[rayl]

@Jawed I don't disagree with you that dollar for performance dollar, GPUs will lose out to FPGAs will lose out to ASICs...That's just the tradeoff between general purpose vs special purpose.... nor do I disagree that the current market does not support nVidia making GPUs viable for DACs (bec nVidia won't make them available in a package suitable for a DAC as the DAC market is too small for them to even care to pursue -- I think that was my initial comment).

But... if someone would to strip apart a DAC and just put the pulse array and other analog portions into a "bare bones" DAC, provide an USB driver with SDK to access it, and maybe an open source project reference implementation for the entire DSP portion -- with all that, I would maintain that there's more than enough power in a $800 nVidia card today to handle the DSP pieces to match the capabilities of BluDAVE. (* perhaps the footnote is, I have not verified the floating point precision needs, but CUDA does do 64 bit floating point.)

Admittedly, this is only of interest to folks using a PC as a source, another reason why an nVidia-based approach is not on the commercial horizon. This type of solution would appeal more to someone wanting to make exploring audio DSP an easy-to-access proposition in the hopes of fostering some general breakthrough (as opposed to for commercial gain).

Of course, this isn't something I foresee Chord doing, or any other existing audio vendor. Maybe if some enthusiast is willing to finance it personally (get Jeff Bezos interested in audio instead of building the world's baddest clock?)....

I was only reacting to your strong pushback that it's just nVidia marketing hype -- Not that we shouldn't expect a near term commercial utilization, a position with which I agree.

I didn't know whether it was really just hype or not, hence I wanted to run some basic benchmarks and seek clarification.

 

[jscmd2000]

Tried optical input to the Dave for the first time last night and was surprised and impressed.
Thanks for that info!!

 

[Jawed]

I am using a Bluesound Node 2 with optical to the Dave. The Node is connected via WiFi. I like the Node because it is compatible with Roon. I don’t have files greater than 192kHz. But what magical optimisations am I missing out on by not spending more on a streamer or posh Ethernet cable?

None.

Next step is Blu 2.

I am ready to purchase a DAVE, the only thing holding me back is the fact that the DAVE was lauched already 2.5 years ago, and that in the meantime there were other developments improving on the sound quality (like the M Scaler in the Blu2 CD-transport). I do not want to invest in Blu2 as i don't own any CDs anymore.
[...]
but the investment in DAVE is quite substantial, therefore my reservation in investing into a 2.5 years old technology.

Blu 2 + Hugo 2 costs slightly more than DAVE. It will be more musical, but perhaps not as transparent and 3-dimensional? The fact that it has a CD spinner, which you can leave turned off, is pretty much irrelevant.

It's quite clear that Chord has many higher priorities than replacing DAVE. For all we know, Blu 2 isn't good enough to extract the maximum performance from DAVE.

Now playing: The National - Guilty Party

 

[flyte3333]

No then external RFI would create currents on Dave's ground plane - it is the solid block of aluminum that screens the PCB ground plane from external RF.

Noted, got it now. So the housing is still critcal for screening airborne RF, even if only using headphones and if the USB source is battery based.

So when will we see an optical interface , standardised and commonly used, that supports up to PCM768kHz ?

Anything < 3 years coming that you can see or is that a crystal ball question?

 

[Jawed]

I don't disagree with you that dollar for performance dollar, GPUs will lose out to FPGAs will lose out to ASICs...

It's worse than that. Watt for watt a GPU is no good either. The power consumption just to turn it on is at least an order of magnitude too much.

I was only reacting to your strong pushback that it's just nVidia marketing hype

You asserted that for DSP (which is normally single algorithm), NVidia is an easy win. It's not even close as I think you now realise. Double precision, especially, is a no-go.

Sure if you're a hobbyist you can build your own upsampler with 1 million taps. I suggest if you want to try this that you use the BruteFIR source code as it appears this is actually the best-optimised FIR code anywhere. Don't bother buying a GPU: double precision is too slow compared to what you'll get on an ordinary CPU.

You still won't have Rob's windowing algorithm for your million coefficients... But it's a good learning exercise.

So when will we see an optical interface , standardised and commonly used, that supports up to PCM768kHz ?

Never, sadly.

AT&T ST connector, 50MHz optical connections were commonly used in high-end hi-fi back in the 1990s to connect transports and DACs (like my Audio Synthesis Transcend/DAX-2 - rotting in the attic). Now extinct, sadly. It's a real pity. How many devices (PCs, Macs) even have an optical output these days? The trend is towards a single connection type that does everything.

Now playing: The War on Drugs - Strangest Thing

 

[flyte3333]

Never, sadly.

AT&T ST connector, 50MHz optical connections were commonly used in high-end hi-fi back in the 1990s

Hehe noted, but I said an optical interface , not the old standard TOSLink interface...

