HQ Player

[Miska]

4 hours ago, k6davis said:

Hmmm... I was hoping for better results. 

 

Given that Apple replaced only their lowest-end CPUs with the M1, they seem to know what they are doing. All top-end is still on Intel CPUs. Now the M1 devices are positioned correctly in their product portfolio.

 

It will take some time before iMac, iMac Pro and Mac Pro will have CPUs replaced with Apple silicon. For example my 27" iMac with i9-9900K CPU (and 48 GB of RAM) has no trouble running poly-sinc-ext2 filter and ASDM7EC modulator to DSD256 output. It was fastest CPU available for iMac at the time, now I think they have updated it to i9-10900K.

 

Also my Mac Mini M1 has 16 GB of RAM which I believe is maximum you can have with M1...

 

4 hours ago, k6davis said:

in typical Apple use cases like video editing, the M1 is reportedly outperforming Intel based Macs, as these metrics would suggest

 

Video editing is typical case for GPU use instead of CPU use. Since Intel's GPUs integrated into their CPUs are really poor, any Intel Mac that has only integrated GPU will suffer on such. The ones that have the bigger discrete AMD GPUs should outperform M1 in these kind of tasks.

 

[SwissBear]

Hi Jussi,

With all due respect and without any attempt to contest your position which is not to validate the Mac Mini M1 as a viable option for HQPlayer oversampling to DSD 256, I would like to share my observations:
- I have been listening to various input formats ranging from redbook to DSD 64 with the following setting:

- I noticed a kind of “white noise” a few seconds after changing input format. When restarting the track, the restitution was exempt of any glitch for the following tracks
- I would like to precise that my Mac Mini is completely headless (no screen no keyboard), with Screen Sharing not used, and Activity Monitor not launched. Only two applications running: Roon and HQPlayer
- I tend to consider that oversampling to DSD 256 is viable under the condition that input format is stable
- Oversampling to DSD 128 is viable whatever the changes in input format…
Hope this helps users to decide for themselves.
Pierre

 

[Miska]

5 minutes ago, SwissBear said:

Hi Jussi,

With all due respect and without any attempt to contest your position which is not to validate the Mac Mini M1 as a viable option for HQPlayer oversampling to DSD 256, I would like to share my observations:
- I have been listening to various input formats ranging from redbook to DSD 64 with the following setting:

- I noticed a kind of “white noise” a few seconds after changing input format. When restarting the track, the restitution was exempt of any glitch for the following tracks
- I would like to precise that my Mac Mini is completely headless (no screen no keyboard), with Screen Sharing not used, and Activity Monitor not launched. Only two applications running: Roon and HQPlayer
- I tend to consider that oversampling to DSD 256 is viable under the condition that input format is stable
- Oversampling to DSD 128 is viable whatever the changes in input format…
Hope this helps users to decide for themselves.

 

The difference here to my settings is that "Multicore DSP" is set to checked (force-enable all parallelizations), while I use it set to grayed (auto-detection). And I have poly-sinc-ext2 also for Nx rates.

 

Tomorrow I'll check also with "Multicore DSP" checked what kind of results I get.

 

[SwissBear]

1 hour ago, Miska said:

 

The difference here to my settings is that "Multicore DSP" is set to checked (force-enable all parallelizations), while I use it set to grayed (auto-detection). And I have poly-sinc-ext2 also for Nx rates.

 

Tomorrow I'll check also with "Multicore DSP" checked what kind of results I get.

 

Also Mac Mini and NAA have fixed IP address...

[Uni]

6 hours ago, SwissBear said:

Hi Jussi,

With all due respect and without any attempt to contest your position which is not to validate the Mac Mini M1 as a viable option for HQPlayer oversampling to DSD 256, I would like to share my observations:
- I have been listening to various input formats ranging from redbook to DSD 64 with the following setting:

- I noticed a kind of “white noise” a few seconds after changing input format. When restarting the track, the restitution was exempt of any glitch for the following tracks
- I would like to precise that my Mac Mini is completely headless (no screen no keyboard), with Screen Sharing not used, and Activity Monitor not launched. Only two applications running: Roon and HQPlayer
- I tend to consider that oversampling to DSD 256 is viable under the condition that input format is stable
- Oversampling to DSD 128 is viable whatever the changes in input format…
Hope this helps users to decide for themselves.
Pierre

 

 

@SwissBear  Can you try to set SDM oversampling to poly-sinc-ext2 (not 2s) and check the performance and see if the fans kicks off?

