New OSX Opensource audiophile player : Audirvana

[Riot Nrrrd]

@Jud:

 

Well, most of the OS is 64-bit, but actually the kernel by default is 32-bit in order to run 32-bit compatibility kernel modules. If you don't need the modules, you can choose to run the kernel in pure 64-bit mode.

 

I did some research, and it turns out we're both sort of right

 

http://support.apple.com/kb/ht3770

 

says

 

These Macs use the 64-bit kernel by default in Mac OS X v10.6:

 

- Mac Pro (Mid 2010)

- MacBook Pro (Early 2011)

 

These Macs use the 64-bit kernel by default in Mac OS X Server v10.6 (they can also use the 64-bit kernel in Mac OS X v10.6, but do not use it by default):

 

- Xserve (Early 2008) and later

- Mac Pro (Early 2008) and later

- Mac mini (Mid 2010)

 

These Macs support the 64-bit kernel, but do not use it by default:

 

- iMac (Early 2008) and later

- MacBook Pro (Early 2008 through Mid 2010)

 

That said, I still think your findings are trippy, but then we can't argue with our ears, can we?

 

 

 

[khollister]

Well, just got back from a short vacation and tried 0.8.0 vs 0.7.4. They definitely sound different. While 0.8.0 seems to have a little better clarity and definition to the instruments, it also has a little more edge and hardness to the treble. I recall many versions ago that Audirvana had this same quality which was why I was using Amarra and Decibel (ayrewave then). Audirvana somewhere along the line got the same sweet, delicate highs that Amarra has, and now the new compiler appears to have gone backwards a bit on this particular aspect.

 

I am certainly keeping 0.7.4 around since I have absolutely no issues/bugs with it on my system. 0.8.0 does have a certain "presence" to voices that I like however.

 

I have no idea at this point what is "accurate" versus "euphonic". The older version compiled with GCC has more "tubelike" upper mids and highs to my ear, which I'm a sucker for.

 

Damien - it would be nice if you could build 0.8.0 with GCC as well as LLVM so we know 100% for certain we are hearing a compiler difference (which really hurts my head, BTW) rather than something else in the baseline.

 

 

 

[Riot Nrrrd]

@church_mouse

 

David, are you having the problems with Audirvana when you aren't running iTunes as well?

 

I've got an old MacBook Pro with only 2 GB of RAM as well, and so far I've only seen Audirvana put up the red CPU overload indicator once - but I'm probably not using it anywhere near as much as you are. Also, I have never run Audirvana as the only program running - I usually have at least Mail.app and a Web browser and Terminal running.

 

Can you identify your Mac mini from the list on this page?

 

http://support.apple.com/kb/ht3476

 

If you could increase the RAM from 2 to 4 GB it would probably be a big help.

 

Also, is this happening to you when Audirvana is the only program you have running?

 

If that is the case, you might want to try running Activity Monitor (from your Applications -> Utilities folder) and experiment with sorting on the Real Mem and Virtual Mem columns.

 

Watch the behavior over time as you play things. See whether or not you can see the numbers grow for whatever is at the top of the sorted lists. Also check out the System Memory tab at the very bottom of the window; see what you have for "Free" (in green) and "Inactive" (in cobalt blue).

 

If you are only setting Audirvana to use 256 MB and have nothing else running, it's pretty surprising to hear that you are having CPU overload issues.

 

The most common impact on running programs is when memory is exhausted and the system has to page memory pages in and out of memory to/from disk; if you can avoid that, it's far more likely that you can avoid the dreaded red indicator.

 

 

[57gold]

Also hear a bit of hardness to treble...also a bump in overall clarity versus the penultimate update.

 

[iamoneagain]

I'm not sure if I like the new sound better or not. It does seem to have more weight and is still very smooth but previous version had more air and a more liquid sound. Might go back to previous version just to confirm.

 

[Jud]

Riot Nrrrd writes:

 

That said, I still think your findings are trippy, but then we can't argue with our ears, can we?

 

Hey, I thought they were trippy, so how can I complain when you agree with me? :-)

 

But I was thinking (always dangerous) - when Audirvana is built with a different compiler, we are actually listening to a different binary, right? So it probably shouldn't surprise us that what is actually a different program being executed sounds different.

 

The same may apply (I don't know for certain) with regard to the 64-bit kernel. Whether it's a different kernel extension (64-bit vs 32-bit) acting as the audio driver, or perhaps fewer kernel extensions running overall (having the same effect on audio as running fewer programs), there are probably quite logical and prosaic reasons for the difference in sound.

 

 

 

[khollister]

It is also darker and the depth folks are commenting on comes with a "hooded" quality to the midrange that I also hear in Fidelia.

