HQplayer 4 and Mac-mini M1

[camott]

47 minutes ago, AudioDoctor said:

 

I do, and saying "they just use off-the-shelf USB cards (Amanero and now JL Sounds)." shows a fundamental misunderstanding or lack of knowledge perhaps, of whats actually inside the DAC and how Lukasz makes it sound the way he wants, and how he makes it do what he wants it to do.

 

There are two clocks on that Amanero board that cost 470 euros a piece a few years back, probably more now. This is just one example. Thinking that Lukasz just throws an off the shelf board in there and calls it a day is laughable.

 

The Amber 3 I recently owned clearly had an off-the-shelf JL Sounds USB module in it. Prior to that they were using Amanero boards. The Pacific is obviously a much more $$ DAC than the Amber and maybe he modified them but I don't believe this is the case in all of his DAC's. Regardless, Lampizator does not employ any form of galvanic isolation on the USB input that I have ever seen or heard of. And thus there is potential for electrical noise transfer from the upstream endpoint (NAA or otherwise) and resulting SQ differences. If the Lampizator USB implementation is so good then why are you using I2S and bypassing it???

 

On the other hand, some DAC's do employ full galvanic isolation on the USB inputs and the recent Holo DAC's in particular are known for their ability to obfuscate any input differences. (both USB and other inputs via PLL). So instead of " the Holo DAC is putting its sound onto everything" it's actually the opposite. This was simply the point of the original comments that you then disagreed with.

 

For what it's worth, I actually really liked the organic sound from my Amber 3 and think about getting the Baltic 3 sometimes. There is definitely something about the sound profile Lukasz imparts on his DAC's, above and beyond tube coloration. But I think they have build issues and the noise floor was insanely high. (Ever wonder why Lukasz has never published a single technical measurement or simple number like SnR?). They are completely different beasts from the Holo DAC's of course - complete fodder for objectivist vs subjectivist arguments.

 

[camott]

@Miskajust tried out HQP 4.15.0 on the M1 and the CPU core allocation changes you made have increased performance significantly. Core usage has gone from 4 pegged out cores to a broader distribution across all 8 cores and I can now run 192 -> DSD256x48 with ASDM7ECv2 and filters like gauss-long without issue.. Kudos!!

[camott]

Though I haven't tested it on 4.16.2, I was having dropouts with guass-xla 48k -> 44xDSD256 or 48xDSD256 on recent HQP revisions. (44k source is ok). Only the very first revision of 4.15.0, with the new core allocation model, worked completely without issue. Since then the load is lower overall, but is not distributed quite as evenly across all 4+4 cores. This has been discussed elsewhere ... https://community.roonlabs.com/t/which-hqp-filter-are-you-using/6061/2171. I don't know if @Miska can change the cpu allocation heuristic for just the MM M1 back to what it was in 4.15.0. 🥺

 

[camott]

Here is htop for same track gauss-xla 48k -> DSD256 in 4.15.0 and 4.16.2 (and likewise all versions since 4.15.1).

 

4.15.0 has efficiency cores around 65% and power cores running around 85%.

4.16.2 has efficiency cores around 45% and power cores maxed out just below 100%.

 

I don't actually use gauss-xla anymore so it's not a concern for me ... but might be helpful for others.

 

 

[camott]

And just like that, 4.17.0 release with EC modulator optimizations have solved any performance concerns with gauss-xla on M1. If you run with multicore fully checked it behaves like recent versions, but if you auto/grey check it then it has much improved performance. nice work @Miska!

[camott]

No Mac Mini M1 update today. Boooo. $1999 for the larger Mac Studio M1 Max. $3999 for M1 Ultra (2x M1 Max). They don't really pencil out as well for an HQP box. Will have to wait for the Mac Mini M2 apparently. 

[camott]

Asahi Linux for M1 Macs Alpha release is out! @Miska Any chance HQP desktop or embedded will work on it, with or without a bit of tweaking?? Maybe a bit more performance can be extracted vs Mac OS.
 

Asahilinux.org

 

[camott]

@proulxarts below is an htop capture of MM M1 running roon + HQP, upsampling to poly-sync-gauss-hires, ASDM7ECv2 DSD256x48 ... Roon uses a fraction of the cpu that HQP does.

 

Also, MM M2 might very well drop at Apple's October event ... so you might want to hold off a few weeks? I am hoping that a MM M2 (pro?) will be able to do DSD512 but we'll see ...

 

 

[camott]

14 minutes ago, AudioDoctor said:

 

How did you get htop on the Mac?

 

Homebrew package manager.  https://brew.sh

[camott]

On 1/22/2023 at 4:11 PM, Miska said:

I'm eagerly waiting for first reports by HQPlayer users.

 

Did some quick tests on a base M2 Mac Mini. Can't do DSD512 with ECv2 :(

 

HQPlayer CPU load is roughly 10% better than M1, in line with Single Core benchmark improvements, but not enough to get it over the hump. It looks like one of the four Power cores is maxed. I highly doubt the M2 Pro will be any different - single core clock speed is the same.

[camott]

6 minutes ago, AudioDoctor said:

 

Did that Machine have only 8GB of RAM?

 

Yes and I was sure to monitor this. RAM usage for HQP was under 1GB and overall memory usage a non-issue.

 

 

[camott]

2 hours ago, rjanderjr said:

I have a Mac Studio Ultra with 20 cores, 4 of which are efficiency cores.  This is an M1 chip so the M2 performance cores will process faster.  I can not say if the M2 Mac mini Pro with 8 to 10 performance cores will be able to play DSD 512 using the ASDM7ECv2 modulator, but it should, depending on the source file.  I would guess that it could play any DSD file and upsample to 512, given how the cores are stressed on my Studio Ultra.  But could play a PMC file and upsample it to DSD 512.  That stress tests my Studio Ultra.  It works, but requires a lot of processors running at high load.  See image.  Finally.  I have found that running HQP on Ventura requires more processing power.  I have also found the HQP 4.21 requires more processing power than 4.20.2.  I run into dropouts running 4.21 and upsampling to DSD 512.  Right now I am running Ventura with HQP 4.20.2 with all stereo files and no dropouts. The most efficient is Monterey with 4.20.2.

 

Hi, that's very interesting! The single core speed of M1 Ultra is the same as other M1's afaik. And I was under the impression that the SDM modulation could only use 2 p-cores - which is the big bottleneck with DSD512 (again, from what I understand). And the rest of the performance cores were for filters, etc. [edit - you can certainly see the load distributed across your 16 p-cores, with the 4 e-cores lower.]

 

If more p-cores are making the difference here then yes the M2 Pro, with 6 or 8 p-cores, vs base M2 4 p-cores might do it??

 

 

[camott]

Yes clearly it's not so straightforward. I will go back and hunt down Jussi's posts on this later. One thought is that the 48MB l2 cache is making the difference with the M1 Ultra. Regardless, great that you have DSD512 ECv2 working. I don't think I have seen any other reports of Mac users doing this.

[camott]

5 hours ago, copy_of_a said:

If I may ask - if not DSD512 … can the base MMini M2 with 8GB RAM handle any source file upsampled to DSD256 with poly-sinc-gauss (long) and ASDM7ECv2?

Would be useful to know - many thanks in advance!

 

 

Yes but this was already the case with the M1.