My deep dive into media storage interfaces: Musical differences heard between chipsets, Firewire 400/800, USB, SATA, flash drives, SD cards, and network shares (Warning: may cause seizures in the DBT crowd and flat earth naysayers)

[sandyk]

I am curious to read what happens when Alex powers up using that Linear Power Supply. (wink)

[acg]

Anthony (acg): Nice to meet you, and thank you (and everyone else here) for your insightful contributions. I think is is fascinating that you probed all the voltage rails realtime! Indeed, my friend, the esteemed Mr. Swenson is getting interested in putting both logic and RF probes into a server running different software and buffering techniques.

 

[snip]

 

Thoughts, suggestions? Counterpoints to my notion that speed is not the factor in SQ?

 

Alex, nice to virtual-meet you too. The real reason that I have measured all of the rails in my XXHE PC is so that I can design a full ATX linear psu, which is well and truly underway with a lot of help from my friends. For those that are interested I have solved the ATX startup procedure so will hopefully have the prototype PCB ready before Christmas to send around to my friends and make sure it starts up all their computers. Anyway, that is why I have measured power draw, and I did notice a great deal about how the XXHE software managed computer power whilst I was at it.

 

Alex, I agree with you that the bulk of the gain is likely to be because of reduced electrical noise within the computer...the ram has to run anyway, there is no way around that, so if we can eliminate sata or firewire or usb or whatever else then hopefully we can get a useful positive result. I have been intrigued by a CF card for the o/s and may trial that after my linear ATX psu project is complete. A bit like your SD card, the CF card is slow and steady but simple and low noise.

 

Mayhem, I nearly fell off my chair when you mentioned doing some listening impressions...wow...and they say that pigs don't fly. And no I'm not calling you a pig...you get what I mean...it's just not what I expected. Good stuff!

 

Anthony

[sandyk]

Going all analog path after the DAC on this one Alex with my fully passive 2.5 ways, subs off. The 2.5's feature 2 180mm midwoofers and do a nice flat in room response to 40hz with an F3 around 34hz or so. While these aren't my favorite speakers, they are IMO the most revealing and the flattest anechoic response in my collection.

 

Mayhem13

It seems that you are taking this very seriously. I believe the biggest sole contribution to the SQ of either analogue or digital, is how good the power supply is. A good low Z PSU will also help to remove interaction with other areas of the device. I would also expect both SSD and HDD to sound quite similar if the PSU area is good enough.

As I said in my previous post, I look forward to Alex's report after swapping to a Linear PSU that they have under development.

Regards

Alex

[Superdad]

Ok so I picked up two sticks of 4gb Crucial ram to go with the two I already have for a total of 16gb. I'm fortunate to have two identical windows machines my son and I put together last Christmas with all hardware and OS being the same and both of us run iTunes for our libraries. I'm adding and configuring the BIOS as i type this for some extensive AB listening tomorrow night. Both machines will be connected to my DAC switching inputs to my main system.

 

I 'WANT' to be able to discern a difference between RAMdisk playback and from a standard 7200 HDD so against my usual concerns, I'll be using music familiar to me. I'll try standard Redbook first and HiRes PCM with each against opposing machines. I won't be taking any standard measurements but instead, reporting my findings purely subjectively. That being said, and wanting to hear a difference, my son will be at the helm doing the switching so I won't be aware of which is RAMdisk and which is HDD playback sampling the same tracks of the same resolution across the same interface of one DAC, eliminating as many variables as physically possible. Should have some viable results to report back by the weekend. Stay tuned as subjective reports aren't something I do often, if ever.....but Alex has my interest peaked here.

 

Hi Mayhem,

Glad that this exercise has gotten your interest. I do have a question though, given that you are trying to reduce variables: Why the need to set up two machines? Sure is hard to make them perfectly identical (unless one does a clean OS install and exactly the same things on both).

Why not just make a RAM disk, copy the test tracks to it, then alternate between directing your player to play from it or from the hard drive?

 

With my set up I have 6 Audirvana Playlists on my desktop, each named and populated with tracks located on each source medium (with A+ the playlist looks for the file at whatever location it was dragged from). So I have the same tracks on: SD card, USB thumb drive, internal SATA, RAM disk, external FW400, and a shared directory on my LAN. I can stay in A+ and within 6 seconds load a playlist from any of the media types. Of course for best results I have unplugged things like the USB thumb drive and/or the FW400 drive when comparing among the better ranked locations (not that there was a lot of difference). I suppose I could eject the SD card, but it's pretty passive, and the internal and the LAN are there regardless.