An interesting post from John Swenson a while ago:

“OR come up with a whole new "standard" using existing optical interfaces and cabling that specifically supports source syncing to the receiver. The interesting part about this is that everything exists today to do this, very well and inexpensively. But if anybody tried to implement this what would happen is there would be huge requests to add this, add that, support this, that and the other thing, that soon it would be so complicated that we would be back to complex electronics generating noise in the DACs. If "we" the audiophile world are willing to put together a simple optical interface that just has the music without all the other stuff, it's easy to do and would work VERY well.”

The full post is here:
https://www.computeraudiophile.com/forums/topic/26139-why-not-toslink/?do=findComment&comment=486215

Would love to know if Rob sees any “standard” and common optical interface coming < 3 years from now that supports up to PCM768kHz or if it’s still a crystal ball question at this stage.

 

[rayl]

You asserted that for DSP (which is normally single algorithm), NVidia is an easy win. It's not even close as I think you now realise. Double precision, especially, is a no-go.

I admit that's fine print I did not read earlier. I didn't realize that that's how nVidia differentiates the consumer grade boards (< $1200) from the professional boards. I've only ever tinkered with a rather fancy professional board (that costs ~$4000 at the time) back in the days when I was working with speech recognition. On those, double precision was 1/2 single precision performance (limited primarily by memory bandwidth). I didn't realize that on consumer board, it is more like 1/8th (or even less)... something I just looked up now.

Yah, I guess I grew up as a software guy so would love a platform that permits experimentation for each of my hobbies, of which audio is a big one. But it's kinda purely theoretical if there isn't a good option for the final analog output piece...

 

[miketlse]

Would love to know if Rob sees any “standard” and common optical interface coming < 3 years from now that supports up to PCM768kHz or if it’s still a crystal ball question at this stage.

The creation of any 'standard' requires a cross-industry consensus (component manufacturers, software developers, consumer equipment manufacturers, etc) that there is driver for that standard.
Given that optical outputs are rapidly disappearing from phones and computers, suggests that the consensus is that there is no need for consumer optical connections any longer, because USB interfaces can be used instead.
There will be little or no commercial incentive to develop a new standard - audiophiles interested in high speed optical, probably account for 0.0001% of the consumer electronics market, so will be ignored.

 

[rayl]

The creation of any 'standard' requires a cross-industry consensus (component manufacturers, software developers, consumer equipment manufacturers, etc) that there is driver for that standard.
Given that optical outputs are rapidly disappearing from phones and computers, suggests that the consensus is that there is no need for consumer optical connections any longer, because USB interfaces can be used instead.
There will be little or no commercial incentive to develop a new standard - audiophiles interested in high speed optical, probably account for 0.0001% of the consumer electronics market, so will be ignored.

Unless one goes network based. You can run fiber instead of twisted pair to get the isolation.

 

[miketlse]

Unless one goes network based. You can run fiber instead of twisted pair to get the isolation.

???
You are now mentioning the 'physical components' of a solution - this is completely different to a 'standard' agreed by multiple actors in one or more industries.

 

[flyte3333]

Given that optical outputs are rapidly disappearing from phones and computers, suggests that the consensus is that there is no need for consumer optical connections any longer, because USB interfaces can be used instead.

In this sense, even USB will very soon be a legacy interface, if not already for new devices.

Networked interfaces seem to be the future - wireless for the mass market and maybe ethernet for audiophile gear (and hopefully maybe networked fiber?)

 

[rayl]

???
You are now mentioning the 'physical components' of a solution - this is completely different to a 'standard' agreed by multiple actors in one or more industries.

Other DACs already support network streaming... roon... etc.

 

[miketlse]

You are both completely missing the point that I was making - however it is nearly midnight here, and I can't be bothered to have a long winded debate, so goodnight to you both.

 

[Jawed]

Hehe noted, but I said an optical interface , not the TOSLink interface...

I just described an optical interface that does what you want. It's old and no longer viable!

Yah, I guess I grew up as a software guy so would love a platform that permits experimentation for each of my hobbies, of which audio is a big one. But it's kinda purely theoretical if there isn't a good option for the final analog output piece...

The project I linked will compile on Linux. That means it'll run on a Raspberry Pi. I think there's a decent chance you could get 1 million taps running in real time for 705600Hz upsampling.

You could then send the audio to Mojo, Hugo 2 or DAVE to get the pulse array goodness! But depending on your window function for the coefficients, you might only have the equivalent of 100,000 WTA taps.

It would certainly be a fun hobbyist project.

Now playing: Arca - Piel

 

[jlbrach]

the ps audio dac with bridge 2 does a wonderful job with roon and ethernet connection to your router etc....very well done and convenient and works very well....