HQplayer 4 and Mac-mini M1

[Miska]

18 minutes ago, Kalpesh said:

unfortunately, with MMM1 it's not real for Nx rates and DSD 64. But I can see you have twice as many performance cores : maybe you can ; can you confirm for PCM up to 192 and DSD 64 ?

I wish I could enjoy ASDM7EC Light + Sinc MG without interrupts ; sounds great !

 

 

Yes, these work fine. Note that the PCM cases have Oratory headphone EQ for Sennheiser HD800 enabled, for DSD source I need to disable it (switch to empty matrix profile) or I get dropouts after a while.

 

This also works fine without issues:

 

Note that M1Max has 4x the memory bandwidth of M1. And M1Ultra would 8x the memory bandwidth. This also has it's own impact on performance.

 

[Miska]

4 minutes ago, Kalpesh said:

Thank you. So dropouts with DSD 64 and convolution....

 

I have not tried convolution in this scenario. My headphone EQ is parametric. It always depends on length of the the convolution filter or number of parametric EQ bands.

 

4 minutes ago, Kalpesh said:

Would you expect better from a Mini M2 Pro ?

 

No, I don't think it is any better.

 

4 minutes ago, Kalpesh said:

For whatever reason second hand market is flooded with entry Studio while there is no Mini M2 Pro and I guess you recommend the 500 euros/$ 32 GB RAM option ?

 

My MacBook Pro is with 32 GB RAM. But reason for that RAM is because HQPlayer Mac builds, testing and development is done on this machine.

 

But this is very very power efficient. The laptop is barely warm to touch while doing this. And pretty quiet, fan running at level that I can just hear it when I take headphones off.

 

[Miska]

11 hours ago, semente said:

Can't HQPlayer Embedded be installed on Mac hardware (e.g. Mac mini)?

 

Intel Macs yes, theoretically on M1/M2 as well if you manage to get Linux running on it properly. As long as it runs Linux it is technically possible.

 

[Miska]

6 hours ago, Jud said:

The kernel versions with good M1 support start at 6.2. The version of Ubuntu Miska builds on/for, 22.04 LTS, is on kernel 5.19.x, I believe.

 

Kernel and what runs on top of the kernel are of course two separate things. One issue you would face with running Linux and Embedded on M1/M2 is that Apple has not upstreamed their special compiler optimizations to the open source compiler. I tried, building a mainstream open source compiler on macOS will make HQPlayer performance absolutely suck.

 

Not something that would matter much for running command on terminal, or running something simple. But will matter a lot when you want to squeeze absolutely all the possible processing performance out of the hardware in a realtime application.

 

[Miska]

45 minutes ago, semente said:

I don't see myself using rates above DSD256.

Should I get an M2 or is the M1 enough? (buying used, so M2 not an easy find)

 

M2Pro has the advantage of more cores and higher memory bandwidth that allows wider range of filters and more DSP processing options. But baseline things like default settings work on the old regular M1 as well.

 

[Miska]

1 hour ago, Kalpesh said:

as far as HQP is concerned : M2 Max > M2 Pro > M1 Max >> M1/M2 ?

 

I think M1 Max and M2 Max are pretty much on par. The difference between M1 and M2 is rather small.

 

[Miska]

2 hours ago, Kalpesh said:

Guess Blender is not of significance for HQP then...

 

Is there a useful benchmark in this : https://www.pcmag.com/news/apple-silicon-showdown-all-the-m1-and-m2-processors-tested-and-compared or elsewhere ?

 

Only useful benchmark is HQPlayer itself. None of the benchmarks I have seen so far resemble HQPlayer workloads.

 

[Miska]

2 hours ago, Kalpesh said:

@Miska The choice of modulator being technically driven, what is your recommendation for Holo ?

 

EC Super is very soft (nicely so) and depicts layers very nicely. But I tried a piano record where it stretched left the keyboard compared to ECV3 : Super is more revealing or the same technicality that makes instruments individualised in an orchestra becomes a liability with a single (large) instrument ?

 

For Spring 3 / May, ASDM7EC-light/super 512+fs at DSD512, especially if you are listening mostly classical. But any of the 7th order EC modulators work very well.

 

[Miska]

12 hours ago, Kalpesh said:

Anybody using Sinc Long on Apple Silicon ?

can @352.8 PCM, no more

glimpses at 1.5 M before crash were amazing

Wonder How much is possible with more recent Mac ?

 

If you don't need conversion between rate families, maybe sinc-Lx group instead?