 

What software player and DAC are you using?

 

Have fun,

ALEX

[Superdad]

Alex, nice to virtual-meet you too. The real reason that I have measured all of the rails in my XXHE PC is so that I can design a full ATX linear psu, which is well and truly underway with a lot of help from my friends. For those that are interested I have solved the ATX startup procedure so will hopefully have the prototype PCB ready before Christmas to send around to my friends and make sure it starts up all their computers. Anyway, that is why I have measured power draw, and I did notice a great deal about how the XXHE software managed computer power whilst I was at it.

 

Alex, I agree with you that the bulk of the gain is likely to be because of reduced electrical noise within the computer...the ram has to run anyway, there is no way around that, so if we can eliminate sata or firewire or usb or whatever else then hopefully we can get a useful positive result. I have been intrigued by a CF card for the o/s and may trial that after my linear ATX psu project is complete. A bit like your SD card, the CF card is slow and steady but simple and low noise.

 

Wow Anthony, a big multi-rail ATX supply with start-up logic? That sounds like an ambitious project! I look forward to pictures.

 

If you try a Compact Flash card, via what interface will you connect the reader? Hope it is not USB! Do you have a motherboard with a CF slot? I have seen some embedded processor boards with SD card slots--specifically for booting. So if you can do some solid state storage that does not require a whole interface with protocols--just simple 1-bit transfer like SD or CF that would be great. Anything external would seem to fall short of that.

 

Obviously for real music library storage, flash memory is not practical yet (though they do keep promising 1TB SD cards--for $$$). I now can see the point of of SoTM's SATA drive filter, though I can't comment on its effectiveness. The computer side is still running the SATA bus, so whatever impact that has…

 

Right now I don't mind creating a couple GB RAM disk and copying what I plan to listen to to it. My sessions are not that long these days, and the bump in SQ is certainly worth it.

 

I am going to make a bootable Mavericks volume with my 32GB SD card tomorrow. We'll see how that works out.

 

I noticed that Roch (elcorso) said he is initially not fond of the RAM playback--fining a bit to forward or harsh. I have and greatly enjoy the track he referenced, and yes it can be a bit overwhelming. The fact is, as our systems become more revealing, a lot of recordings can get their flaws exposed. That's not news. But as long as whatever irritating character one hears is not prevalent on all tracks, then I think we are just hearing more of the facts of the recording.

 

To be honest, these past two weeks I have not had time to ventured far past my test tracks and some other favorite music I have that is well recorded. I'm almost a bit frightened to dive into my general collection. I know I'll be thrilled with some and be cringing with others.

 

Ooh, it's late. Time to put kids to bed.

 

Goodnight and good luck.

[mayhem13]

Hi Mayhem,

Glad that this exercise has gotten your interest. I do have a question though, given that you are trying to reduce variables: Why the need to set up two machines? Sure is hard to make them perfectly identical (unless one does a clean OS install and exactly the same things on both).

Why not just make a RAM disk, copy the test tracks to it, then alternate between directing your player to play from it or from the hard drive?

 

 

What software player and DAC are you using?

 

Have fun,

ALEX

 

With two machines and the DAC switching, I can reduce or eliminate any lag when switching sources, and by properly syncing the files to play at the same time familiar segments of the file will be available for direct comparison. With someone else doing the switching, bias will be somewhat removed.

 

As to you statement about the need for clean OS, my logic can't take me to a place to consider inactive OS operations as a cause for SQ variations. If we were talking machines that were underpowered and possibly background operations were slowing things down, I might consider it.....but I wouldn't own or listen to music from a machine like that anyways and these machines are clean i5's with very low ripple PSUs and no video cards. Both also boot from identical SSDs so the whole controller controversy is eliminated as well. Because we have these machines is probobly the only reason why I considered performing this AB test and talking about it here......the conditions don't get much better from an objective standpoint really. I do quite a bit of AB testing at home when designing speaker systems so the premise is something I'm very familiar with. If the conditions won't allow for a viable result, why even bother.

