T+a dac 8 dsd

[tailspn]

...to pick up most of the build list for Jussi's build (except I'm going i7 6700k). It will not exactly be a silent streacom-housed kind of thing

 

Hey ted, consider this:

 

HDPLEX H5 Fanless Computer Case - HDPLEX INC

 

Maybe with this power supply:

 

HDPLEX Fanless 300W ATX Linear Power Supply with Modular ATX Output

 

It'll handle your 91W Skylake without fans, plus room for three PCIe expansion boards on a ATX MB etc.

 

Lots of pics and info here:

 

HDPLEX 2nd Gen H5 Fanless PC chassis official thread - HDPLEX User Community

 

I carry their H1-S with an Asus Q87T Thin motherboard and I7-4770S processor all around the world recording, and it serves very well. I'd use a Mac Mini, bit I need a PCIe slot for the Merging Masscore NIC card.

[ted_b]

Tom,

Nice. It adds several hundred to my build, but interesting modular external ATX linear power ideas. Not sure 300W is enough, but I need to do the math first. I;m thinking GTX960 card and fiber card (my Mellanox runs hot but might do Intel 520 instead).

[ted_b]

Not exactly rave reviews though

http://www.computeraudiophile.com/f10-music-servers/building-hdplex-300w-atx-linear-power-supply-unit-28158/

[sdolezalek]

Tom,

Nice. It adds several hundred to my build, but interesting modular external ATX linear power ideas. Not sure 300W is enough, but I need to do the math first. I;m thinking GTX960 card and fiber card (my Mellanox runs hot but might do Intel 520 instead).

 

Ted: I can't remember clearly, but I thought you were ultimately going to use this build for HQPlayer upsampling but then through an NAA. Is that correct? If so, I'd love to understand how much you (or Jussi or Eurodriver) feel that noise in the upstream processing PC gets carried downstream past the NAA? I think that Eurodriver has commented earlier in the thread that even with an NAA keeping the processing PC quiet is important.

 

I would think that in general as processing power, RAM memory and modules and GPU power (and cooling requirements) and overall system power go up, so does the noise. Depending on how that enters the data stream it presumably can't be taken out by downstream components, whether an NAA or a fiber optic bridge.

 

Don't mean to sidetrack the T+a dac 8 dsd thread, but it seems that DAC is really benefited by feeding it the best DSD512 upsampled input (probably not that different from your exasound).

[YashN]

but eq and trimming (as I stated) is not normally done in analog . It is sent to DXD (at least for almost all our labels that do eq).

 

The EQ doesn't have to be done in digital, it can be done in the analogue domain.

 

For Edits, things can be a little more convoluted but there a couple of ways to go about it, one which doesn't entail the conversion of the whole files but just the parts at Edit points, which can be a good way to proceed.

 

There was talk from Korg about a pure DSD editing software but I don't know what became of that.

 

For pure PCM, I know of a way to get that analogue mixer quality, but it is not very well known. Some people are using these processing algorithms in their work, but I believe they're mostly concerned with PCM.

 

So, most probably the majority goes the known route of doing everything in digital and PCM as it's easy.

 

What I'm saying is that it doesn't *have* to be this way as Cookie showed.

[hifial]

Ted: I can't remember clearly, but I thought you were ultimately going to use this build for HQPlayer upsampling but then through an NAA. Is that correct? If so, I'd love to understand how much you (or Jussi or Eurodriver) feel that noise in the upstream processing PC gets carried downstream past the NAA? I think that Eurodriver has commented earlier in the thread that even with an NAA keeping the processing PC quiet is important.

 

I would think that in general as processing power, RAM memory and modules and GPU power (and cooling requirements) and overall system power go up, so does the noise. Depending on how that enters the data stream it presumably can't be taken out by downstream components, whether an NAA or a fiber optic bridge.

 

Don't mean to sidetrack the T+a dac 8 dsd thread, but it seems that DAC is really benefited by feeding it the best DSD512 upsampled input (probably not that different from your exasound).

 

 

Well yes but it is all relative. I have an exaSound E20 MK III .082 Clock (up for sale here on CA) and when I had my Mac Mini power supply replaced with an Uptone MMK and used an Uptone JS-2 to power the mini it was a big improvement.