 

It will be a long term problem since Apple is deprecating support for GCC going forward in favor of LLVM, but I really think 0.7.4 sounds better IMHO.

 

[JS21]

There definitely is a difference in SQ with the latest 0.8 release. Not saying it's better, just a different "feel". Bass noticeably deeper and maybe thicker. Will be playing with the sub volume to get the best balance...

 

[damien78]

@khollister:

I've uploaded the same 0.8.0 but built with gcc instead of llvm on audirvana.googlecode.com

 

So you can compare for the compiler only impact on SQ.

 

Damien

 

[church_mouse]

If it is of any help, it is a mid-2007 model MacMini.

 

As it is just used for music, I do not have any other apps running except for i-Tunes but switching that off does not stop the CPU overload. I do no upsampling - simply AIFF 16/44 played at that rate.

 

I'll keep an eye on the Mem readings, though I am too much of a computer "numpty" to make a lot of sense of them.

 

Audirvana has never worked for me with any lengthy listening (I tend to sit down for 2-3 hour listening sessions when I get a chance) and it is a shame as I like how it makes music sound in my system.

 

Thanks for replying.

 

[Jeff In San Diego]

Hi Damien,

With all this talk about "bit perfect player" and differences between things that ought to sound exactly the same...I had a thought.

 

Should be pretty easy to insert something at the very end of Audirvana "data path" that saves the raw digital output to a test-file. For example, the method that gets called to fill the I/O buffer could do its normal work and then additionally append those bytes to a "test file".

 

If you did that and then captured a "test file" for the same music file using two different versions of the app (for instance, v0.8 compiled with the two different compilers -- or v0.8 and v0.74). Could then create a "difference file" to more rigorously compare the output of two versions of the app.

 

This might teach us something interesting about how the two versions actually differ. On the other hand, if the difference file is all zeros, it will be a painful bit of placebo for us to swallow.

 

Just thinking out loud. Yes, I know. I could do this myself, and probably should...but I've already started a (non-music-related) project of my own.

 

Keep up the great work! I'm enjoying v0.8 very much.

 

[Jud]

Damien:

 

For those of us who get geeky pleasure from doing this ourselves, does XCode 4 have a switch to allow building with GCC from source at home?

 

[Joebah]

Apparently, one of the other of the "other" players doesn't play nice with my MacBook Pro and leaves messes behind even when closed. So it appears that if I use one of the others before using Audirvana, bad things happen, including the clicks.

 

Solution: restart and open Audirvana first. The sound is back in all its glory, and no clicks.

 

[Bob Stern]

@wgscott: Thanks for pointing this out! I had no idea prices had dropped so much for this size RAM (i.e., two 4 GB cards). It was $300 when I bought my Mini in June.

 

Incidentally, I recommend buying RAM from Crucial (the retail arm of Micron) because Micron is one of Apple's suppliers.

 

[Bob Stern]

@Joebah: Do you mind sharing which player app left your system in a state that created problems with Audirvana?

 

It would be very helpful for the rest of us to watch out for that and confirm whether that app is a problem on other users' systems.

 

[Joebah]

and anyone else that would like the info. I can say that this "fix" has certainly worked for me and I am again fully enjoying Audirvarna.

 

 

 

[iamoneagain]

After comparing both, think I like the newer version better. It actually keeps the air of the previous version but definitely has more bass and impact. This works out better for my mostly rock library. The previous version might be better for classical. I'll have to give a longer listen. I do know I like both versions but than any of the other players.

 

 

 

[Jud]

Micron supplies lots of computer makers. I second the recommendation because they sell very high quality memory (never have had a problem, going back to builds many years ago) at extremely competitive prices - and when I say "competitive," I include the online retailers like Newegg, OWC, etc.

 

[khollister]

After listening to 7.4, 8.0-LLVM and 8.0-GCC, I believe that the 8.0 versions are much closer in sound regardless of the compiler used. 7.4 still sounds different than the GCC build of 8.0 to me.

 

Damien - the only change was shuffle play?

 

 

 

[Jud]

Found the setting to switch compilers, and built Audirvana with all 3 possible settings in XCode 4 - llvm 2.0, gcc 4.2, and llvm gcc 4.2.

 

I understand what khollister is saying about the gcc version - it has an appealing almost ethereal quality with its big soundstage and shimmering presentation of vocals.

 

I have to personally come down in favor of the llvm version, though. I don't think it loses anything in the soundstage, and vocals sound as if they're coming from a human with a mouth, throat and chest, not so much of a disembodied presence floating on top. When I put on "Hold On" from Tom Waits' "Mule Variations" it sounded a heck of a lot like Tom was there in my family room singing to me. (Damn fine song with more than a drop of sentiment to it like a lot of Waits' best, makes me wish I was young enough that they'd played it at my senior prom, and I didn't even go to my senior prom.)