 

Software player is good ol iTunes as IMO boutique software players are nothing more than algorithmic DSP......sorta rebels without a cause of equations without variables. As to the DAC, it's a modded Firewire interface originally created for digital mastering. As a side note, both PCs will be plugged into the same Lab grade power conditioner.

 

Today is a full day for me at work, so no listening until 8pm or so.....I need some memory buffer to forget the days activities and clear my head before trying to be critical of anything else.......works well for me in my personal life too!

 

On a side note, I might not even post my results straight away. There's still some things that I'm not sure of such as how iTunes handles the file for playback and how it natively assigns virtual memory for the file before playback which is a huge bias for me. As I mentioned earlier, my logic is I since iTunes is already buffering via virtual memory in RAM, what's the value in a RAMdisk?..............and that's what I'm hoping to find out. This thread intrigued me from the get go as faster processing with less mechanical moving parts 'should' produce the finest digital output possible......but this is relatively elementary stuff with today's tech. When you posted improved results using SD cards and RAMdisk, I felt a need to explore the possibilities. This also opens up a new area of CA for me with a server machine using a library of internal mSATAs. Such a machine with a power sipping mobile CPU could run in a very small fanless enclosure making for one hell of an AV server.

[sandyk]

With two machines and the DAC switching, I can reduce or eliminate any lag when switching sources, and by properly syncing the files to play at the same time familiar segments of the file will be available for direct comparison.

 

Barry D. can do that with his work PC using special Mastering software . Both tracks don't run together, but the other one picks up seamlessly where the other left off.

[Jud]

With two machines and the DAC switching, I can reduce or eliminate any lag when switching sources, and by properly syncing the files to play at the same time familiar segments of the file will be available for direct comparison. With someone else doing the switching, bias will be somewhat removed.

 

As to you statement about the need for clean OS, my logic can't take me to a place to consider inactive OS operations as a cause for SQ variations. If we were talking machines that were underpowered and possibly background operations were slowing things down, I might consider it.....but I wouldn't own or listen to music from a machine like that anyways and these machines are clean i5's with very low ripple PSUs and no video cards. Both also boot from identical SSDs so the whole controller controversy is eliminated as well. Because we have these machines is probobly the only reason why I considered performing this AB test and talking about it here......the conditions don't get much better from an objective standpoint really. I do quite a bit of AB testing at home when designing speaker systems so the premise is something I'm very familiar with. If the conditions won't allow for a viable result, why even bother.

 

Software player is good ol iTunes as IMO boutique software players are nothing more than algorithmic DSP......sorta rebels without a cause of equations without variables. As to the DAC, it's a modded Firewire interface originally created for digital mastering. As a side note, both PCs will be plugged into the same Lab grade power conditioner.

 

Today is a full day for me at work, so no listening until 8pm or so.....I need some memory buffer to forget the days activities and clear my head before trying to be critical of anything else.......works well for me in my personal life too!

 

On a side note, I might not even post my results straight away. There's still some things that I'm not sure of such as how iTunes handles the file for playback and how it natively assigns virtual memory for the file before playback which is a huge bias for me. As I mentioned earlier, my logic is I since iTunes is already buffering via virtual memory in RAM, what's the value in a RAMdisk?..............and that's what I'm hoping to find out. This thread intrigued me from the get go as faster processing with less mechanical moving parts 'should' produce the finest digital output possible......but this is relatively elementary stuff with today's tech. When you posted improved results using SD cards and RAMdisk, I felt a need to explore the possibilities. This also opens up a new area of CA for me with a server machine using a library of internal mSATAs. Such a machine with a power sipping mobile CPU could run in a very small fanless enclosure making for one hell of an AV server.

 

If you *really* would like to make this a total PITA for yourself...

 

- You could try installing XXHE. It's a memory player. In fact with the Playback Drive setting you could install XXHE in two different directories, with one set to play back from RAMdisk, the other from regular disk. XXHE can be set to play back a limited piece of a track, which is nice for A<->B purposes. And if you did, I could suggest to you some other XXHE settings you might try that Could Not Possibly Have Any Earthly Effect On The Sound, and see whether you hear the impossible.

