HQPlayer 4 - Master Thread

Just got back to giving HQPlayer another try (current version 4.13.1). For the first time I can see running it continuously as a default for my Roon headphone sessions. My primary DAC is a Metrum Adagio (still on DAC2, waiting for DAC3 upgrade in the mail) via I2s at 192khz (shaper Gauss1, limit 20bit). Right now listening to my current SR1 chain Adagio->Ayon Spheris III->HSA-1a. Very nice and fatigue free sound for many different genres (unlike past HQPlayer experiments).

3 takeaways to share –

  1. Several of the new filters bring distinctly new possibilities over previous versions, including sinc-L, sinc-MX, and ext3. One that I had overlooked and has now become my favorite: poly-sinc-gauss-xl.

  2. Feasibility of HQP has increased significantly for me since I moved to a Mac Mini M1 (16GB RAM) on my desk, which is completely silent and sufficiently powerful to run HQP in the background (unlike Intel/AMD machines where that would have sent the fans spinning or otherwise impeded smooth system usage). Mac Mini M1 base configuration can be had for less than $600k and the form factor is brilliant (not many smaller systems around). Upcoming rumored M1X or M2 revision should stretch to higher bit rate DSD resampling as well (not sure if that wouldn’t still be asking too much from the current gen - fortunately I don’t need DSD).

  3. Read the Friendly Manual - specifically pages 16-19 (filter choices). I recently started to experiment with apodizing (apod) vs non-apod filters, this list will help you find equivalent options and more systematically determine what works best for you in your system. For the longest time I was simply overwhelmed by the many choices, investing 10 minutes to understand the documentation was foundational to being able to getting value from HQP (instead of trying recommendations from other people, which was my previous approach that got me nowhere).

3 Likes

What Intel/AMD configs have you tried, specifically?

Mostly MacBook Pro - including e.g. maxed out 2019 16" Core i9 2.4 Ghz … Even on my 32 Core 3970x Threadripper desktop (with RTX 3090 Radeon GPU), the per-core/GPU load from HPQ does lead to increased temparature/fan activity. With the Mac Mini M1 “The Beast” mostly stays off during daily productivity sessions (of course always off for critical headphone listening).

Seems like there’s something non-obvious going on.

The M1 has an advantage in efficiency but Threadripper has 5 times the multi-threaded performance.

If the M1 was at 100% cpu the Threadripper would only need to be at 20% which doesn’t seem too hard to manage.

Maybe the M1’s AI cores and/or GPU can be used to tilt the scales.

(drifting off topic) Threadripper thermals are sensitive to selective (e.g. 1-2) core utilization. HQPlayer doesn’t spread the load across 32 … think of it as “hot spots” in both utilization and overall thermals. I could tweak the fan profile to keep things quieter - back to the original observation though that Intel/AMD CPUs with workloads such as HQP are far more likely to switch to active thermal management compared to ARM/M1. My takeaway remains that M1 Mac is an incredibly compelling choice if you are looking for a HQP engine as part of your hifi setup - one of the quietest, cheapest, fastest, smallest, and best looking boxes you can buy.

Might be off-topic for you.

It’s exactly the topic to me.

It’s just one example of a case where a general purpose benchmark doesn’t give results that translate well to certain, specific, workloads.

In this case, it might well be that the HQPlayer code allows for high levels of vectorization and/or is SIMD in nature (which I would assume is the case, since HQPlayer can use GPUs … and that’s typically how they’re employed), and thus will favor units like the M1 over a high-end Intel or AMD processor.

I’ve got workloads that perform better on certain <$50 processors than they would on a $5,000 64-core Threadripper. I’ve also got workloads that require clusters of big high-core-count conventional machines and that likely wouldn’t complete at all (within my lifetime) on those same <$50 processors no matter how many of them you tried to tie together.

3 Likes

It’s useful for people who depend on HQPlayer and want to minimize the hardware footprint.

I found a few references to machine learning apps running well on the M1 so that fits in with the kind of processing that HQPlayer does.

1 Like

Do you use hqplayer in over sampling mode? If so how much OS?

I don’t have a DAC that works well with DSD; out of curiosity, I tried DSD256 (my Chord DAC will not go higher on Mac OS) and found similar CPU utilization as with corresponding PCM upsampling.

For some of my 192khz constrained (SPDIF or AES) DACs I generally run 176/192 kHz OS; one big
disturbance in the force is the very much not welcome MQA format (requiring painful HQP filter changes away from the baseline configuration). Since Tidal started converting their full catalog, I am more likely than not to just cancel the subscription – HQPlayer magnifies the deficiencies of MQA compared to RedBook CD format (e.g. yesterday I compared the 1991 original CD FLACs for Metallica’s eponymous album with the sound from the 2021 remaster in MQA – 1991 won with a large margin – both were night and day inferior to the HDTracks 96khz master).

1 Like