 

Vocals seem to have gained some detail and even more microdynamics. On Jakob Dylan's "Nothing But the Whole Wide World" from "Women and Country," Neko Case's backing vocal and its blending with Jakob's lead came through a lot more clearly with the llvm version.

 

I'm not getting the harshness a couple of folks have complained of. Probably down to the Vandersteen speakers, and, I believe, the tremendous musicality of the Omega Mikro Zephyr digital S/PDIF cable. It'll do harsh if it's fed junk, but give it something well recorded and it really sings. (T-Bone Burnett, who is determined to fight against the plague of compression in most modern recordings, produced the Jakob Dylan album. Waits' wife Kathleen Brennan produced "Mule Variations.") Maybe the async interface and consequent jitter reduction from the Musical Fidelity V-Link also keep the sound from being too strident.

 

Just to be thorough I tried llvm gcc 4.2, but instead of being the best of both, it was the worst of the lot IMHO.

 

[iamoneagain]

Hmm, after playing some slower tracks I may have changed my mind and actually prefer GCC version. The bass almost seemed bloated on several tracks and the sound is not as black as previous version. I guess it's a good thing we have both to choose from.

 

So I don't keep flip flopping, I'm going to spend a few days with this to get a better feel. But I really don't want to switch players depending on what song I'm playing. I just want to stick with one.

 

[Jud]

Jeff in San Diego writes:

 

With all this talk about "bit perfect player" and differences between things that ought to sound exactly the same....

 

Where'd you get that idea? Any reasonable CD player that doesn't have horrible dropouts overwhelming its error correction can pass bit perfect information from the CD, and they don't all sound the same. Or even more to the point, what about digital cables? Except for one that simply doesn't work, they will all pass bit perfect digital information, but I certainly hear vast differences between them.

 

The ability to pass bit perfect information is a kind of minimum guarantee that the player won't make wholesale changes/distortions to the sound that you don't intend, but nothing more than that. In fact, many people really like the sound of upsampling software and hardware, which by definition is *not* bit perfect, so bit perfect is no assurance of the best sound from a given software or hardware player, let alone a point where all the different players sound identical.

 

[audiozorro]

Well I just checked my MacBook Pro that has been running Audirvana nonstop for 3 days now. My computer had the Forced Quit Window open with the message "Your Mac OS X startup disk has no more space available for application memory."

 

These are the results from the Activity Monitor for Audirvana:

 

Audirvana - (Not Responding)

Threads - 3

Real Memory - 5.29GB

Virtual Memory - 49.26GB

Kind - Intel (64 bit)

 

I previously reversed all the wgscott scripts to their original state before playing Audirvana last Friday, so the problem seems to be solely due to Audirvana playing a number of albums in repeat mode.

 

I know when I left for work this morning the Activity Monitor showed that the hard drive had over 47GB free. This evening my free space is less than 600MB.

 

I don't know what's causing this memory leak but intend to try another player running 24/7 for a few days to see if the problem is limited to Audirvana.

 

[wgscott]

Are the static libraries you compile and distribute with the source code recompiled with the newer compiler? Also, why not dylibs?

 

[Johnny Moondog]

SidHo... Don't know if this will help but here's my experience thus far.

 

I have 8GB in the Mach2 (MacMini 2010). I've been using Audirvana for the past 3 weeks with the default settings: Exclusive access mode, Integer mode, Use max I/O buffer size, with the 'Maximum memory allocated for tracks pre-load' set to 4GB.

 

The Audio Filters tab is set to 'Apple CoreAudio', Quality 'Best', but I've never used Forced Upsampling so my understanding is that those Sample Rate settings are not being used?

 

I was listening entirely to Red Book CD files with the exception of the Plant/Krause 'Raising Sand' recording at 96 KHz 24 Bit. Never had an instance of CPU Overload. A couple of days ago I purchased The Rolling Stones 'Aftermath' at 176 KHz 24 Bit and finally saw the CPU Overload red flag appear. The first time was during playback of the albums last track. It locked up and froze the MacMini. I noticed the track is 11:13 long and suspected that was the problem. Too much 176 KHz data to load up into memory play?

 

So I increased 'Maximum memory allocated for tracks pre-load' to 6GB. Since then I've still encountered the Red Flag freezing the MacMini while trying to play the Stones tracks. But it seems to be from moving too quickly. If I load the tracks into the playlist and then immediately press play the Red Flag appears and the MacMini is frozen. But if I wait a few seconds and then press play, the Red Flag appears briefly then disappears and playback begins. I haven't tried that last track of 11 minutes.