[mayhem13]

Yes I'm sure Alex and I can do it too....but that's an added software environment with unknown processes......we're after viable results from a stable test platform where the differences may be so insignificant as to go undetected or not present at all. Stitching file segments together without a gap would not present a set of samples for comparison. One could argue that it would be the task of the listener to isolate and identify the passages within the song......good luck wit alla dat!

[Jud]

Yes I'm sure Alex and I can do it too....but that's an added software environment with unknown processes......we're after viable results from a stable test platform where the differences may be so insignificant as to go undetected or not present at all. Stitching file segments together without a gap would not present a set of samples for comparison. One could argue that it would be the task of the listener to isolate and identify the passages within the song......good luck wit alla dat!

 

Yep, the reluctance to go with something unfamiliar when trying to hear differences that may be subtle (or nonexistent) is understandable. (You might think about trying it after you do the experiment with iTunes, to see whether different software brings different results. OTOH, getting XXHE installed and running for a first time user may not be trivial, and this is supposed to be interesting and fun; so totally up to you.)

 

To be clear, I was talking about setting up XXHE to play the same part of the same track in both cases, rather than playing sequential pieces of a track.

[mayhem13]

If you *really* would like to make this a total PITA for yourself...

 

- You could try installing XXHE. It's a memory player. In fact with the Playback Drive setting you could install XXHE in two different directories, with one set to play back from RAMdisk, the other from regular disk. XXHE can be set to play back a limited piece of a track, which is nice for A<->B purposes. And if you did, I could suggest to you some other XXHE settings you might try that Could Not Possibly Have Any Earthly Effect On The Sound, and see whether you hear the impossible.

 

I considered it.....but since there's soooooooo many variables in Pete'sware, I'm concerned I'll wind up farther from where I started. If iTunes is such the crappy environment that many claim it is ( I don't), then differences in SQ 'should' be more apparent.

 

Notice I chose to use the word difference ( trying to stay objective).......as Alex trends towards improved.........a very very gray area where tastes and personal preference are concerned....there's truly no reference available. I love it when guys start gushing about how a track sounded just like so and so's violin live at such and such......what complete and utter nonsense trying to develop an auditory comparison from a live event as a member of the audience. That's about as far from obtainable reference as possible and IMO useless to this hobby. Try having a steak at home and that same steak seated at Peter Luger's.........wanna guess which is gonna taste better?

 

Let me iterate that this isn't a criticism of Alex (SuperDads) findings or method, contrary.....I wish I could feel that way about gear sometimes but alas......the objective side of my brain always takes the jelly outta my donut!.....and from a comparison standpoint, a 'difference' might be more telling than an improvement for some of our fellow members.

[mayhem13]

. (You might think about trying it after you do the experiment with iTunes, to see whether different software brings different results. .

 

I have tried some software players Jud when I first joined CA. Before posting, I tried quite a few of the suggested players and other tweeks. When I started posting afterwards if you remember with the earlier 'regime'.....well ...it wasn't pretty and I got sent to my room a few times ( this situation has improved so hats of to Chris for lightening the hand a bit).

 

My 'conclusion' to the players was YES...there is a difference, BUT was it better? For me the answer was 'depends'.......and I concluded that the conditional pivot would be the room. I could hear what was going on was primarily algorithmic DSP and as such came to the conclusion that some would benefit depending in their listening environment and tastes while others might not be impressed......including myself. My logic says that I can treat my space physically in a much more useful and controlled manner than an algorithm treating a file.

 

Many here like to talk about the time domain as the major influence in how we hear or perceive music in a dimensional space......and I'll agree.......as long as those can define this time domain which IMO those that practice this technique have yet to do. Us speaker designers work with something a bit more tangeable we like to call power response which we find very useful and measurable based on sound acoustic principles. In essence, we're describing the same thing but the time bandits make more assumptions that explain a lot less....all unsubstantiated or without a viable measure.

[VandyMan]

I'd expect a software player with a properly implemented buffer to work better than a RAM disk because a RAM disk will have driver software between it and the application.

[Daudio]

a RAM disk will have driver software

 

Yep, it has to translate a HD address (volume/cylinder/track/sector) to a memory address (offset/bank/whatever). I wonder how simple or complex that process is ? It should be very simple, but could be complex if it has to traverse many layers of OS and driver overhead to get to a straightforward algorithm.

