Taiko Audio SGM Extreme : the Crème de la Crème

[Taiko Audio]

SO QUESTION: The Taiko team has made a number of hardware and (it seems now even more) software design choices that have resulted in significant improvements in playback quality. Curious about how the choices between options are decided? Are they made strictly from subjective (individual or group) listening sessions? With the same or various systems/rooms? Or are some objective measurements also included in the decision?

We do indeed have quite a number of design projects in progress, some nearing completion and some scheduled to start. It will take us well into 2025 before we have completed all, and likely there will be new projects. Right now we have 9 hardware design projects running and/or scheduled to start in the next few months pending engineering capacity availability, which we have not disclosed anything about (and not going to on short notice either). We have finished the 5th switch design iteration which went into presale now, we will add the 4th switch design as well because we can offer it at almost half the price. The CPU direct to A or D interface card and the Router are launching in the next few months, as well as the Battery Power Supply which is more then just a Battery Power Supply as it has quite an advanced regulator section which is also employed in the "5th gen", or "Big Switch", lowering overall noise levels to below what Linear Power Supplies are capable of, we will provide more detail on all of that when the time comes. We're pulling an additional new project to the front of the queue, we actually had not scheduled capacity for yet being the design of interface cards for use with alternative interfaces, like MSB Pro ISL, other forms of I2S and probably dual AES/EBU.

Next to this we have the software projects. Edward, Wilson, Christiaan and the others some of which you've met on Discord are working on XDMS, I expect this will keep running for years. Meanwhile I'm working with 2 other software design times on remote hardware control, firmware, drivers and even new protocols we're designing from scratch for audio streaming.

Listening remains part of it, but to a lesser extend, as we are at a stage were we actually have a very good understanding of what defines sound quality and all the mechanisms affecting it. Listening is now largely validation / confirmation of new designs. Listening was much more important, even critical in earlier stages of our "evolution" when we were in full on "trial&error" mode, there is not much of that left today. In stead we are now actively engineering new designs from scratch to solve the issues we've uncovered, as we've definitely reached the limits of what is achievable with clever combining existing "off the shelf" circuit designs or building blocks.

 

[Christiaan Punter]

Hi all, here are the latest versions of the three summary PDF documents (Updated 17-01-2023)

- TACDA & TACDD DAC & Interface Summary
- Taiko Network Card & Switch Summary
- Taiko Router Summary

 

[Nils]

With my Holo May with 32 bit PGGB redbook files, more than half my music doesn’t play on XDMS (probably 80-90% doesn’t play)- is this related to the large file issue?

I’m perfectly ok with fully gapped playback, one second or not, as a start… the Holo May is just gathering dust now unfortunately.

@jelt2359 in the most recent XDMS release a change was made to the configuration/size of the RAM disk used by Extreme during playback, before the change tracks of about 6.30m in DSD256 could be played - anything longer was cut short/truncated.

After the change Kind of Blue/All Blues DSD256 11.35m plays fine.

There is always a technical limit though - where it is depends on bitdepth and sample rate and the content/track itself.

 

[microstrip]

I don't know whether or not streaming can be considered bit-perfect as the signal chain is unclear, and also, I don't have all the answers. But I guess even if it could theoretically be bit-perfect, for a given service it still might not actually be, due to any processing that may have taken place.

Well, I must say I am astonished by your answer. This seems to suggest that we are in a high level of ignorance considering streaming. Although CD was not theoretically bit perfect - level c2 correction (interpolation) changed bits, level 1 was bit perfect and early in the process research by users and reviewers showed it to be bit perfect, unless the disk was in poor conditions and/or the players was defective. This was later statistically confirmed checks performed on rips.

Qobuz sells the same files they stream - we got the information from Eliot in another thread that they stream the files exactly as they get them from the recording companies. Can't it be compared with streaming content? Can't we compare streams of the same track coming from different providers?

