USB Nervosa Thread Decrapifiers, pro interfaces, and bears oh my

Not that anyone here needs further convincing, but it's nice to hear more directly from the industry, aside from @baldr , describing how shit USB is.

Start at 6:20, click on the "watch on vimeo" button. FYI, they were talking about PS Audio's bridge/ethernet interface prior.

 

Privacy settings on the video prevent us from watching it.

 

Just from watching it embedded. If you click the video, you can view it on Vimeo.

 

Sorry. I didn't realize it was not his upload....

 

I also enjoyed his rant about "spdif narrows it all down into a cluster eff..."

 

Is that the guy who came here banging on about his "credentials" and blathering about crazy audiophool tweaks that were so expensive they had to be good, though he apparently understood them like a Labrador understands a mass spectrometer? If it isn't, then he's a larval form of the exact same kind of cockrocket.

 

That must be rb2016.

 

I believe it is, indeed, the very same chap.

Now I have to make more tea ... as my keyboard is now dripping with the last one.

Yep, this guy!

...

I had a fascinating* chat with a guy I met during a work-related function the other day. He was going on about need to have de-fragmented memory in order to ensure the lowest possible jitter for USB interfaces. This is a novel concept, and wrong in every important respect. Given that there is typically four-fifths of f**k-all relationship between the apparent organization of memory (as seen from the code-side of an x86 process - which all run as virtual CPU contexts, with what appears to be linear memory, unbeknownst to them) and the actual layout in physical memory.

Quite how differences in memory-fetch timing that might result from the page-addressing and memory virtualization (I'm not talking about swapping, but the virtualization of the overall address space) can possibly effect the jitter a DAC sees on the far end of a USB connection is beyond me. There's a buffer in the source USB implementation, even at a hardware level, and there's another in the receiver, both of which operate in a default FIFO manner (so couldn't become fragmented even if that mattered). And then, of course, there's no f'ing sample-jitter on USB in the first place (that doesn't occur in/after the receiver) as it doesn't carry a sample-clock.

My first thought, beyond raw incredulity, was to ask "Do you actually understand what any of what you just said means?"

I suppressed that desire, and instead went with: "Do you know what English words mean?"

--

*In the sense that you couldn't pull yourself away in case he said something even more ridiculous than his last statement.

 

The sad thing is he'd spent so much time, energy and money perfecting his Rube Goldbergian decrapification chain (the pursuit of which I'm sure continues to this day) and was using like a Lite Dac 60.

 

Come on, guys. This is obviously the killer setup (at least according to rb2016):

...this would be a good time to summarize where I'm at as far as the Ultra+ USB chain, and maybe some guidance to others.

For Non-DIYers (that is those who need only a plug and play solution):

Stage 1a:Singxer F-1 (the heart of the chain) or SU-1 (but this is a less ideal choice without power supply mods or replacement).
Split USB cable like the LH labs 2G - or cheaper Forza Audio Twin Copper split USB. Data leg to the PC, power leg to a LiPo Battery (many cheap ones on Ebay - I had good success with a couple of Xomai 1600 mAh ones. They last for a day or two (24/7)- so then I just swapped batteries) or a cheap R-Core linear PS set to 5VDC(Ebay again less then $100 for a LT1083 version or the better $150 TeraDak R-Core DC-30W).

Comment:This alone is a killer good USB chain for not much money

Stage 1b:Add an iFi iPurifier 2 right into the F-1. Major SQ improvement here for low cost.

Stage 2: Add the Startech USB2G4LEXT2 or ICRON Ranger 2304GE-LAN. Be sure to get the 4 Port version - those extra ports will come in handy later.

This alone will improve the SQ nicely. But to get the full SQ boost you need a LPS on the REX. Suggest the ZeroZone R-Core set to 16VDC, same for a DC-30W or a MIEYAN. Why 16VDC and not 24VDC? The REX seems to operate fine with this reduced voltage - and allows adding a 15VDC (the highest voltage available) LT3045 box to lower DC noise and increase PSRR (power supply ripple rejection).

Second major SQ improvement - add a high capacity USB stick - or better a smaller capacity (due to cost) SLC mSD (SLC is a very special NAND chip design and key). You can find legitimate ones at Digikey.


https://www.amazon.com/StarTech-com...F8&qid=1493904291&sr=8-1&keywords=USB2G4LEXT2

https://www.icronshop.com/icron-brand/icron-ranger-2304ge-lan

Stage 3: Add a W4S Recovery or Uptone ISO Regen. The ISO Regen is brand new and I have not had a chance to hear one - but judging from my experience with the Amber Regen and the major design improvements to the ISO Regen - this may be another killer stepup. But it's not cheap! Look to find many good condition Recovery's on the used market and low cost as an alternative.

Note the Recovery needs between 7-9VDC for power. The ISO Regen 6-9VDC. LPS is needed (the included SMPS is OK as a temporary solution - but will not give you the SQ boost a decent R-core LPS will).

