Older Intel CPU working with 256 ASDM7EC

[gsbrva]

Hi all, I have been lurking on here for years but have not posted.  

 

If this has been gone over before, I apologize.  I did quite a few searches trying to find a solution to my problem of too old/slow CPU playing DSD256 with ASDM7EC.  I did not want to upgrade the computer so I experimented a bit and got it working on 44.1. 

 

My computer is an older Intel 6-core that was strong in its day:

 

I7-6850K high end desktop CPU 32 GB memory (not overclocked 3.8ghz boost clock). 

Audio out is a EVGA NU Audio Pro  

CUDA -  6GB 1060 loaner upgrade.  (thanks 4est!).

 

I borrowed the graphics card in hopes it would be the solution.  It was not although CUDA is needed for reliable playback with my CPU.  I could not get this combo to play DSD 256 with the ASDM7EC until now.  At best it would play unreliably.

 

My struggles led to trying this Intel app:

 

Intel® Turbo Boost Max Technology 3.0 Application / Driver

I used the driver panel to assign HQPlayer4Desktop.exe to my CPU's best performing cores.  This assignment seems to exclude other system processes from using the two cores for left and right channels and the HQ Player process stays locked on the assigned cores.  These are the settings I'm using:

 

Enable checked

add app HQplayer4Descktop.exe

Evaluation Interval 10 (doesn't make a difference in this case)

Utilization Threshold 80  (just needs to be in a range that detects the load)

Affinitize all Demanding work  unchecked  (you don't want other processes getting the CPU resources)

Affinitize to all cores checked (if your CPU has two fast factory marked cores you can select the other option, mine did not although one core is allowed to clock 4.0 ghz)

 

In HQ Player 4.8.0:  

 

Multicore either grayed or unchecked (not sure if this matters with the core assignment)

CUDA offload checked (unreliable without)

Adaptive output rate checked (unreliable without) My DAC supports both rate families.

Filters - probably any but sinc-L and the non-2s versions of xtr (had trouble getting core assignments of the threads)

Modulator ASDM7EC

 

I did not get the good playback until I would monitor two cores running at near capacity and the other 4 cores quiet.  If the process jumped from core to core playback was still unreliable.  On this CPU which is Broadwell family, I get reliable playback starting at around 3.7ghz boost clock for 44.1 source.  The factory Turbo Boost is 3.8ghz which works consistently for me with no drop outs.    There seems to be a noticeable efficiency gain for not jumping around between the cores. 

 

I realize that Miska has already gone to some lengths to make sure HQPlayer is using the cores efficiently.  I'm not entirely sure why this works so much better for me than than the multicore options.  Maybe it is particular to this CPU family.

 

It would be great if somebody else could confirm these results or show me how to get the same results with the HQ player settings instead.  Thanks!

[Miska]

On 12/20/2020 at 9:13 PM, gsbrva said:

Multicore either grayed or unchecked (not sure if this matters with the core assignment)

 

Yes it does, unchecked makes minimum amount of work split to multiple cores. While grayed does auto-configuration based on detected CPU topology. Checked makes all work split explicitly enabled regardless of what kind of CPU topology is, or if it is known to make performance suffer. Core assignments are based on how much work split there is.

 

On 12/20/2020 at 9:13 PM, gsbrva said:

There seems to be a noticeable efficiency gain for not jumping around between the cores.

 

Certainly, because there's always notable work migration cost. This is one reason why HQPlayer locks tasks to specific cores.

 

Without this "core pinning", Windows has bad habit of shuffling work between cores such that the performance suffers a lot because of it.

 

[toddrhodes]

Was going to see if this might work for my older 6700k but no dice, it's only for slightly newer I7/i9 CPUs.

 

But... I believe the ability to run those EC modulators is solely dependent on ST performance (one core per channel) and, frankly, clock speed. My 6700k at 4.2 can easily do ASDM7EC at 128/48 and plays it perfectly at 256/44.1 but every 10-15 minutes, I get one tiny dropout on 48k material at 256. So what I do is set my rate to 256/44.1 and ASDM7EC as my modulator, and check auto-rate family. That way I get 256/44.1 and 128/48K. Works like a champ and frankly, unless I could reliably O/C this thing to 4.5GHz, I think that's where I'm going to live for awhile. And my Ryzen 3700x wouldn't do any better, nor will the 3900x I've got coming because while they will spank the holy hell out of the 6700k in everything else, this one, single-threaded performance task (modulator performance) doesn't really care how many threads and such we have. 