I think that this a fundamental aspect in streaming. If we do not get the same bits from streaming using a dCS, an Wadax or a Taiko we are exactly in the same of position as vinyl listeners - never twice the same sound.

Apologies for my insistence, IMHO we can only support something if we know the problems and the answers.

 

[MarkusBarkus]

the design of interface cards for use with alternative interfaces, like MSB Pro ISL, other forms of I2S and probably dual AES/EBU.

...two selfish, enthusiastic thumbs up on this project!

 

[jelt2359]

@jelt2359 in the most recent XDMS release a change was made to the configuration/size of the RAM disk used by Extreme during playback, before the change tracks of about 6.30m in DSD256 could be played - anything longer was cut short/truncated.

After the change Kind of Blue/All Blues DSD256 11.35m plays fine.

There is always a technical limit though - where it is depends on bitdepth and sample rate and the content/track itself.

Thanks for th info- I will try again.
Do I need to change the Ramdisk size in the extreme?

 

[Christiaan Punter]

Well, I must say I am astonished by your answer. This seems to suggest that we are in a high level of ignorance considering streaming. Although CD was not theoretically bit perfect - level c2 correction (interpolation) changed bits, level 1 was bit perfect and early in the process research by users and reviewers showed it to be bit perfect, unless the disk was in poor conditions and/or the players was defective. This was later statistically confirmed checks performed on rips.

Qobuz sells the same files they stream - we got the information from Eliot in another thread that they stream the files exactly as they get them from the recording companies. Can't it be compared with streaming content? Can't we compare streams of the same track coming from different providers?

I think that this a fundamental aspect in streaming. If we do not get the same bits from streaming using a dCS, an Wadax or a Taiko we are exactly in the same of position as vinyl listeners - never twice the same sound.

Apologies for my insistence, IMHO we can only support something if we know the problems and the answers.

I'm not suggesting anything nor saying you can't compare this to that. All I am saying is that I do not *personally* know the answer to your question. Please note that this answer is not a representation of the available knowledge at Taiko. I'm sure that Emile will have more to say on this topic. All I aimed to respond to is that bit-perfection is not the main aspect to focus on. You can have two sources that are both bit-accurate, yet different sounding. It's the same with CD rips using different software, and when comparing CD transports or music servers. Bit-accuracy certainly means a lot and one should strive for it, but it is not the complete story.

 

[Christiaan Punter]

By the way, I have compared several Qobuz downloads in WAV and FLAC versus CD rips and the same streaming content and found that, in general, Qobuz sounds really good. When comparing the FLAC download to the same streaming file, I consistently preferred the downloaded file (just slightly crisper and livelier), and also preferred the WAV version of the download over the FLAC version.

When comparing a streaming file to a CD rip, I usually preferred the CD rip, but this is tricky as you can never be sure that the source is 100% the same.

An interesting point to make here is that Emile is working on a software implementation that will greatly reduce the influence of the file handling property responsible for the audible difference between WAV and FLAC, and with that, also further reduce the gap between local and streaming.

 

[Zeotrope]

Are you saying that streaming, such as Qobuz, is not bit perfect?

No, what I mean is that moving the bits from the internet or a local HD to a DAC is fraught with issues, and why we are all Taiko customers.
But copying the bits from a CD to a local HD seems like a much easier problem to solve.

 

[cat6man]

When comparing the FLAC download to the same streaming file, I consistently preferred the downloaded file (just slightly crisper and livelier), and also preferred the WAV version of the download over the FLAC version.

This has got to be frustrating for dealers who almost all use streamers. Recently I heard some speakers with a Bartok streaming from Qobuz. When I asked to try some tracks from a USB stick, the dealer nearly fell off his chair at the difference. Once you've heard the local WAV to FLAC difference in SQ, you can't unhear it.