[VandyMan]

It is not very complex. I recall a RAM disk driver for MacOS 7 was only a few K. Still, if max performance with lowest latency is your goal, an application buffer is probably the way to go. Also, RAM disks are not very memory efficient unless they use compression (which many do) because they use track/sector approach and can become fragmented just like a physical disk.

[sandyk]

and from a comparison standpoint, a 'difference' might be more telling than an improvement for some of our fellow members.

 

A good point.

I was reminded of this yesterday when searching through my "archives" I rediscovered a copy of the track "Feelin' the Same Way-Norah Jones" that was ripped on a Mac by "Silverlight" (Geoff) from NYC back in 2009.

It sounds noticeably different to the same track that I ripped on my PC in Australia a couple of days later.

This example is unusual in that it was ripped from different copies of the same hybrid disc, using different types of computer on 2 different continents. Yes, the checksums ARE the same, and Geoff stated at the time that my rip sounded better than his.

Alex

[acg]

It is not very complex. I recall a RAM disk driver for MacOS 7 was only a few K. Still, if max performance with lowest latency is your goal, an application buffer is probably the way to go. Also, RAM disks are not very memory efficient unless they use compression (which many do) because they use track/sector approach and can become fragmented just like a physical disk.

 

XXHighEnd gets around this by loading the _entire_ playlist onto the playback disk (in this case a ramdisk) which is organised as contiguous memory.

 

EDIT: XXHE also reformats the Playback Drive (Ramdisk) each time a new playlist is played.

[acg]

Wow Anthony, a big multi-rail ATX supply with start-up logic? That sounds like an ambitious project! I look forward to pictures.

 

 

There will be pictures.

 

If you try a Compact Flash card, via what interface will you connect the reader? Hope it is not USB! Do you have a motherboard with a CF slot? I have seen some embedded processor boards with SD card slots--specifically for booting. So if you can do some solid state storage that does not require a whole interface with protocols--just simple 1-bit transfer like SD or CF that would be great. Anything external would seem to fall short of that.

 

CF via motherboard sata connection I think would be the best best, especially given your impressions in this thread.

 

I noticed that Roch (elcorso) said he is initially not fond of the RAM playback--fining a bit to forward or harsh. I have and greatly enjoy the track he referenced, and yes it can be a bit overwhelming. The fact is, as our systems become more revealing, a lot of recordings can get their flaws exposed. That's not news. But as long as whatever irritating character one hears is not prevalent on all tracks, then I think we are just hearing more of the facts of the recording.

 

I feel that a change like a ramdisk is best appreciated over a number of days or weeks on as much music as possible. Like you say Alex, if you can detect a flavour or colouring or lack thereof, then you are better situated to know if it works for you. The ramdisk works for me, but it may not work for others depending on their preconceived ideas of how they want their music to sound or what works in their system.

 

PeterSt has some interesting ideas about noise in the digital system which basically goes that sometimes it is best to keep 'nice noise' than to remove it and reveal the 'horrible noise' it masks.

 

Anthony

[sandyk]

PeterSt has some interesting ideas about noise in the digital system which basically goes that sometimes it is best to keep 'nice noise' than to remove it and reveal the 'horrible noise' it masks.

 

Anthony

Like a Sonar operator hearing another , perhaps unfriendly submarine "pinging" you ?

Guys like Paul and Miska will tell you that there can be intelligible (or otherwise) information well below the noise floor.

Alex

[mayhem13]

Ok, work is over. Dinner has been consumed and fresh tracks of DSOTM ripped to both machines from Redbook and HiRes PCM BD disc. Let the listening begin! I'll start with a nice Penfolds Shiraz and go straight to Breathe.

[acg]

Ok, work is over. Dinner has been consumed and fresh tracks of DSOTM ripped to both machines from Redbook and HiRes PCM BD disc. Let the listening begin! I'll start with a nice Penfolds Shiraz and go straight to Breathe.

 

Nice choice of music Mayhem...

[levandier]

Hi Alex,

 

Thanks for your very informative thread starter! I did a similar comparison between USB 2, FW 400 & 800 and Ethernet a while back and I felt Ethernet sounded the best so I've been using it ever since. After reading your post, I pulled out an SD card, put it in the back of my 2012 Mac Mini and gave it a try. WOW! You are absolutely right - what a big improvement. Everything just sounds more right. So thank you again for sharing your findings.