[SwissBear]

I spent the whole evening yesterday listening to music with HQPlayer in DSD 256. This was a great experience :-)
I have to admit that I have been facing some small glitches, mostly clicks and pops, during this time. These were random, very seldom, and there were also very few periods of white noise. But compared to what other processors have produced in this field in my environment, I still consider this M1 machine as promising.
I can only dream that we will find a way to tune the system and turn unnecessary services off (https://gist.github.com/pwnsdx/1217727ca57de2dd2a372afdd7a0fc21#gistcomment-3467443), or be able to boot in a near future from an ARM 64 Linux USB device, with HQPlayer on it (https://forums.macrumors.com/threads/ubuntu-linux-virtualized-on-m1-success.2270365/).
In the meantime, as Jussi has explained, DSD 128 remains the production environment, and DSD 256 more of a teaser for future developments.

 

[Luca72c]

Anybody had the chance to test HQPlayer on new AMD 5000 cpus yet?

I find those, more than other cpus, to be really promising to push a little further computing support for HQP heavy calculations at present: they're actually in top positions about single thread cpu performance, still avoiding absurd power consumption and heat generation...

Any news about that?

Are they available for purchasing in USA at decent prices? Here in Europe they're still very hard (and overpriced) to obtain

[guiltyboxswapper]

31 minutes ago, Luca72c said:

Anybody had the chance to test HQPlayer on new AMD 5000 cpus yet?

 

I have a 5800x arriving tomorrow (couldnt get my hands on a 5900x at all).  Will try it out mid next week with HQPE.  Currently running a 3900x no problem mind.

[Luca72c]

8 minutes ago, guiltyboxswapper said:

I have a 5800x arriving tomorrow (couldnt get my hands on a 5900x at all).  Will try it out mid next week with HQPE.  Currently running a 3900x no problem mind.

 

Thanks, very good!

Are you able to run flawlessly PCM -> DSD256 using EC modulators and non-2s xtr filters on your 3900x PC?

Even if applying convolution?

[guiltyboxswapper]

3 minutes ago, Luca72c said:

Are you able to run flawlessly PCM -> DSD256 using EC modulators and non-2s xtr filters on your 3900x PC?

 

It runs for about 5-10s then hiccups for 1s.  Though bare in mind I have "quiet" cooling activated so I maybe facing throttling, and my ram is set to 3200mhz.

 

Will try that combo again, I suspect the 5800x might cut it this time.  Hopefully.

[SwissBear]

11 hours ago, Uni said:

 

@SwissBear  Can you try to set SDM oversampling to poly-sinc-ext2 (not 2s) and check the performance and see if the fans kicks off?

 

I just tried. I kind of noticed that a 400 % load of HQPlayer was the threshold when clicks and pops started. And this is what happens with such oversampling filter. I did not stay long enough to evaluate fan behavior as this was not so pleasant to listen to  :-)

[Miska]

17 hours ago, Miska said:

Tomorrow I'll check also with "Multicore DSP" checked what kind of results I get.

 

I can confirm that with "Multicore DSP" checked, both ASDM5EC and ASDM7EC work from RedBook to DSD256 for me too.

 

Seems like ARM has less cross-core penalties in this case than x64 architectures. I will adjust the automatic configuration to take this into account as well.

 

[Miska]

Some cases put the CPU load close to the maximum, but it still works and fan doesn't get too loud

 

[SwissBear]

25 minutes ago, Miska said:

Some cases put the CPU load close to the maximum, but it still works and fan doesn't get too loud

 

Did you also get cracks and pops ? White noise ? Any idea how to get rid of that ?

[Miska]

8 minutes ago, SwissBear said:

Did you also get cracks and pops ? White noise ? Any idea how to get rid of that ?

 

None, everything completely smooth. NAA is UpBoard running my NAA image and DAC is Holo Audio Spring.

 

[k6davis]

5 hours ago, Miska said:

Seems like ARM has less cross-core penalties in this case than x64 architectures. 

 

This is great news, because Apple is making machines with more cores, and soon.

 

The current M1 has 4 high performance cores and in the image Jussi posted, you can see that 4 of the cores are doing the heavy lifting. The next Apple Silicon chip, rumored to be called the M1x is due to arrive in Q1 2021 and will have 8 high performance cores. Systems with 16 and even 32 total cores are on the roadmap for later in 2021.

 

According to Bloomberg, "[Apple] said it expects to finish the transition away from Intel and to its own silicon in 2022."   