I've found the delta in SQ to vary with system from clearly audible (ibasso dx300 portable in the car system) to significant to OMG.
I can listen to streaming only if I don't know what the local wav file sound like. If taiko can eliminate the delta in SQ, I'll be a happy camper, (as long as there are no gaps)

 

[Steve Williams]

An interesting point to make here is that Emile is working on a software implementation that will greatly reduce the file handling property that is responsible for the audible difference between WAV and FLAC, and with that, also further reduce the gap between local and streaming.

Wow.....no flies on Emile as he just keeps moving at warp speed. Everything comes to fruition on or before 2025 which means there will be special Christmas expectations in the next few years. I am still loving XDMS. FWIW Emile shared with me years ago the very earliest prototype of XDMS. To me it was readily apparent that XDMS had the sparkle and promise of becoming something great and then to be part of the alpha test group and watch it's very development before our eyes for me has been a joy to behold but an even better joy to hear. The promise of even more toys under development in the pipe line is amazing as just the new builds every week bring something new to the table and fortunately Emile will never say his plate is full. He has a mind that is always on the go to make things better. I have no idea what is planned but it has gotten so good that I always have a smile on my face with the expectation that the best is yet to come.

 

[oldmustang]

I don't know whether or not streaming can be considered bit-perfect as the signal chain is unclear, and also, I don't have all the answers. But I guess even if it could theoretically be bit-perfect, for a given service it still might not actually be, due to any processing that may have taken place.

Indeed, bit-perfect in reference to what? As Christiaan writes, in most cases we have no idea as to the provenance of most music provided via streaming. Some services have indicated (mostly for rock titles) a particular remastering ("2019 remaster", "Steven Wilson remaster") and if one really cares they could compare a bit-perfect rip to a download of the same resolution in the case of Qobuz at least. But what would that prove? We still wouldn't know if what is streamed is the same as what is downloaded or comes on a CD.

For Tidal, with its increasing embrace of MQA is bit-perfect even a thing? And more generally, what virtue does "bit perfect" confer in and of itself?

Steve Z

 

[Tuckia]

Is an internet data stream sent "real time" like an optical read? Or is it sent as a data package with robust redundancy/error correction? Both?

Either way, I doubt whether or not a stream is "bit perfect" is the primary issue of concern for highest SQ. Nor is the file type terribly important at a certain level of equipment performance. What is important is the level of noise introduced by the data transmission chain - power supplies, processor noise, RFI, EMI, etc. - which obscures the reading/decoding process within the receiving equipment - streamers, dacs, etc. Without this noise, simple equipment could perform quite well. Bit perfect buried in a sea of undifferentiated noise, makes noise. SQ is variable throughout the day as noise levels fluctuate.

If a stream is copied locally by the end user, with robust/redundant methods, then the data should be bit perfect. This is how many have created their local stored files. Then the primary battle is dealing with local self noise - same culprits - processors, power supplies, RFI - but not multiplied as much as it is through a real time stream. I believe streaming companies forbid this type of data retention however due to copyright issues. How to catch, process, and release the data legally is the issue and subject to review by others.

If a WAV or FLAC file contains the same information, but in different forms (if both can produce bit perfect files after processing), and each file sounds different during playback, aren't you just hearing the effects of the processing equipment? So each piece of processing equipment adds its own sound, but it's not inherently a file characteristic. Results will vary.

Disclaimer: I read this on the internet while staying at a Holiday Inn Express.

 

[Nils]

Thanks for th info- I will try again.
Do I need to change the Ramdisk size in the extreme?

You only need to upgrade to most recent version of XDMS

 

[dminches]

You only need to upgrade to most recent version of XDMS

What was the old size and what is the new size of the ram disk?