 

But the PC we used at lmitche to demo the T+A DAC 8 DSD was not heavily modified. Just as bare bones as we could but with enough processing power to do the up-sampling to DSD512. Total cost was around $600-700.

 

Now, if we went further and tricked out the PC with better this and that, yea another step up in sound improvement.

 

But the sound of the T+A with just a plain bare bones PC, No special power supply, was way above my exaSound.

[EuroDriver]

Edward, I am 24 hours away from heading to my Microcenter to pick up most of the build list for Jussi's build (except I'm going i7 6700k). It will not exactly be a silent streacom-housed kind of thing; it will be a minitower with possibly a almost-silent Noctua case fan, etc. Why do I mention this? Cuz it then requires an NAA (as it will be sequestered to my home office due to possible noise). I know you mentioned a few times that you guys have found the single pc to have better sq, when done right. I am not sure I can build a multichannel-upsample-ready power pc and keep it silent enough to be my single pc, but the NAA approach (which I currently do except my HQP i7 3770S machine died) works perfectly. But..I am still all ears if you have info I should know.

 

Hi Ted,

 

I suspect your 3770S may have died from being overworked. It does happen depending on how intensive the computation is (closed form fans please note) and how long it's been running

 

Your old PC might make a good NAA with a new CPU

 

After our DSD 512 Fest in Munich, let's talk about what we have on our overflowing shelves which might work for you for multichannel

 

Recently we have been testing out a Octocore i7, which does not cut it SQ wise for 2 channel, but might be interesting for you for multichannel. (no easy way to attach an OCXO)

 

Best

 

Edward

[EuroDriver]

Well yes but it is all relative. I have an exaSound E20 MK III .082 Clock (up for sale here on CA) and when I had my Mac Mini power supply replaced with an Uptone MMK and used an Uptone JS-2 to power the mini it was a big improvement.

 

But the PC we used at lmitche to demo the T+A DAC 8 DSD was not heavily modified. Just as bare bones as we could but with enough processing power to do the up-sampling to DSD512. Total cost was around $600-700.

 

Now, if we went further and tricked out the PC with better this and that, yea another step up in sound improvement.

 

But the sound of the T+A with just a plain bare bones PC, No special power supply, was way above my exaSound.

 

What's nice about the T+A DAC 8 DSD, is that the sound it delivers can be enjoyed at so many levels of PC tweaking and optimization. This DAC at DSD 512 reveals everything. We hear the differences between Haswell and Skylake, standard crystal, TCXO and OCXO, RAM tweaks, BIOS tuning, and OS optimizations

 

At Larry's get together the sound quality that impressed was being delivered with a BOM of around $ 700, but it was with no HDD or SSD. Larry had the machine loading the Win 10 OS from his network, very elegant and great for SQ.

 

DSD 512 seems to lower the noise floor so that all the rocks are revealed, and then you can have the fun of removing the rocks one by one :-)

[bmoura]

After our DSD 512 Fest in Munich, let's talk about what we have on our overflowing shelves which might work for you for multichannel

 

Recently we have been testing out a Octocore i7, which does not cut it SQ wise for 2 channel, but might be interesting for you for multichannel. (no easy way to attach an OCXO)

 

Do keep us up to date on Multichannel solutions for the e28 and the NADAC. I'd like to hear more about that as well.

[ted_b]

The EQ doesn't have to be done in digital, it can be done in the analogue domain.

 

For Edits, things can be a little more convoluted but there a couple of ways to go about it, one which doesn't entail the conversion of the whole files but just the parts at Edit points, which can be a good way to proceed.

 

There was talk from Korg about a pure DSD editing software but I don't know what became of that.

 

For pure PCM, I know of a way to get that analogue mixer quality, but it is not very well known. Some people are using these processing algorithms in their work, but I believe they're mostly concerned with PCM.

 

So, most probably the majority goes the known route of doing everything in digital and PCM as it's easy.

 

What I'm saying is that it doesn't *have* to be this way as Cookie showed.

 

Yes, good stuff; true all that. I was simply/mostly responding to the DXD discussion and our labels to-date. And...what will be really cool is when full DSD editing will be accomplished in DSD. Folks we've mentioned are already working on it. Let's just hope the market is there.