• MacRumors: Apple Working on Next-Gen Apple Silicon Chips for MacBook Pro, iMacs, and Mac Pro Due to Launch Next Year
• Bloomberg: Apple Preps Next Mac Chips With Aim to Outclass Top-End PCs
• 9to5Mac: Apple developing industry-leading CPUs with as many as 32 performance cores, targeting iMac and MacBook Pro

Of course, we'll have to see how it all plays out, but if this performance scales well, Apple could be the first to make a chip that's capable of HQP DSD512 EC. 

[Miska]

8 minutes ago, k6davis said:

This is great news, because Apple is making machines with more cores, and soon.

 

The current M1 has 4 high performance cores and in the image Jussi posted, you can see that 4 of the cores are doing the heavy lifting. The next Apple Silicon chip, rumored to be called the M1x is due to arrive in Q1 2021 and will have 8 high performance cores. Systems with 16 and even 32 total cores are on the roadmap for later in 2021.

 

Adding more cores won't help for modulators. Modulators can run either on one or two cores per channel.

 

If they cut at all on the clock speeds or single core performance to trade for more cores, HQPlayer performance will suffer. So the high performance core clocks need to be same or more, with possibly additional cores for things to improve.

 

Overall, this is very complex topic without straightforward answer. We can already see this from Intel vs AMD. Cache sizes and bus architecture (latencies), etc matter a lot.

 

[k6davis]

17 minutes ago, Miska said:

Adding more cores won't help for modulators. Modulators can run either on one or two cores per channel.

I see. I thought that your recent posts meant that modulators ran on multiple cores more efficiently on the ARM chips than on the x86. I must have misunderstood you. 

 

17 minutes ago, Miska said:

If they cut at all on the clock speeds or single core performance to trade for more cores, HQPlayer performance will suffer. So the high performance core clocks need to be same or more, with possibly additional cores for things to improve.

Yes, more cores with fast clocks is what I would like to see.

 

What I don't expect any time soon is an 8 GHz chip x86 chip from AMD or Intel. Apple is at least doing something different. So there is hope. I hope. 🙂

[Miska]

47 minutes ago, k6davis said:

I see. I thought that your recent posts meant that modulators ran on multiple cores more efficiently on the ARM chips than on the x86. I must have misunderstood you. 

 

Yes, multiple, 2 instead of 1 per channel. That is the most that would make sense. You have that option also x64 if you make "Multicore DSP" checked instead of auto. In most cases, it will make performance suffer. On this particular arm64 case it seems to improve over the other alternative. Good side is that so far there is only single CPU variant so the result is pretty straightforward.

 

[Luca72c]

1 hour ago, k6davis said:

What I don't expect any time soon is an 8 GHz chip x86 chip from AMD or Intel. Apple is at least doing something different. So there is hope. I hope. 🙂

 

To be honest, if computing power has raised relatively slowly in the last years we probably have to blame the lack of serious competition against Intel primacy. Apple itself has settled on Intel cpus for a long while, without even attempting to endanger Intel supremacy about processors. In nearly 6 years we have seen an increase in actual enthusiast cpu top performance around 2x, really not much in so long time (reference: passmark benchmarks).

We have to thank AMD effort (starting from a seriously lagging position) to break Intel dominance in the last 2/3 years, if we have now seen an increase of cpu performance of almost 4x in less than 2 years, and the pace of this performance increase is actually accelerating. What AMD has done, considering its limits in dimensions, economical power and technical potential is outstanding and very brave. And competition obviously pays off heavily for real users...

Intel conservative rest has been so insanely long and short-sighted that even ARM producers were able to reach Intel cpus' performance in last years, so giving the chance to Apple to join the competition, and that's very good too.

So at this time all 3 major contenders (but others are stepping in) have strong interest to increase cpu performance in a significant manner and i think all 3 have good chances; but i doubt any of them will have a real strong advantage over the others. Apple is just starting now (even if it has good experience about mobile cpus), so it is now a bit backwards and has honestly quite to run to reach performances of - say - AMD 5950X or Threadripper cpus or even a good optimized i9-10900K. Rocket Lake (with a 2 digits IPC increase) and AMD Zen 4 are coming out this year (Q1 and Q3), so new Apple enthusiast chips have to be very strong to seriously compete performance wise: please remember we have 16 cores going up to more than 5 Ghz on 5950X already available from AMD, with top IPC. Rocket Lake could be even better about IPC and frequencies (rumors are maybe 5,5 ghz). Both Intel and AMD have recently focused on their weaknesses and are working hard to overcome them.

Things are moving fast now, not only at Apple's... It's more than 1 year that we have cpus capable of PCM -> DSD256 using EC modulators without any lag and with sufficient spare power to run convolutions, so M1's results are rather good (considering it's a notebook cpu) but certainly not game changers and not going to push HQP hardware support much forward. Apple has frequently done things different (that's its primary marketing tool), but we have to see if the difference is beneficial for HQP or not, this time: it seems primary focuses for Apple have always been graphics/video/multimedia/ease of use, not hard floating point computation.