 

[Miska]

On 6/2/2023 at 11:15 PM, Schafheide said:

On my Mini M2 Pro, using Desktop v4 or v5 - when I O/S DSD64 files to DSD256, when I pause the playing, I can just hear very faint, random "ticks" (like tiny scratches on an LP).  This does not happen when I O/S to DSD512 or when I play PCM or DXD files.  Any thoughts?

 

One typical reason for such are lost USB packets, causing 125 µs long dropouts. Such are usually hardware related. For example RPi3 based NAAs suffer from such and for that reason are not supported for NAA use.

 

[Miska]

11 hours ago, Schafheide said:

I tried this on my Mini M1 8GB - the clicks are still there (they are also there when I play the file, when it is on an attached USB drive).

I am thinking that this only became obvious after an update to macOS Ventura (not certain which one).  

Not sure if I can easily go back a couple of those updates?

 

I cannot reproduce this on my Macs. Is your DAC locally connected? If not, I would try with DAC connected locally through USB.

 

[Miska]

4 hours ago, Tihon said:

@Miska do you have any info comparing M1 Max vs M2 Max or M1 Ultra vs M2 Ultra?

 

Not really.. I don't have any M2 machines yet myself. Just the original M1 Mini (16 GB) and then this newer MacBook Pro with M1Max (8p+2e, 32 GB RAM).

 

4 hours ago, Tihon said:

My MBP (M1 Max, 32Gb RAM) can run HQPlayer with ASDM7EC-light at 512 or ASDM7EC-super at 256 with poly-sinc-gauss-long and poly-sinc-gauss-hires filters.

 

That's the same I have and can do.

 

4 hours ago, Tihon said:

I wonder what is achievable with new Mac Studio with M2 Ultra or M2 Max. ASDM7EC-super at 512? ASDM7EC-light at 1024? Single core performance difference between M1 and M2 generation is around 11%. Memory bandwidth for M2 Ultra is 2x higher than M1 or M2 Max.

 

I doubt DSD1024 is possible even with M2Ultra, because the single core performance difference is rather small. But since it has really fast memory access and more performance cores, it will likely help a lot with big and heavy filters and convolution. And of course with multichannel, if one prefers such.

 

4 hours ago, Tihon said:

By the way, am I right that with DSD upsampling modulator is the most demanding part and use of convolution/filters do not make a big difference, because these operations are performed by different cores?

 

Those big filters are rather heavy, and also convolution/EQ for DSD sources can get really heavy.

 

For modulators, I think M1Max has already things covered for stereo.

 

[Miska]

8 hours ago, Tihon said:

Hmmm.. I thought the modulators were the most demanding part that limited max rate. And that is the reason why my M1 Max runs ASDM7EC-lite at 512 but ASDM7EC-super only at 256. On the other hand, @rjanderjr is able to use -super at 512 rate with M1 Ultra Studio, but our processors have similar single-core performance, difference is only in core numbers and memory bandwidth.

 

Adding more cores than what M1Max already has likely doesn't improve modulator case for stereo output notably. Memory bandwidth may have some impact, but shouldn't hugely impact modulators that should mostly fit CPU caches. But memory bandwidth plays role with filters/convolution.

 

8 hours ago, Tihon said:

By the way, what do you think about memory requirements? It's pretty obvious that with bandwidth it's "higher -> better", but how much memory we need in case of DSD1024 upsampling with reasonable filters (poly-sinc family). Is 32Gb enough, or we will need 64 or 128?

 

It depends on what you want to do. It is possible to exhaust 32 GB with HQPlayer. But for most common use cases, 16 GB should be enough on macOS. If you run other applications together with HQPlayer, then more memory may be useful. For example browsers can be memory hogs if you have a lot of tabs open.

 

[Miska]

43 minutes ago, Schafheide said:

I have a problem converting PCM96 to PCM1536 on my Mini M2 Pro.

There is a dropout approx every 10 secs.

It has no probs converting PCM or DXD to DSD512 (using poly-sinc-gauss-hires-lp and ASDM7EC-light 512+fs).

I would appreciate if someone could share their successful settings on their Mac.

 

Same filter in both cases?

 

[Miska]

27 minutes ago, Schafheide said:

Yes, same filter (although I have tried others).

 

Strange, since poly-sinc-gauss-hires-lp is not even particularly heavy.

 

27 minutes ago, Schafheide said:

I just noticed that the manual say that HQP will only convert PCM files up to PCM768.

But it will convert DSD files up to PCM1536.

 

Does it? I have some mistake of forgotten ancient thing somewhere then. Currently PCM is supported up to 2M and DSD up to 2048.