[ted_b]

Hi Ted,

 

I suspect your 3770S may have died from being overworked. It does happen depending on how intensive the computation is (closed form fans please note) and how long it's been running

 

Your old PC might make a good NAA with a new CPU

 

After our DSD 512 Fest in Munich, let's talk about what we have on our overflowing shelves which might work for you for multichannel

 

Recently we have been testing out a Octocore i7, which does not cut it SQ wise for 2 channel, but might be interesting for you for multichannel. (no easy way to attach an OCXO)

 

Best

 

Edward

 

I suspect you are right. It had been doing both 2 channel and multichannel upsampling/processing for a very long time, and left powered on 24/7 (dc-dc 19V pico psu into a Hynes SR7EHD monster external ps; SSSD was battery powered). My NAA is a simple Atom-based Caps Carbon with JCAT card (also WS2012 R2, AO, SSD battery powered like it's now dead cousin).

 

My 2 channel and multichannel systems are one, so I can't do something that steals sq from 2 channel. And of course, the T+A is on my immediate radar.

[bmoura]

Yes, good stuff; true all that. I was simply/mostly responding to the DXD discussion and our labels to-date. And...what will be really cool is when full DSD editing will be accomplished in DSD. Folks we've mentioned are already working on it. Let's just hope the market is there.

 

Agreed. I'm sure that some of the producers and recording engineers working with DSD today would love to see full DSD editing. Bring it on!

[chrille]

Hi Chris,

 

Here's a link to the ARDA AT1201 A/D converter data sheet, four of which are used in the Horus 8 channel input card:

 

http://www.profusionplc.com/images/data%20sheets/at1201-ds.pdf

 

This is in pdf form, so you'll have to download it to view it.

 

The block diagram shows that the actual A/D function is performed by a separate 6 bit Sigma-Delta modulator for each channel, outputting six mathematically related Pulse Density Modulated data streams at the chosen bit rate. These streams can then be reformatted, or converted to respectively DSD or PCM (DXD if 352.8KHz sampling rate is chosen). As an aside, regardless that there are on chip data reformatters and/or converters, Merging chose to do that processing using their own algorithms in a neighboring gate array processor. Horus does not natively A/D convert to DXD, but to multi-bit DSD like bit streams at the chosen bit rate, later reformatted or converted to the desired format.

 

The file size of DXD is larger than DSD due to the nature of the formats. DSD is a continuous bit stream whose bit density is proportional to the analog level modulating the bit clock. There's no redundancy, but neither is there a digital (processible in a digital computer) value represented. DXD (PCM) is like motion picture film, where each frame (sample) is a complete stand alone digital word (2's compliment binary word of X bits depth/resolution) measuring the analog value at that sample time. One of its characteristics is 90+% of the value (the bits making the digital word) of the binary word data is redundant to its neighboring samples. PCM is an EXTREMELY inefficient coding process, but it does yield data that is processible in a digital computer.

 

Most labels, save one diehard, do not record in DXD, if they can avoid it. It makes no sense to do the conversion with the limited processing power on the card, when it can be done later using all the resources and flexibility of the host computer if required for post processing. The advantage of recording in DXD is a greater number of analog channels can be accommodated in the record process. You'd be amazed of the number of recordings made with 100 channels, or more.

 

Hope this helps,

 

Tom

 

Thanks a lot Tom,

I won't pretend though that I understand all of the information in the pdf.

Nor that I grasp all of what you are saying in this post.

 

And not having heard any native DSD 256 recordings at all so far apart from downconverted to DSD 128 via Hugo and DAVE, who am I to say which format comes closest to the real thing?

For me the music and performance come first and I enjoy well recorded albums of acoustic music in both DSD and PCM formats. But I have to say that DXD via DAVE was some of the most natural and realistic digitallly recorded reproduction of music I have heard, very impressive indeed.

My non technical brain is still a bit puzzled by the fact that DXD files are bigger than even DSD 256 obviously.

And if I read the pdf numbers correctly it seems to me as if the multibit 24/384 has a dynamic range of 124 dB, whereas the DSD output drops 10 dB?

Some "digital experts" equate dynamic range with resolution. And if such a stand holds true wouldn't 10 dB more dynamic range from mutibit also mean higher resolution?

 

Regarding the 100 mics, I haven't seen that many in use. But often far too many. As far as I am concerned I would much prefer a return to the old simply mic'd two/three mics for stereo and five for surround.

Two mics yielded excellent , still admirable stereo results already in the late 1950s but has basically been abandoned by everyone since those last truly stunning direct cut LPs by Doug Sax of Sheffield Labs.