 

And I don't think, or at least I didn't think this had anything to do with CUDA offload either. Stuff like Sinc-L and possibly the poly-sinc filters could use help with CUDA Offload, but not the modulators. But maybe it's a scenario where if you offload enough work from the CPU with CUDA, that then there is enough resources available in those primary cores to do more work for the modulators? I'm just speculating, and bored, so take this all with a grain of salt :D

 

Either way, I'm pretty impressed with my aged 6700k. Of course it's doing NOTHING else but HQP, but still, better than just sitting on a shelf collecting dust I guess.

[Miska]

18 hours ago, toddrhodes said:

Was going to see if this might work for my older 6700k but no dice, it's only for slightly newer I7/i9 CPUs.

 

But... I believe the ability to run those EC modulators is solely dependent on ST performance (one core per channel) and, frankly, clock speed. My 6700k at 4.2 can easily do ASDM7EC at 128/48 and plays it perfectly at 256/44.1 but every 10-15 minutes, I get one tiny dropout on 48k material at 256. So what I do is set my rate to 256/44.1 and ASDM7EC as my modulator, and check auto-rate family. That way I get 256/44.1 and 128/48K. Works like a champ and frankly, unless I could reliably O/C this thing to 4.5GHz, I think that's where I'm going to live for awhile. And my Ryzen 3700x wouldn't do any better, nor will the 3900x I've got coming because while they will spank the holy hell out of the 6700k in everything else, this one, single-threaded performance task (modulator performance) doesn't really care how many threads and such we have. 

 

And I don't think, or at least I didn't think this had anything to do with CUDA offload either. Stuff like Sinc-L and possibly the poly-sinc filters could use help with CUDA Offload, but not the modulators. But maybe it's a scenario where if you offload enough work from the CPU with CUDA, that then there is enough resources available in those primary cores to do more work for the modulators? I'm just speculating, and bored, so take this all with a grain of salt :D

 

Either way, I'm pretty impressed with my aged 6700k. Of course it's doing NOTHING else but HQP, but still, better than just sitting on a shelf collecting dust I guess.

 

Since running modulators for stereo on Intel/AMD depends on performance of two cores (and on four cores on Apple silicon), and thus Turbo boost clocks they can have, offloading may release enough work from the remaining cores that the critical cores can keep up with high enough clocks.

 

Have you considered trying this same case with Ubuntu Studio 20.04 (and possibly with my custom kernel)? It could be just enough more efficient OS to make it work better.

 

[toddrhodes]

4 minutes ago, Miska said:

 

Since running modulators for stereo on Intel/AMD depends on performance of two cores (and on four cores on Apple silicon), and thus Turbo boost clocks they can have, offloading may release enough work from the remaining cores that the critical cores can keep up with high enough clocks.

 

Have you considered trying this same case with Ubuntu Studio 20.04 (and possibly with my custom kernel)? It could be just enough more efficient OS to make it work better.

 

I absolutely have wanted to do that but as expressed in PM a couple days ago across multiple replies, unfortunately right now is not a good time to buy a second full license for experimentation. And half an hour of playback at a time is not necessarily enough to properly evaluate. 

 

I do have Embedded set up on a flash drive and can boot to it with my 6700k machine, but even if I find that it works and I like it, a second license which renders my first license effectively moot, isn't something I can do right now, so it doesn't really even make sense to try it since I can't really use it.

[Miska]

11 minutes ago, toddrhodes said:

 

I absolutely have wanted to do that but as expressed in PM a couple days ago across multiple replies, unfortunately right now is not a good time to buy a second full license for experimentation. And half an hour of playback at a time is not necessarily enough to properly evaluate. 

 

I do have Embedded set up on a flash drive and can boot to it with my 6700k machine, but even if I find that it works and I like it, a second license which renders my first license effectively moot, isn't something I can do right now, so it doesn't really even make sense to try it since I can't really use it.

 

I don't mean Embedded. HQPlayer 4 Desktop license covers all three platforms. Ubuntu Studio is a lightweight desktop OS.

 

[toddrhodes]

5 minutes ago, Miska said:

 

I don't mean Embedded. HQPlayer 4 Desktop license covers all three platforms. Ubuntu Studio is a lightweight desktop OS.

 

 

I can try, but I'm completely linux ignorant. That's why embedded was so nice, lol. But let me see if I can figure it out and report back. Thanks for the clarification!

[toddrhodes]

Ok so that was really easy. I'm off and running from a USB thumb drive in Ubuntu 20.10 with HQP Desktop 4.81, listening to 256x48 ASDM7EC now on the 6700k machine, no GPU/CUDA offload. Only a few minutes of playback thus far but no dropouts yet. I'm still going to see if I can push this CPU up to 4.4 or 4.5 on air.

 

Sounds terrific, though! Using XTR-LP-2S for both 1x and Nx SDM output.

 

Now i have to learn a bit more about Ubuntu, just so I'm at least somewhat familiar with what I'm doing... Like maybe adding VNC or something to it for remote management. Right now I have that going into my AVR via HDMI out but I really don't want to use that.