 

Now I have a question for you. Your comment about bus speeds got me thinking if it would be possible to get the sound via Ethernet closer to that of the SD card by playing with the speed settings in system preferences so I quickly switched back and forth between 10, 100 and 1000 and full/half duplex and I'm pretty certain it had an effect on the sound (oddly enough it had an effect on the sound even when playing files from the SD card - but maybe it's not that surprising since I run my Mac Mini headless and so I have to use screen sharing via Ethernet to control it). I'm going to listen more carefully when I have time this weekend, but I'm curious if you also find that changing the Ethernet speed settings has an effect on the sound, and if so, what do you think the optimal setting is?

 

Thanks!

[Superdad]

Thanks for your very informative thread starter! I did a similar comparison between USB 2, FW 400 & 800 and Ethernet a while back and I felt Ethernet sounded the best so I've been using it ever since. After reading your post, I pulled out an SD card, put it in the back of my 2012 Mac Mini and gave it a try. WOW! You are absolutely right - what a big improvement. Everything just sounds more right. So thank you again for sharing your findings.

 

Now I have a question for you. Your comment about bus speeds got me thinking if it would be possible to get the sound via Ethernet closer to that of the SD card by playing with the speed settings in system preferences so I quickly switched back and forth between 10, 100 and 1000 and full/half duplex and I'm pretty certain it had an effect on the sound (oddly enough it had an effect on the sound even when playing files from the SD card - but maybe it's not that surprising since I run my Mac Mini headless and so I have to use screen sharing via Ethernet to control it). I'm going to listen more carefully when I have time this weekend, but I'm curious if you also find that changing the Ethernet speed settings has an effect on the sound, and if so, what do you think the optimal setting is?

 

Glad you are enjoying the thread Levandier. This this topic certainly is more fun than arguing over cables or music formats ;-).

 

I can't answer your question at all because you are ahead of me on the ethernet thing. I would not even have thought to ratchet it down. As I's said before, I thing it is more about noise/activity inside the computer than it is about the speed (fast or slow) of any one interface. So if futzing with ethernet speed and duplex affects that to a large degree then I could fathom there might be a benefit sonically.

 

But I have to prove everything for myself, and that one is a couple of places back in the queue for me. I too run my mini "headless," so while I think the best test of ethernet interference/sonic impact would be to start playing from the SD card and then yank the ethernet cable, that presents a control issue for me. Maybe I'll hook up a little monitor.

[Jud]

Glad you are enjoying the thread Levandier. This this topic certainly is more fun than arguing over cables or music formats ;-).

 

I can't answer your question at all because you are ahead of me on the ethernet thing. I would not even have thought to ratchet it down. As I's said before, I thing it is more about noise/activity inside the computer than it is about the speed (fast or slow) of any one interface. So if futzing with ethernet speed and duplex affects that to a large degree then I could fathom there might be a benefit sonically.

 

But I have to prove everything for myself, and that one is a couple of places back in the queue for me. I too run my mini "headless," so while I think the best test of ethernet interference/sonic impact would be to start playing from the SD card and then yank the ethernet cable, that presents a control issue for me. Maybe I'll hook up a little monitor.

 

Funny you should mention yanking cables. I got an AQ Cinnamon FW cable and two 256GB SD cards today and have had a brief chance to listen. Prefer Cinnamon FW to Forest FW I've been using. Preferred SD card on most selections, but then ran into a track where SD and Cinnamon were much closer. SD didn't seem to have the usual clarity. Got up from the sofa in the middle of the track and saw the FW cable was still plugged in, though I'd unmounted the drive. Pulled out the FW cable, sat back down on the sofa, and SD card was back to sounding its best.

[Jud]

It is not very complex. I recall a RAM disk driver for MacOS 7 was only a few K. Still, if max performance with lowest latency is your goal, an application buffer is probably the way to go. Also, RAM disks are not very memory efficient unless they use compression (which many do) because they use track/sector approach and can become fragmented just like a physical disk.

 

Would fragmentation matter at all if there's no physical read head?

 

Edit: BTW, if you use a memory player to play back music stored on a RAMdisk, you're still pulling the file into the player from the application buffer; it's just that it's pulled into that buffer from a RAMdisk rather than a physical disk.