I would really like to hear that same approach from another "Sax" with a slightly different spelling using the highest possible resolution recording rate.

Not that I don't like his DSD 64 recordings, I love most of them both as recordings and musically. But I do suspect that higher res than DSD 64 with its inherent noise problems combined with even simpler miking would yield even more realistic results.

I am of course aware of the better safe than sorry and economical aspects involved. But why not please those headphone freaks among us with at least a dummy head take of things!

I absolutely love the few examples of that approach from native DSD.com

Both the Beethoven Heiliger Gedankensang especially the DSD 128 version and the two Brahms quartets are wonderful in their coherence and beauty.

Cheers from "Dummyhead Chris" currently next to the fireplace in an old English bungalow in chilly Cameroon Higlands Malaysia after a weekend of wonderful live concerts in steaming hot KL.

[ted_b]

Ted: I can't remember clearly, but I thought you were ultimately going to use this build for HQPlayer upsampling but then through an NAA. Is that correct? If so, I'd love to understand how much you (or Jussi or Eurodriver) feel that noise in the upstream processing PC gets carried downstream past the NAA? I think that Eurodriver has commented earlier in the thread that even with an NAA keeping the processing PC quiet is important.

 

I would think that in general as processing power, RAM memory and modules and GPU power (and cooling requirements) and overall system power go up, so does the noise. Depending on how that enters the data stream it presumably can't be taken out by downstream components, whether an NAA or a fiber optic bridge.

 

Don't mean to sidetrack the T+a dac 8 dsd thread, but it seems that DAC is really benefited by feeding it the best DSD512 upsampled input (probably not that different from your exasound).

 

Answered here, so as not to hijack any further

http://www.computeraudiophile.com/f11-software/design-pc-server-roon-and-hq-player-25796/index5.html#post531477

[tailspn]

...still a bit puzzled by the fact that DXD files are bigger than even DSD 256 obviously.

And if I read the pdf numbers correctly it seems to me as if the multibit 24/384 has a dynamic range of 124 dB, whereas the DSD output drops 10 dB?

 

Hi Chris,

 

I apologize in advance to this thread originator for these off topic posts, and this one will be my last here.

 

Regarding DXD vs. DSD file sizes, think of the motion picture film analogy I mentioned earlier. With two adjacent frames of the motion picture film or file, how much "change" information/data is in the adjacent frames verses constant information/data? 10%, 1%? That constant or non changing information/data is carried along through adjacent frames, or in the case of PCM the adjacent samples in the data words. That's allot of redundant data! DSD, regardless of bit rate, is only representing the change information/data, so it's obviously much more compact and efficient as a recording and transmission media.

 

Please notice I didn't say 100 mics, but 100 or more channels. A typical studio recording may have 2 to 30+ mics, but hundreds of tracks representing many takes. When mixing in post processing all of these tracks need to be represented and processed at the same time. That's where PCM shines, and the lower the sample rate, the greater the number of tracks can be handled. Of course this has little to do with classical music acoustic recording, but you'd be surprised how much post processing is even done there.

 

But I agree completely with you about minimal mic recording. I dabbled with it many years ago in stereo, and was never satisfied with the results, despite the best Schoeps mics and Ampex professional tape machines. That's when I sold it all, and bought an airplane. Surround sound has changed all that for me, and finally allows truly realistic recordings to be made using minimal mic techniques. We'll be offering a recent Mahler 3 recording made at 256fs DSD with five DPA mics on NativeDSD in the hopefully near future, but only in surround. All the spacial reality and magic collapses when mixed to stereo. Still great sounding though

 

Enjoy your stay in Malaysia Chris,

 

Tom

[guiltyboxswapper]

How can running closed form algorithms destroy a cpu? Do we even know for sure it's the cpu that's dead? Would love miska's input on this :-D

[guiltyboxswapper]

I'm glad someone else said it, you don't need a super tricked out pc to make this work it's best. Besides, if the pc is really that noisy and difficult to stabilise, why not use something else that's designed for the job of audio performance first?