 

[Miska]

1 hour ago, Schafheide said:

I have a problem converting PCM96 to PCM1536 on my Mini M2 Pro.

 

NAA connected over wired ethernet? 

 

My MacBook Pro M1Max has no trouble, it is almost idling:

 

This is playing FLAC from SMB share to NAA output.

 

Also works with metering enabled:

 

(efficiency core load is web browser with some 50+ tabs open)

 

[Miska]

16 minutes ago, Schafheide said:

See the manual 5.0.2 page 6 Item 1.1  DSDIFF and DSF Playback ..........

 

Page 6 talks about NAA?

 

But that is talking about DSD-to-PCM conversion. Output rate of the conversion is 1/16th of the source rate. Thus 176.4k for DSD64, etc... It is not limited to those mentioned examples. Which can be further converted to any other rate.

 

[Miska]

2 hours ago, AudioDoctor said:

The Holo Red is based on a Raspberry Pi4 compute module, and it was my understanding that the Pi4 was only capable of 768 pcm  and 512 DSD.

 

With NAA OS, at least mine works up to 1.5M PCM and DSD1024... 🙂

[Miska]

1 hour ago, juniorbudel said:

Guys, I'm thinking of buying a Mini M1 or M2 to use as a HQPLAYER server, my current server runs upsampler for DSD 256 with EC7 light only and it still stutters in some moments. Could it be that with the Mini I will have more reserve to run with EC 7 V3, for example? For purchase here the M1 and M2 are practically the same price, has anyone compared? Can the M2 run filters and modulators more freely?

 

I don't have any M2 hardware, but based on reports, in some cases M2 seems to be slower the M1 and in some cases faster. But the difference should be mostly rather small. Biggest advantage of M2 Mini models is with M2Pro which provides more memory bandwidth and performance cores.

 

Either way, I would strongly recommend getting 16 GB model if you go with M1 or M2.

 

[Miska]

6 hours ago, juniorbudel said:

I'm thinking how much it could improve with the direct connection between the server and the Optical.

 

There is no benefit in direct connection. It will just create problems and it is unsupported. Networking systems are not designed for direct point-to-point connections, but instead for networks.

 

[Miska]

On 7/14/2023 at 10:56 PM, Kalpesh said:

I have just ordered a Mac Studio M1 Max from Apple's refurb store : don't know if it will come with Monterey or Ventura. In any case is it worth to bother change ? What's the ideal OS with HQP5 ?

 

I'm using the latest to test... When a new release comes out, before upgrading wait a bit for possible problem reports. You can for example ask here for my test results. Once I've successfully moved over to latest, it is certainly safest choice.

 

Ventura introduced some CoreAudio bugs on Intel based Mac platforms, but HQPlayer now has workaround for those. No issues with Apple silicon though. Probably they just don't test as much on Intel hardware anymore.

 

[Miska]

3 hours ago, Kalpesh said:

Thank you Jussi. Is there any hope Sonoma's gaming feature benefitsHQP/brings some kind of CUDA offload ? May I thus come to regret the entry level GPU number of cores ?

 

No, I don't think it would have big impact on HQPlayer. Maybe just reduce likelihood of drop-outs. Apple GPUs are only used for graphics by HQPlayer, not for any CUDA style computation. So for HQPlayer use the number of GPU cores doesn't really matter.

 

[Miska]

7 hours ago, Tihon said:

I saw that. I wonder if @Miska found out how to make HQPlayer work with M2 Ultra.

 

From HQPlayer's perspective it is no different from any other M1/M2. HQPlayer knows the cores and everything else. But Apple doesn't let applications have much overall control over many things. So if something doesn't perform, we just need to blame Apple for not doing a very good job for a particular hardware.

 

M1 series also has some bugs - location services screws up WiFi network latencies big time. SSH is jerky and laggy as hell over WiFi for that reason. And NAA in unusable over WiFi for the same reason on M1. Apple just suggests to disable location services as a workaround. Which means that "Find my Mac" gets disabled.

 

[Miska]

10 hours ago, Tihon said:

No performance loss? If I recall correctly, someone complained about a drop in performance after switching to Ventura.

 

I didn't notice anything on that front. Only the new CoreAudio bugs they introduced for Intel platforms.

 

[Miska]

14 minutes ago, Kalpesh said:

SDM upsampled SDM : Max I got is a mere disappointing DSD 128 and that is with the LIGHT 7EC ; on the other hand I was

 

What other processing are you running?

 

This works fine for me on MacBook Pro M1Max 32GB:

Try with this integrator and rate converter!

 

Here's the load: