The best DAC for under $10k is only $2k

Bit of a click-bait title but hear me out friends and I’ll get right to the answer.
RME ADI-2 DAC FS + HQPlayer Embedded

That’s right. This combo is simply spectacular! It’s now my primary signal chain.

First the parts list and the prices I paid without tax and any shipping (USD):

  • DAC RME ADI-2 DAC FS (AKM) $999
  • MB ASUS Prime Z690-P D4 $149.99
  • CPU Intel i7-12700K $324.00
  • Case be quiet! Pure Base 600 $79.92
  • PSU Thermaltake GF1 Snow $89.99
  • RAM Kingston Fury Beast RGB 16GB x2 $105.98
  • AIO Corsair iCUE H150i RGB $159.99
  • Paste ARCTIC MX-6 $9.79

Total: $1,918.66

Parts from the bin or things I already had that cost me zero:

  • Complete Pi 4 as NAA Endpoint USB to RME
  • microSD Card in USB adapter to boot Embedded
  • Ethernet cables, USB cable, monitor, keyboard, etc.
  • Another computer to burn images
  • Roon, network, internet, etc.
  • HQP Embedded License

Performance?
I can do DSD256x48 (max RME supports) without issue. Here is example:


Story:
I got interested in HQPlayer back this summer by loading Embedded in a VM on a small Ryzen 5 PRO 3350GE desktop machine. This was pointed at my Schiit Yggdrasil OG. This, obviously, was PCM only and the Ryzen did just fine. However, the changes in sound quality were not worth the cost / hassle. The Yggdrasil OG has a very excellent, to my ears, filter and I’ve always been happy with its sound.

But, I’m a huge fan of DSD and I knew my next step was HQP + a DSD DAC. I went on the hunt for an ADI-2 with AKM chip and, by almost luck, I found one new in box at a retailer who honored the RME price of $999 at the time. Great, I had my DSD DAC for $999.

This allowed me to play with the DSD modulators on the Ryzen machine and I quickly learned 2 things. First, I really liked the poly-sinc-gauss filters; especially -xla. Secondly, the Ryzen 5 did not have the horsepower to do this at DSD256. It was close, very close, at DSD128 but it would thermal throttle and fall over. I didn’t use the RME or HQP much knowing I needed more power grunt

Bring on Cyber Monday! I woke Cyber Monday and decided, if I could put a system together for under $1000, I’m pulling the trigger and building a dedicated HQP Embedded machine. Frantic research and tracking prices and my cart soon had the above combination of parts. Then it was time to sit and wait. Come to find out my ordering wasn’t perfect, that’s a different story, and some parts took a couple weeks but here we are.

I’ve been using this system for about 2 weeks listening to various formats and I’m very happy with it. I suspect it’s about maxed out at DSD256. But, really, that’s OK because my next DAC will probably be Holo May KTE @ DSD1024. That HQP box + DAC is an easy $10k. So, for now, this is the best DAC under $10k and I only spent $2k to get here.

Happy listening to me! I’m loving HQP and what it’s bringing out in the music. I’ve run just darn every filter that looks interesting and certainly have my favorites for various genre at this point. It really is a game changer and absolutely proves there will never be one filter to rule them all. Having the ability to switch the filter is the right answer. It makes the RME a total chameleon. It’s the ultimate sound quality weapon and combination and I can’t believe it’s only $2k!

Well, that’s my story. Happy Listening!

Some notes on the various parts:
The motherboard and CPU combo should let me overclock at 5Ghz or slightly higher. I’m not doing this right now and am simply using the bios settings to let it boost to 4.9Ghz. It rarely ever boosts to those frequencies and is happy siting at 3.6 most of the time.

AIO was purchased to give me better thermals if / when I overclock. The CPU runs really cool, like ~40c. I saw it hit ~70c once, ~75c once, but next time I ran the command it was below 60c both times. I’m happy with the thermals.

I’m running the RAM in the following profile:
DDR4-3000 CL16-18-18 @ 1.35V
I am testing lower latency vs higher clocks. I don’t know if this makes a difference but it’s working so great.

RGB
I didn’t want RGB but that’s what was on sale so I have some RGB now.

PSU
This is a 750 watt CPU which will allow me to add a GPU at some point. That should get me more life out of the thing. Might even let me do DSD512? I’ll need a different DAC and a GPU to test. Maybe next Cyber Monday.

Noise
I’m really impressed with the be Quiet! case. It’s certainly not something I’d want in my music room but it is significantly quieter than I was expecting. I expected a wind tunnel but it’s seriously just a low hum from across the room it’s currently in. Highly recommend these cases.

Ask me any questions here as I’m excited to respond. I’ll also be talking about system, and showing it off, tomorrow at @Rugbys Roon Talk 20:00 UTC. Look for a post in Off Topic.

11 Likes

Some various stats from this system at different bitrates:




4 Likes

You could have spend less if you are not using the DSP features and the headphone amp of the RME.

The topping E70 Velvet has the newest akm 4499ex chip and supports DSD512. costs 500 bucks.

A M1 Mac mini should be able to handle DSD256 as well :slight_smile: … if not the M2 should be out in a few months

I have 2 x ADI-2’s

But in the 20-100kHz region, my SMSL D-6 outmeasures my ADI-2 in most measurements when fed DSD512 (running in DSD Direct mode). ADI-2 is has a bit better filtering of ultrasonic noise.

D-6 has dual AK4493S chips - so cheap! With balanced outputs. Need to make sure you load the latest firmware though for best performance.

They all sound identical to me though - it’s just more pleasing to get to DSD512 in DSD Direct and without tradeoff in performance…

Having said that, my ADI-2’s will be with me forever ! It’s a classic.

