I just spend the last 8 hours helping a friend with his computer. It is an Athlon Thunderbird 1,4 GHz on an Asus board with VIA chipset.

He resently began ripping everything to FLAC. He also replaced his Live! card with a 6Fire from Terratec. His problem was that he got horrible skipping errors with FLAC and 6Fire, typically with 10 minutes intervals. He got the same errors with the Live! card and FLAC, but there could be hours between the errors.

No errors when playing Ogg Vorbis or wave files.

He claimed not having any heating problems. His CPU was only about 65 °C, which he told me was more than acceptable for an Athlon Thunderbird. He made some very innovative cooling with a flex pipe.

I speculated about corrupt FLACs and ran flac -t on the server. No problems. Then I wanted to stress the CPU on the client and ran flac -t on the same file in a loop. To my surprise, I got a MD5 checksum mismatch in 50% of the tests. Then I tried to encode a wave to FLAC with -V (verify). It failed in 25% of the tests. This was WinXP. We then booted in Win98 and got no errors at all.

Very strange indeed. His machine has run flawlessly for three months. It never crashed and he never got strange messages.

At this point I was ready to give up on FLAC. I was thinking in lines of the windows binaries were incompatible with Athlon. So I installed MAC. This was a good thing. When I ran verify on the encoded files, I was told where the error occured, i.e. 20% and Winamp gave a warning 20% into the song.

This convinced me it was a hardware error. We underclocked the pc to run 1050 MHz/100MHz and suddenly all problems went away. The problem caused by overheating VIA chip on mainboard. We manually cooled that chip with another fan and could reduce the flac -t errors to 0. MAC still failed.


My observations, comments and questions:

* You might have defective hardware without knowing it.

* Why does the 6fire card fail more frequently than Live! ?

* MAC has a great advantage over FLAC in being able to tell exactly where the file is corrupt -- and stopping playback in Winamp.

* Why did FLAC playback fail, while wave and Ogg Vorbis could play forever with no errors? Does it skip a frame when found defective?

* Why does flac -t only fail in 50% of the tests?

* Why does MAC verify fail every time the same place?

* May I suggest an option for the FLAC Winamp plugin: Fail on ERROR/WARNING. It would be very helpful to know see when the error was detected instead of having to listen and listen.

Wow, interesting observations... this could serve as a warning for overclockers or for people trying to build quiet/under-cooled PC's.

So with the overclocked machine, encoded files would have errors, and MAC would report these errors reliably but FLAC wouldn't report them all the time? And by reducing the system temperature, FLAC files no longer encoded incorrectly, but MAC still did?

Wow, interesting observations... this could serve as a warning for overclockers or for people trying to build quiet/under-cooled PC's.
Overclockers, not underclockers

So with the overclocked machine, encoded files would have errors, and MAC would report these errors reliably but FLAC wouldn't report them all the time?
I am not about to make such a statement.

And by reducing the system temperature, FLAC files no longer encoded incorrectly, but MAC still did?
No, by reducing the system temp on the normal clocked system, flac -t worked fine, but MAC failed.
Both MAC and FLAC worked fine on the underclocked system.

@Annuka

i think he should try memtest86 on his computer.

65ºC is actually way too much for an old 0.18µm Athlon Thunderbird if it is the idle temperature... Older Thunderbirds can have some serious data corruption above 70ºC in full load... And it's quite likely that the temperature goes beyond that in full load (as I said, if the idle temperature is 65ºC)... Idle shouldn't be above 50ºC...

FLAC and MAC are deterministic systems (assuming that the whatever is executing the code is deterministic), maybe it just happened that the hardware problems were triggered differently or sporadically by flac code while mac code always triggered it.



That is a good idea. I don't like it as a default behavior since a player is supposed to deal with errors as much as possible without disturbing playback, but as an option it is fine.

There are many options I would like to add to the winamp plugin (like the ones in the XMMS plugin) that I haven't because I don't know MFC GUI stuff. At one time I tried to borrow from the MAD plugin but I got that dirty disgusted feeling I always get whenever I delve into a Windows API, so I'm hoping one day someone just submits patches for all this stuff. Once the basic framework is there for the configuration windows I can at least add to it easily.

Josh

why don't you delve into wxWindows or something similar instead, Josh?

Would it be possible to implement those options thru a .ini or .conf file until someone submits that patch?

I had a horrible problem when I upgraded my system where MAC files kept getting corrupted, yet I could still compress/listen to MP3's & OGG's with no problems (that I noticed anyway).
Lo & behold, I got a copy of Memtest86 as mentioned by harashin and it turned out I had some bad memory chip. Replacing the memory fixed all my problems.

Unfortunately the memtest website was down during the hours of distress. Bad memory was the first thing that came to my mind, so we put in another module we knew worked perfectly. The VIA chip on the board got very hot. I am pretty sure the problem lies there.

You mean do wxWindows dialogs for the Winamp plugins? I think that may be a distribution problem for people who are just used to dropping in_flac.dll into the right dir.

I actually like wxWindows though, that's what I'm developing fui (my transcoding GUI) in.



I guess that's doable. Related question, does winamp2 have a configuration interface like XMMS?

Josh

I haven't really tried wxWindows, but from their FAQ:



So I guess wxWindows can be statically linked, so people would still only need to drop in_flac.dll into the right dir.