler and it can run high boost clocks.

 

We have seen it before that from the specs it is hard to estimate how the performance will be under real usecase.

 

 

I understand. So maybe it would be advisable to overclock base frequency on all cores around 4 ghz (if possible) sacrificing boost capability... 

But we'll see when the new cpus (including comet lake 12 core i9s) will be available for testing

[Miska]

15 minutes ago, Luca72c said:

I understand. So maybe it would be advisable to overclock base frequency on all cores around 4 ghz (if possible) sacrificing boost capability... 

But we'll see when the new cpus (including comet lake 12 core i9s) will be available for testing

 

Yeah, at the moment 9900K(S) seems to be roughly the best performer.

 

There are 10 - 18 core Cascade Lake ones, but those have fairly low base clocks... (3.7 - 3.0 GHz) Still better than 3.0 GHz of my old 10 core 6950X.

 

Socket 2066 would offer double the memory bandwidth, but seems to be hard to find high clock parts for that socket.

 

I'm waiting for Xeon W-2245 to become available to build a new workstation around that.

 

[Luca72c]

Anyway, first tests show that an OC base frequency of around 4,4ghz can be reached and held by 3950x on all 16 cores at the same time (assuming proper cooling):

https://www.tomshardware.com/reviews/amd-ryzen-9-3950x-review/2

[Samuel T Cogley]

First of all, thanks everyone for making this thread.

 

My setup:

 

Windows 10

ASUS X99 Deluxe

i7 5930k

NVidia RTX 2080ti

64GB of RAM

 

...been doing all my testing so far using an iFI iDAC2.  Going to roll in a Topping D50S very soon.  I can sometimes sustain the ASDM7EC, but not consistently.  Got much better results by turning off "Intel SpeedStep Technology" in the BIOS.  I'm going to be building a new system based on in i9 9900K.  Looking forward to hearing the results.

[barrows]

Hey folks, so what is the consensus?  Some seem to say that I7-9900K works for EC7 at DSD 256, others not so much?

 

I guess an I9-9900K would definitely be fine?  I am right in this, for ASDMEC7 @ DSD 256?  I want to get a machine built up for this, but I require DSD256 (for DSC-2balanced) so I do not want to make any mistakes.  I would also like to run ROON not he same machine.  Thinking 16 Gb RAM to be fully safe.

[randytsuch]

26 minutes ago, barrows said:

Hey folks, so what is the consensus?  Some seem to say that I7-9900K works for EC7 at DSD 256, others not so much?

 

I guess an I9-9900K would definitely be fine?  I am right in this, for ASDMEC7 @ DSD 256?  I want to get a machine built up for this, but I require DSD256 (for DSC-2balanced) so I do not want to make any mistakes.  I would also like to run ROON not he same machine.  Thinking 16 Gb RAM to be fully safe.

I7-9700K? 

 

I have the I7.  I use hqp embedded.  If I use hqp OS booted from a usb stick, I get no stuttering at 256 ec7.

My bios is slightly tweaked, but I was tweaking it when I was trying to get debian and hqp embedded to run ec7 with no stuttering, and I left it as it after running hqp os.  I think a newer version of hqp embedded came out that is supposed to be better with deb but I have something working so I have no interest in trying it.

 

Seems like anyone running hqp desktop has no problems with an 9700k and 256 ec7.

 

Randy

[barrows]

Yeah, I would want to running embedded...  I guess the safest bet is just to go with I9-9900K

[bodiebill]

1 hour ago, barrows said:

Yeah, I would want to running embedded...  I guess the safest bet is just to go with I9-9900K

 

Here I7-9700K with 16GB RAM and no BIOS tweaking, just turbo mode, works flawless with EC7 and DSD256. Temperatures 40-65 C.

[AudioXP]

On 12/2/2019 at 9:50 PM, bodiebill said:

 

Here I7-9700K with 16GB RAM and no BIOS tweaking, just turbo mode, works flawless with EC7 and DSD256. Temperatures 40-65 C.

 

Exactly the same here, with convolution for room correction on top 🙂

[Samuel T Cogley]

Just a follow up to my last.

 

The Topping D50s is impressive for the price.  Can sustain ASDM7EC using poly-sinc-ext2 with Redbook input material.  When I switch to something in the 48kHz domain (48, 96, 192), I get some stuttering.  I'm thinking this is because additional, non-integer work needs to be done there.

 

Just because I could, I tried some DSD512 encoding.  Even with Redbook input material, could not use any EC shaper without stuttering.  This was not unexpected.  When I get my new i9/9900K system built, I'll try this again.