1 Like

Not surprising to me and great suggestion. As you said, RME is the classic. I completely missed the ADI-2 hype and it was one of those things I wanted only after they became unobtainable (AKM version). When I found one I jumped on it immediately and that got the ball rolling. Good to know there is even a cheaper solution on the DAC side of this set-up for those just getting started.

1 Like

I wanted to run embedded with ability to add Cuda cores in future so mini wasn’t an option for me. I believe this system should get me to DSD512 with minimal upgrades if I stumble onto a DAC that supports such things.

Let the adventure continue!

1 Like

Nice post IP with a lot of good information.

I have been playing with a dedicated M1 Mac Mini as a headless HQPlayer setup.
I have it driving my Matrix Mini I pro 3 at DSD256 over USB (it won’t go higher on DOP) and my Zen Stream and DAC over the network at DSD256

It doesn’t really go higher without dropouts, but I am happy with DSD256 and the sound from both devices is great.

I have a 13th gen desktop i5 set-up that I am also going to test as well which might be an alternative, but I need to put the time in and see which I am going to buy a license for :grin:

Glad too see that you have your solution working so well though :+1:t2:

1 Like

Looking at your loads, your machine should do DSD512 with the same settings just fine. It will roughly double all the loads, but that will still leave plenty of margin.

1 Like

But only with correct firmware version! I don’t know what SMSL now ships with, but as you know it matters a lot in this respect.

1 Like

They added the fixed DSD512 firmware officially on their website

So hopefully that means it now ships properly

But indeed, DSD512 is shockingly bad without this latest firmware :grinning:

From memory, best to fix output to 44.1kHz x 512

1 Like

I have to admit I fail to see it on their site… Only DAC firmware update I can see is for “DO200 MKⅡ”.

I have to mention it to SMSL.

Even when I first reported the DSD512 issue their first response was that this was a product designed for Shenzhen Audio so I should contact them.

But once I reached SMSL engineering manager they were good to communicate with and they fixed it quickly.

So Shenzhen have the correct firmware now public here:
DOWNLOAD CENTER - LOWEST PRICE GUARANTEE – SHENZHENAUDIO

But I will mention to SMSL also to add to their website also.

I did check the firmware and measure it after I saw people mention on ASR that there was as a new firmware released end of November

1 Like

Interesting read. Do you folks know of comparable 8-channel units you would recommend?
8 channels at DSD256 would probably be unfeasible on the pc side, but maybe DSD128?

Just curious why you are a huge fan of DSD?

Isn’t all the original digitally mixed and mastered music recorded, stored and streamed in PCM?

Is it because of some DAC shortcomings in conversion of PCM or is there another reason?

Passband ripple on the best linear phase ES9038 filter is 0.002dB which seems far more than sufficient considering speaker response and room anomalies will totally dwarf that.

Curious as to what it is you hear that makes the ADI DAC sound so much better by converting the original format to DSD?

No, take a look for example at:

Most DACs contain resource constrained DSP. Best DACs allow you to bypass any DSP and instead operate bit-perfect, allowing you to do all the necessary DSP (oversampling and modulators/noise-shapers) in unconstrained external device.

2 Likes

Curious to know if native DSD recording is directly recorded in DSD (no conversion from PCM but straight from analog direct to DSD)? If so then how do they EQ the individual channels in the mix?

I agree that many DACs are not well designed - so that indeed is a good reason for external pre-processing. It begs the question why people buy these DACs. The plethora of sub optimal filtering available on certain DACs makes it worse - linear phase being the only filter that preserves highly important phase information and minimum phase being a coloration of the original.

For $10k I think you could do better. Holo May plus HQplayer would fit into that budget. But if you are actually looking for best value DAC under $10k, you might be right.

3 Likes

I follow PS Audio’s Octave records and have a couple of their Albums that were recorded and mixed purely in DSD. They have a YouTube channel that has a lot of short video’s talking about it. Quite interesting watching, as it doesn’t seem to be easy to mix in DSD and it has taken them a few years to get there.

1 Like

I can answer that, my old DAC was an RME ADI-2 DAC (AK4490 version), and I just upgraded to a Holo Audio Spring 3 KTE. Which one is the “best value” is subjective, but the Holo Audio certainly sounds much better than the RME (as it should, costing more than 3 times as much).

You can do that with my HQPlayer Pro, as well as with other HQPlayer products, but only Pro outputs files. But most of the time there’s no need to do EQ for recordings.

But for example I do headphone and room correction EQ in my playback chains also for DSD → DSD path. One reason I developed HQPlayer because I wanted all the DSP available for all the cases without conversion to some pathetic low rate PCM.

Because there are no DACs on the market that wouldn’t suffer from the mentioned problems. Especially if you don’t have unlimited funds.

Modern CPU + GPU combination can give you pretty unbeatable price-performance ratio and power of over 100 billion transistors. You are not going to easily find a DAC with comparable computational performance. When did you last time see a DAC with several GB of RAM and GHz clocks?

In addition, challenge for DACs is that they operate synchronously and cannot afford going back in time, change strategy and reprocess the data again while still meeting delivery deadline for output, for example if they see unwanted behaviour in analysis of the output data. (they typically don’t even do advanced analysis of the output data)

Analog filters are minimum-phase, you can find those in analog equipment, including all analog DAC reconstruction filters and also in loudspeaker cross-overs. With linear phase the challenge is the unnatural pre-ringing in time domain. You could consider that as coloration too.

Solution is to use DSD recordings, because then you don’t need any steep anti-alias / decimation filters and thus you avoid all the filter challenges to begin with.

44.1k and 48k sampling rates are the biggest challenges as you have variety of potential time-frequency-phase issues.

3 Likes