[evalon]

Hi all, 

 

Quote

Ok so that was really easy. I'm off and running from a USB thumb drive in Ubuntu 20.10 with HQP Desktop 4.81, listening to 256x48 ASDM7EC now on the 6700k machine, no GPU/CUDA offload.

 

Well, that sounds really interesting as I would also prefer to use a slightly older computer for HQPlayer. I was wondering, though, what happens with e.g 192kHz files - have you tried this? 

 

Cheers & Happy New Year,

 

Jesper 

[toddrhodes]

4 hours ago, evalon said:

Hi all, 

 

 

Well, that sounds really interesting as I would also prefer to use a slightly older computer for HQPlayer. I was wondering, though, what happens with e.g 192kHz files - have you tried this? 

 

Cheers & Happy New Year,

 

Jesper 

 

No issues really, and here lately I've been on a 768 PCM + Sinc-L + LSN15 kick so it's even easier on the CPU/RAM. I have Auto-Rate checked so it will step it down to the proper rate family, as I see no real point in going from 44.1 to 48k family at such high PCM or DSD rates to begin with. I would think the easier computation would "sound better" than a strict adherernce to the 48k rate class or something.

 

Hope that makes sense, but yea - no issues playing native DSF or 24/192 once I dialed everything in well enough.

[evalon]

... Hi & thanks for your prompt reply 😉 ... which also made me realize that I had been imprecise in the wording of my question: What I meant was how your PC works with converting e.g. 192 kHz files to DSD256 ASDM7EC? Can it do this without drop-outs or the like? ... I'd like to be able to play DSD at as good/high a resolution as possible ... 

 

Cheers,

 

Jesper 

[toddrhodes]

No worries, my reply could have been more clear as well. Playing any 48k rate - 48, 96, or 192 at 256 with ASDM7EC is doable but it taxes the hell out of the system and is on the ragged edge of perfect. So, I deploy auto rate and a 256/44.1 rate so that on those 48k rate families, I step down to dsd128/ASDM7EC. To me it still sounds brilliant, but the raw horsepower just isn't there on the 6700k even in a minimal install, in all operating conditions.

[juanitox]

 with the 6700K    DSDEC  256 works flawless  at  44 /48 / 96 PCM   dends on the Dac but sounds very good with T+A  and HOlo spring..

[evalon]

@toddrhodes & juanitox: Thank you both for your feedbacks. I will just consider what may be most feasible, however, over a quiet New Year I have thought about various additional options. And given that prices on cpus are continually going down I am now also considering buying a mini-itx motherboard and the I9-9900k if it can do DSD256 EC modulators without the additional graphics card (i.e. with the on-cpu UHD 630 graphics processor).

 

Compared with an additional graphics card this will (significantly) reduce power consumption and - hopefully - give some extra speed capacity for ?? upcoming processing needs. And as I never play games I don't really need a dedicated graphics card unless needed for HQPlayer ... 

 

@Miska: To this end ... Can I ask you Miska, if you think that HQPlayer will do DSD256 EC modulators (including whatever "fringes - i.e. additional sound quality improving processing that may be needed for HQPlayer") with an (almost) dedicated quality Z390 motherboard (mini-itx), 8GB Kingston HyperX RAM 2666 MHz DDR4, fast SSD HD, an i9-9900k or 'ks processor (with built-in Intel UHD Graphics 630)? Clocking to 5 GHz will be possible, an optimized Win10 is likely to be used (Ubuntu 20.10 is less feasible I guess, as I will also be using this PC for a Kontakt Player digital piano).

 

Happy New Year - for a hopefully better 2021 

 

Cheers,

 

Jesper 

[Miska]

6 minutes ago, evalon said:

@Miska: To this end ... Can I ask you Miska, if you think that HQPlayer will do DSD256 EC modulators (including whatever "fringes - i.e. additional sound quality improving processing that may be needed for HQPlayer") with an (almost) dedicated quality Z390 motherboard (mini-itx), 8GB Kingston HyperX RAM 2666 MHz DDR4, fast SSD HD, an i9-9900k or 'ks processor (with built-in Intel UHD Graphics 630)? Clocking to 5 GHz will be possible, an optimized Win10 is likely to be used (Ubuntu 20.10 is less feasible I guess, as I will also be using this PC for a Kontakt Player digital piano).

 

For some of the more heavy filters or certain convolution cases you may need GPU, but for most things it should be fine. And in any case you could later add GPU if needed. Mini-ITX boards still typically have one 16x PCIe slot.

 

[evalon]

Thanks Miska for your being so forthcoming in replying to these PC-related questions. Not being very computer savvy myself it is a great help to get feedback from you (& other people here) ...  

 

Cheers,

 

Jesper