[mansr]

Regarding DXD vs. DSD file sizes, think of the motion picture film analogy I mentioned earlier. With two adjacent frames of the motion picture film or file, how much "change" information/data is in the adjacent frames verses constant information/data? 10%, 1%? That constant or non changing information/data is carried along through adjacent frames, or in the case of PCM the adjacent samples in the data words. That's allot of redundant data! DSD, regardless of bit rate, is only representing the change information/data, so it's obviously much more compact and efficient as a recording and transmission media.

 

I think you're oversimplifying a bit. While DSD, in a sense, records only changes, the amount of change encoded in each sample is tiny so a whole lot of samples are needed to represent a large change. That's why DSD64 is the bare minimum which achieves only slightly better than CD resolution despite being 4x larger. PCM at 88.2/24 is by any reasonable measure better than DSD64 while needing 25% less space.

[matthias]

My non technical brain is still a bit puzzled by the fact that DXD files are bigger than even DSD 256 obviously.

 

AFAIK,

 

DXD 24/352.8 is 8.4672Mbit/s

 

and

 

DSD256 is 11.2896 Mbit/s

 

so

 

DSD256 files are bigger than DXD files?

 

Matt

[tailspn]

...the amount of change encoded in each sample is tiny so a whole lot of samples are needed to represent a large change.

 

Well, they're not samples as they are in PCM, just the presence or absence of bits dependent on the level of modulation, but otherwise you're correct.

 

The actual size of a 256fs DSD file is about 95% the size of the same file in 32bit DXD. A recent album I worked on was 8.488GB in stereo DXD 32 bit, and the DSD was 8.09GB. The DSD would be correspondingly larger if compared to 24bit DXD, but that's not the format we deal with.

 

Specs aside, there's a big subjective difference in sound quality between an original 256fs DSD recording, and its 352,8KHz PCM DXD conversion. At 88.2KHz PCM, IMO, the difference is even greater, in favor of the DSD. Converting down to 64fs DSD does lose spatial resolution, but not to the degree of a 4X subsampling of 352.8KHz PCM to 88.2KHz.

 

As always, YMMV

[craighartley]

[. We'll be offering a recent Mahler 3 recording made at 256fs DSD with five DPA mics on NativeDSD in the hopefully near future, but only in surround. All the spacial reality and magic collapses when mixed to stereo. Still great sounding though

 

Tom

 

Tom,

Do I remember right you were also doing this binaural?

Cheers

Craig

[EuroDriver]

How can running closed form algorithms destroy a cpu? Do we even know for sure it's the cpu that's dead? Would love miska's input on this :-D

 

One of our team members has extensive experience with a University computing center server farm. When you are doing intensive computing, the CPU's don't last for ever, especially if the cooling is not adequate and the individual core temperatures get too high. RAM also does not last for ever.

[EuroDriver]

I'm glad someone else said it, you don't need a super tricked out pc to make this work it's best. Besides, if the pc is really that noisy and difficult to stabilise, why not use something else that's designed for the job of audio performance first?

 

Quite correct, you can get super results with $ 700 worth of parts + case. We did that 2 weeks ago at Larry's

 

But when you have a mobo power supply with microvolt ripple, add a OCXO with its own power supply, a dedicated power supply for the SSD, select the best sounding RAM, tune the RAM settings, build a optimized OS for the software that's running, the sound quality rises to another level

 

That's what we have been doing since July 2015, and will be announcing is 3 weeks time :-)

[guiltyboxswapper]

One of our team members has extensive experience with a University computing center server farm. When you are doing intensive computing, the CPU's don't last for ever, especially if the cooling is not adequate and the individual core temperatures get too high. RAM also does not last for ever.

 

But that's not the fault of an algorithm (in case of insufficient thermal cooling), that's a poorly engineered system. Ofcourse parts don't last forever, but its rare (easily <1%) even if the CPU is loaded 100% of the day. CPUs are designed to withstand long loads, when in a sufficiently cooled system. Closed-form won't cause premature death...

 

(and we still dont know if the CPU or something else is dead!)

[guiltyboxswapper]

That's what we have been doing since July 2015, and will be announcing is 3 weeks time :-)

 

But you don't actually understand why your system sounds good, even if it may sound better than average. Good luck if anything goes wrong with one...

 

Oh and it's still a PC running Windows, not an appliance. That'll be a tough sell. I'd have been interested if it was an actual appliance that you turn on and forget about. Linux is far easier to turn into a 'forget-about-it' appliance, as well as hyper-tune to your hearts content. Missed opportunity IMHO.