I've heard it mentioned that the Nvidia soundstorm re-samples 44.1kHz to 48kHhz during playback (and does a pretty ordinary job of it). Does this mean that I'd be better off to use the DSP resampler (set at 44.1k to 48k) in Foobar2k during playback?

BTW. My motherboard is Asus A7N8X-deluxe rev2, and the sound is the onboard soundstorm.

Second Question. If I do resample 44.1 to 48k in foobar then does this have any impact on which of the output options I should use, that is direct-sound versus wave-mapper.

I doubt if you will hear ANY differences in sound regardless of the options you will choose.

I remember it made a measurable (and just audible) difference when I tested Realtek's
AC97 codec with RMAA, showing higher intermodulation distortion and sounding
too sharp in the high end at 44 kHz. So, to my opinion, it makes sense simply trying. Whether you use the wave mapper or direct sound engine, makes a difference as well,
but it all depends on your reproduction chain. I could only hear a difference with a high
quality headphone.
There were some complaints in the winamp forums a while ago about the features of
kmixer.sys (part of direct X) and they didn't really like it.
Another hint would be to try another player, like VUplayer, that employs the bass.dll from un4seen developments. I don't advertise here, but I use it for testing purpose, and it can resample as well.

I have the same board and also use the Soundstorm when I'm not in the studio. I have tested it with the udial sample, and while it's resampling is not perfect, I deem it good enough to not eat cpu cycles resampling in foobar.

Yeah it was the intermodulation distortion that made me consider using re-sample. According to Rightmark Audio Analyser (RMAA) loopback test the intermodulation distortion is way higher at 44.1kHz compared with 48kHz. It was 6.5% at 15kHz using 44.1kHz sample rate but only 0.02% when using 48kHz.

The truth however is that I cant hear the difference either way. Still the resampler doesn't seem to use as mauch CPU power as I thought, actually I can hardly notice it, so I figure I may as well use it to be safe.

6.5% IMD will likely be clearly audible. However, were you using the same sample rate on the input side of the cable during the RMAA test. What you are seeing might be caused by the ADC, rather than the DAC.

I think listening tests are in order (6.5% IMD will be trivial to ABX).

Yep I agree, but sorry I didn't post full details. When the card (or onboard) resamples the effect on IMD is fairly insignificant below a certain frequency (about 5kHz) but increases strongly with increasing frequencies beyond that point. It's under 0.05% at 5kHz but increases to 0.6% at 10kHz and reaches 6.5% at 15kHz.

The test that measured this is called the "IMD Sweep Tones" test. This "imd sweeptones" test is the only one where the 44.1kHz results really depart drastically from their 48kHz counterparts. Most of the figures like the noise floor and dynamic range are just a little bit better at 48k, and the "normal" IMD tests (which I assume just uses two static tones) show neglegable difference between the 48k and 44.1k results


It was a loopback test so you're quite correct, it's impossible to tell if the problem arises in the DAC or in the ADC or if it's a combination of the two.

If anyone wants to see what I'm referring to about the high frequence IMD problems then take a look at the following LINK.

To quote from the above link :


To see the corresponding graph srcoll about three quarters of the way down the page in that link and you'll see the image (it immediately follows the above quoted text). Notice how the "sweep tones IMD" stays low over most of the frequency range but really starts kicking up at the high end. The curve for my soundstorm is a little like this one but very much worse. In particular my sweeptones IMD starts kicking up at a much lower frequency than the one illustrated. Apparently this is indicative of a poor internal sample rate conversion algorithm.

BTW the "sweeptones" problem is completely absent if the test is conducted at a 48kHz sample rate, which pretty much clinches it that it's an internal SRC issue.

Here's some extra info on re-sample cpu usage if anyone is interested. I made these measurements because when I first enabled resample to 48000 in foobar I was expecting some fairly hefty cpu usage but I could barely detect any increase in task_manager (performance).

I then found that a more sensitive method of measuring the impact on available CPU was to run a benchmark program during mp3 playback and record the changes in benchmark figure in response to changes in parameters like re-sample etc.

The benchmark that I choose was cpumark99 and the results (very repeatable) are summarized in the chart that I made below. The six test condition shown are :

1. Nothing else running. This is a highest cpumark that my system can achieve.
2. An old version of Winamp with very low cpu usage.
3. Foobar 0.83 without any DSP plugins.
4. Windows Media Player 8. No vsiualization.
5. Foobar 0.83 with 48000 resample
6. Windows media Player 8 with visualizatin (ambience random) running.



BTW My system is an AthlonXP2600 (overclocked to 2230MHz) running on an A7N8X-deluxe motherboard.