 

[2ndLiner]

We do indeed have quite a number of design projects in progress, some nearing completion and some scheduled to start. It will take us well into 2025 before we have completed all, and likely there will be new projects. Right now we have 9 hardware design projects running and/or scheduled to start in the next few months pending engineering capacity availability, which we have not disclosed anything about (and not going to on short notice either). We have finished the 5th switch design iteration which went into presale now, we will add the 4th switch design as well because we can offer it at almost half the price. The CPU direct to A or D interface card and the Router are launching in the next few months, as well as the Battery Power Supply which is more then just a Battery Power Supply as it has quite an advanced regulator section which is also employed in the "5th gen", or "Big Switch", lowering overall noise levels to below what Linear Power Supplies are capable of, we will provide more detail on all of that when the time comes. We're pulling an additional new project to the front of the queue, we actually had not scheduled capacity for yet being the design of interface cards for use with alternative interfaces, like MSB Pro ISL, other forms of I2S and probably dual AES/EBU.

Next to this we have the software projects. Edward, Wilson, Christiaan and the others some of which you've met on Discord are working on XDMS, I expect this will keep running for years. Meanwhile I'm working with 2 other software design times on remote hardware control, firmware, drivers and even new protocols we're designing from scratch for audio streaming.

Listening remains part of it, but to a lesser extend, as we are at a stage were we actually have a very good understanding of what defines sound quality and all the mechanisms affecting it. Listening is now largely validation / confirmation of new designs. Listening was much more important, even critical in earlier stages of our "evolution" when we were in full on "trial&error" mode, there is not much of that left today. In stead we are now actively engineering new designs from scratch to solve the issues we've uncovered, as we've definitely reached the limits of what is achievable with clever combining existing "off the shelf" circuit designs or building blocks.

This is such bittersweet news, Emile. It's so exciting to hear of and imagine what the future holds but sad to know I may not be here to to hear it! Grateful, nonetheless, to have my seat on the Taiko Train!!!

 

[Nils]

What was the old size and what is the new size of the ram disk?

Before the upgrade to most recent XDMS, tracks < 1GB played fine but larger tracks were truncated -> I suspect/guess that old size was 1 GB.

My largest track is just under 2GB and after upgrade plays fine.

According to Windows [pic] - appears new size is 4GB.

Please correct me - @Taiko Audio...

 

[Bobvin]

This has got to be frustrating for dealers who almost all use streamers. Recently I heard some speakers with a Bartok streaming from Qobuz. When I asked to try some tracks from a USB stick, the dealer nearly fell off his chair at the difference. Once you've heard the local WAV to FLAC difference in SQ, you can't unhear it.

I've found the delta in SQ to vary with system from clearly audible (ibasso dx300 portable in the car system) to significant to OMG.
I can listen to streaming only if I don't know what the local wav file sound like. If taiko can eliminate the delta in SQ, I'll be a happy camper, (as long as there are no gaps)

A question then... if I have a CD rip to lossless FLAC, and convert to WAV via some software converter, does the newly minted WAV file sound the same as a rip directly to WAV? I have a lot of lossless FLAC files. And what do I gain and/or lose making a change besides possible sound quality improvement?

 

[cat6man]

A question then... if I have a CD rip to lossless FLAC, and convert to WAV via some software converter, does the newly minted WAV file sound the same as a rip directly to WAV? I have a lot of lossless FLAC files. And what do I gain and/or lose making a change besides possible sound quality improvement?

I haven't explored it myself, but I often rip to FLAC, edit the metadata, then convert to WAV before copying to my music drive.

I only gain SQ and lose memory space. Convenience of course varies in importance by user and is not a trivial issue. I have 16TB on my extreme and am using about 9TB right now.

I'd suggest taking ten of your favorite tracks and doing the comparison yourself. In my view, the difference is on the rough order of TAS to xdms. Of course, as Tolstoy famously said "all audiophile systems are different in their own way..."
YMMV

 

[oldmustang]

A question then... if I have a CD rip to lossless FLAC, and convert to WAV via some software converter, does the newly minted WAV file sound the same as a rip directly to WAV? I have a lot of lossless FLAC files. And what do I gain and/or lose making a change besides possible sound quality improvement?

Why not give it a try and hear for yourself?

Steve Z