So first cpu to run PCM -> DSD512 using EC modulators could be Apple, but it could be Intel or AMD too. Probably whatever it should be the others will follow in a short time.

[Miska]

11 minutes ago, Luca72c said:

To be honest, if computing power has raised relatively slowly in the last years we probably have to blame the lack of serious competition against Intel primacy. Apple itself has settled on Intel cpus for a long while, without even attempting to endanger Intel supremacy about processors. In nearly 6 years we have seen an increase in actual enthusiast cpu top performance around 2x, really not much in so long time (reference: passmark benchmarks).

We have to thank AMD effort (starting from a seriously lagging position) to break Intel dominance in the last 2/3 years, if we have now seen an increase of cpu performance of almost 4x in less than 2 years, and the pace of this performance increase is actually accelerating. What AMD has done, considering its limits in dimensions, economical power and technical potential is outstanding and very brave. And competition obviously pays off heavily for real users...

 

Yes, this is very expensive and technically challenging area. Only companies with billions of $ to spend on R&D can afford to do it.

 

At the moment the competition comes from two opposite directions. Apple is scaling up from mobile to bigger CPUs while Intel equivalents are scaled down from bigger CPUs and they are now meeting in the middle. Each having their own strengths.

 

It will be interesting to see how well Apple manages to scale ARM up towards bigger workstation/server CPUs. While Intel has not been so successful scaling down into mobile (phone/tablet) CPUs. This is kind of meet-in-the-middle so far.

 

[k6davis]

7 hours ago, Luca72c said:

We have now seen an increase of cpu performance of almost 4x in less than 2 years, and the pace of this performance increase is actually accelerating.

 

Really? I use the i7-9700K, which was released just over 2 years ago. It does DSD256 ADSM7EC x48 (poly-sinc-ext2, no convolution, 2ch) with any file format. If CPU performance had increased by 4x, wouldn't we be at DSD512 EC by now? 

 

7 hours ago, Luca72c said:

please remember we have 16 cores going up to more than 5 Ghz on 5950X already available from AMD, with top IPC. Rocket Lake could be even better about IPC and frequencies (rumors are maybe 5,5 ghz)

 

Since I already have DSD256 EC, I'm only interested in upgrading (eventually) to a chip that can go beyond that. The x86 chips are unable to split the modulator workload to multiple cores per channel, so additional cores don't help with my use case. Jussi has said that x86 will need 2x clock speed to go from DSD256 EC to DSD512 EC, so clock speeds like 5.5 Ghz won't help me. Intel & AMD will eventually get there, but I'm not aware of anything on their roadmap that will improve my simple HQP use case.

 

14 hours ago, Miska said:

I can confirm that with "Multicore DSP" checked, both ASDM5EC and ASDM7EC work from RedBook to DSD256 for me too.

 

Seems like ARM has less cross-core penalties in this case than x64 architectures. I will adjust the automatic configuration to take this into account as well.

 

Apple Silicon may not be the solution either, but this is what's exciting to me. Jussi's test today showed that on the ARM chip, the modulator effectively used more than 1 core per channel. This is new. All 4 high performance cores in the chip were utilized. Will forthcoming Apple Silicon chips with 8 or 16 high performance cores offer any performance improvement? Is it possible that more than 2 cores per channel could be utilized? Perhaps not, but I'm looking forward to finding out. 

[The Computer Audiophile]

I want to setup a Raspberry Pi cluster super computer that can run HQP and handle anything such as EC modulators at DSD1028. Would be cool. 
 

OK, back to our regularly scheduled programming. 

[AudioDoctor]

23 minutes ago, The Computer Audiophile said:

I want to setup a Raspberry Pi cluster super computer that can run HQP and handle anything such as EC modulators at DSD1028. Would be cool. 
 

OK, back to our regularly scheduled programming. 

 

I can just imagine the Raspberry Pi company sitting around wondering why all their stock has been bought up and what some mad scientist might be up to now...

If you do this, wait till 2020 ends before starting. We've had enough this year.  ;-)

[Account Closed]

21 minutes ago, The Computer Audiophile said:

I want to setup a Raspberry Pi cluster super computer that can run HQP and handle anything such as EC modulators at DSD1028. Would be cool. 
 

OK, back to our regularly scheduled programming. 

DSD 1024 me thinks. Seriously, it’s pretty cool that we now have a $700 solution for DSD 256/ASDM7EC .