So if you have a 5VDC R-core/DC30W - the voltage can be adjusted to 8VDC say and used to power either. I have had a much more expensive LPS-1 (and it was just slightly better then a HW capped DC-30W - but that was before the nice improvement with the LT3045 8VDC board - now I imagine it would be a tough call) and it's much more expensive and will need it's own LPS for energizing (that may void it's warranty!). Or you could buy a AC line isolator and use the included SMPS energizer.

Stage 4: Add a iDefender between the F-1 and iPur2 so like this RUR>iDef>iPur2>F-1. Add a LPS to power the LEX. Move to an external SSD drive for the OS. Upgrade the SATA data cable to a Coboc flat silver shield, Akasa SATA, or Amphenol Dual line Blue Silver shielded. Other enhancements - silver teflon DC cable from the DC-30w or other LPS to device. $10 on Ebay.

I wanted to shorten the above to a simple chain diagram but found it impossible to do!

 

That's his "plug and play" solution, by the way. His DIY version is crazy!!

 

Anybody try it out?

Spoiler

 

Not just any old Lite Dac 60. His had a "upgraded cyro'd fuse holder" among other things.
https://www.head-fi.org/f/threads/lite-dac60-pcm1704-r2r-tubed-dac-mod-project.740362/

 

Maybe I can do a simplified chain:

Computer>StarTech LEX>StarTech REX>RUR>iDef>iPur2>F-1>DAC

Or:

Computer>StarTech LEX>StarTech REX>ISO Regen>iDef>iPur2>F-1>DAC

Apparently you need linear power supplies for the StarTech REX, the RUR or ISO Regen, and the Singxer F-1.

This is what I want and the audio quality should easily surpass that crap chain above:

Roon endpoint/DLNA (Ether in/AES out) > Yggdrasil

I'd be happy to run an LPS on the Roon endpoint/DLNA box.

 

While I like my Aries via AES into Yggdrasil, fronted by Roon, over the network ... the differences the Aries brings vs. say, a competent, isolated, USB -> S/PDIF bridge when dealing with a re-clocking DAC are comparatively minimal (definite diminishing returns). Stick a suitable coax-to-AES transformer on the end of that and you're pretty much going to be indistinguishable in any normal use-case (detailed auditions are different to just listening for pleasure).

Of course, the trick is finding a competent, isolated, USB -> S/PDIF bridge when you need one ...

The real key is not to use USB on the DAC-end of the chain. The benefit to removing USB further up the chain is present, but smaller. And using any form of USB -> USB "magic-box" should really be limited to when you HAVE to have USB and you're really just fixing USB-specific issues that, in most cases, would be just as well addressed with a decent quality, cleanly powered, USB hub.

 

Of course. That's why I am very happy with my Singxer F-1 feeding SPDIF coax to my Yggdrasil. It's simple and effective! As we have discussed elsewhere, I am in need of an Ethernet solution as my source material/player will soon be in a different room than my audio system. Hence the desire for something like the Aries with AES out to simplify the chain and get the best sound quality possible.

 

The memcpy guy?

 

And just to throw an added bit of head scratching into this mix…

I have been noticing, for while now, that just shutting down and re-starting my player after the total amount of used memory climbs above ≈1.5 times the amount used at startup, the SQ does noticeably improve.
And as the amount of additional memory being used continues to increase, the greater the amount the SQ increase there seems to be.
And if I take it a few steps further and restart the browser and/or my Mac, these restarts also result in a noticeable change as well.

This seems to point to memory fragmentation of some sort.
And my Mac is running on SSD's that are connected via the PCIe bus and has more than twice the amount of ram available than what is actually being used.
IOW it isn't a matter of resource scarcity nor a bottleneck in feeding the digital audio into or out of the hardware.

Now, it takes ≈ 24hrs+ of continuous use (my system stays on all the time with very few exceptions) for the memory to 'creep up', for these changes to be noticeable.
And it's usually the browser that causes the greatest amount of memory creep, even more than the player

And no these changes aren't huge, but that this happens at all, is more to the point.

JJ

 

It could "point to" any number of things that occur when a system has not been restarted for some time. The fact that you observe memory allocation increase with up-time is circumstantial.

The memory usage of a program should not "creep up" over time if it is not doing appreciably more work. Such behavior is indicative of a memory leak.

 

Yes those aspects can certainly make possible contributions.

And the browser memory usage increases alone seem to be the single greatest contributions to increased memory usage.

The thing is running a system for 24 hrs is not a long time and it alone doesn't seem to be the issue as in some cases I have run it for a much greater length of time and as long as the total memory usage creep doesn't exceed that factor of 1.5 times, it doesn't seem to matter.

What is going on is complete conjecture at this point as I have no way to track any of this down and other than these observations, which are consistent over months and months.
And so what I do now is usually just restart the player, and the browser, which takes like less than 30 seconds and things are back to being 'better'.

I don't expect anyone to solve this, but rather I figure by alerting those who do run their systems for extended periods of time, to be able to see if a quick restart does make any difference, at all.

It does work for me and perhaps it'll be noticeable/beneficial for others as well.

JJ