 

 

[4est]

17 minutes ago, Samuel T Cogley said:

Just a follow up to my last.

 

The Topping D50s is impressive for the price.  Can sustain ASDM7EC using poly-sinc-ext2 with Redbook input material.  When I switch to something in the 48kHz domain (48, 96, 192), I get some stuttering.  I'm thinking this is because additional, non-integer work needs to be done there.

 

Just because I could, I tried some DSD512 encoding.  Even with Redbook input material, could not use any EC shaper without stuttering.  This was not unexpected.  When I get my new i9/9900K system built, I'll try this again.

 

 

I do not think anyone has achieved 512 with EC.

[bodiebill]

My setup:

Audiolinux headless with HQPlayer embedded latest version

ASRock Z390M-ITX/AC

i7-7900K

1x Apacer DDR4 RAM 2666MHz 8GB (non-ECC)

 

Since putting in the Apacer RAM I get occasional tiny dropouts when upsampling to DSD256 with ASDM7EC. I never had these before using hyperthreading and turbo mode without overclocking.

I just reset the BIOS to the optimal settings and at first that seemed better, but now the occasional dropouts are back.

I will try to put in a second RAM (8GB => 16GB) and see if that helps, but I expect not as 8GB should be sufficient for HQPlayer?

Any advice?

 

[bodiebill]

Doubling RAM 8GB => 16GB did not help. The dropouts even occur when upsampling to only DSD128. Also, they are unlike the dropouts I experienced with EC7 upsampling to DSD256 when my system was not yet powerful enough.

 

As I upgraded to HQPlayer version 4.13 at the same time as I put in the Apacer RAM: could it be a problem with version 4.13?

[bodiebill]

So i went back from AL + HQP 4.13 to Jussi's 4.12 image. Still tiny dropouts with ASDM7EC / DSD256 😐

 

So what now? Should I try to overclock? I have no experience with that. Can anyone advise how to do this for an ASRock Z390M-ITX/AC motherboard?

[bodiebill]

Apologies if I am talking to myself, but it seems that subtle overclocking did the trick. The only setting I changed in the ASRock Z390M UEFI BIOS was going from Load-Line Calibration level 4 to level 3. This applies additional voltage to the CPU under load to provide more stability.

Fingers crossed...

[randytsuch]

Seems like you're running into what I was, guess an 7900k is at the edge for 7ec at 256dsd.

 

Using HQP OS is what fixed it for me.   Just upgraded to the newer version, still working fine.

[bodiebill]

2 hours ago, randytsuch said:

Seems like you're running into what I was, guess an 7900k is at the edge for 7ec at 256dsd.

 

Using HQP OS is what fixed it for me.   Just upgraded to the newer version, still working fine.

 

Perhaps, but I am somewhat puzzled as the dropouts are tiny and only occur after 40 seconds or so. And they also occur with dsd128 and temperatures remain moderate (<60C). So I suspect it is something else.

[randytsuch]

21 hours ago, bodiebill said:

 

Perhaps, but I am somewhat puzzled as the dropouts are tiny and only occur after 40 seconds or so. And they also occur with dsd128 and temperatures remain moderate (<60C). So I suspect it is something else.

Yes, I think you're right, I missed that you get dropouts at 128.  I only had dropouts at 256 and 7ec.

[Miska]

On 12/21/2019 at 3:51 PM, bodiebill said:

My setup:

Audiolinux headless with HQPlayer embedded latest version

ASRock Z390M-ITX/AC

i7-7900K

1x Apacer DDR4 RAM 2666MHz 8GB (non-ECC)

 

Since putting in the Apacer RAM I get occasional tiny dropouts when upsampling to DSD256 with ASDM7EC. I never had these before using hyperthreading and turbo mode without overclocking.

I just reset the BIOS to the optimal settings and at first that seemed better, but now the occasional dropouts are back.

I will try to put in a second RAM (8GB => 16GB) and see if that helps, but I expect not as 8GB should be sufficient for HQPlayer?

Any advice?

 

What kind of CAS latency does this memory have compared to earlier one? Both configurations with 4 memory modules?

 

This CPU requires that it is running at high enough boost clocks for the relevant cores in order to perform ASDM7EC successfully.

 

What kind of filter are you using?

 

[Miska]

On 12/11/2019 at 6:59 PM, 4est said:

I do not think anyone has achieved 512 with EC.

 

Only way I currently know is to process offline with HQPlayer 4 Pro, to 512 or 1024. But doesn't work in realtime with current CPUs. And I can tell I certainly have tried to extract every last bit of performance out of the CPUs. But I still have some CUDA rocks unturned.