Stereo Tool

Other than hammering my CPU, was the disk writer output in Winamp supposed to be beneficial in some way? If yes, I'm not sure how...other than I can now process a few MP3s, convert them to WAV, and run the offline Dynamic Range analyzer and see what RMS level I'm getting... Looks good so far. Highest output has been around -7, with others lower at -8 and -9.

So, how was writing out to disk supposed to eliminate the whole "Task Manager" thing?

You need to measure time how much seconds ST needs to convert that to wav.

For example, if you want to compare ST 6.10 and ST 6.20 with same preset and same filters on, the converting time will tell you exact difference.

OK. Lacking a stopwatch, I used the file attributes of when the file was created, and then the difference between that and when the file was modified.

Timbaland - Carry Out

6.10

Created - 7:25:15 PM
Modified - 7:27:23 PM
Time difference = 2m 8s (128 seconds)

6.20 (BETA125)

Created - 7:36:51 PM
Modified - 7:39:23 PM
Time difference = 2m 32s (152 seconds)

Increase of 24 seconds, or about 16% of the 6.10 baseline. Both of these tests were at 4096 with the GUI in the System Tray.

I took a supper break, and can now do tests with the GUI up, and will edit in a bit.

Edit: Since I ended with 6.20 installed...

6.20 (BETA125) - GUI up

Created - 8:33:53 PM
Modified - 8:36:26 PM
Time difference = 2M 33s (153 seconds), so there is no meaningful difference.

Speculation: Disk or memory I/O performance is a limiting factor. Gearing down to 2048.

6.20 (BETA125) - 2048 samples and GUI up

Created - 8:54:40 PM
Modified - 8:57:09 PM
Time difference = 2m 29s (149 seconds)

6.20 (BETA125) - 2048 samples and GUI in tray

Created - 9:03:59
Modified - 9:06:27
Time = 2m 28s (148 seconds)

Again little difference in GUI up v. down, but notice also little total difference in the min and max times with 6.20 (5 second max delta).

Gearing down to 512 for another speculation.

So there is difference between 6.10 and 6,20. Will leave that to you and Hans
Disk no. First it goes to memory so memory waits disk. Then memory yes. But i think even SD ram is faster then tris writing

Final test:

6.20 (BETA125) - 512 samples, GUI up

Created - 9:12:21
Modified - 9:14:36
Time difference = 2m 15s

Maximum difference between 4096 and 512 samples is 18 seconds.

Theory: Test is inconclusive. Nullsoft Diskwriter plugin and/or other internals of Winamp are likely acting as a limiter.

Supporting evidence:

6.10 - 512 samples, GUI in tray

Created - 9:27:26
Modified - 9:29:28
Time = 2m 2s

So, there is a total of 6 seconds between 6.10 at 512 samples and 6.10 at 4096 samples. That indicates that either the load doesn't increase as claimed when moving between the sample rates (latency), or the diskwriter plugin is the performance governor. Also, 6 seconds is subject to rounding, so it could be 5 or 7, but taking it as 6, that is a 5% variance, which is enough to be considered within a +/- sampling variance, meaning I could potentially test more and end up with an even smaller difference.

Bottom line, the test shows an increase in 6.20 vs. 6.10, and I think it is generally agreed upon that there is an increase, but I doubt the ability of this particular test to indicate the amount of increase in any reliable manner.

See how long it takes to write the file with NO dsp enabled at all.

If it's nowhere near as long as the times you're getting with Stereo Tool, then it's not a bottleneck. As you probably know, there's write caching in several places, including on the drive itself. So... see how long it takes to write it with NO dsp.

No DSP enabled is very fast 5-6 seconds.

Enabling just Nullsoft's DSP with the delay setting bumps that up to 9 seconds.

So what's your thought about why so little difference between 512 and 4096 samples? You mentioned write caching, but how about read caching / cache misses and high CPU / cache utilization?

I'm going to try a longer song (Pink Floyd - Echoes) so I can watch some things while it's running (maybe).

Edit:

I'm using Process Explorer, and am noting interesting things. Just having ST loaded - visible, within Winamp, with processing disabled, causes a dramatic increase in Page Faults. Across 5 runs with no DSP loaded, the page faults were 18000 for each run. With the DSP loaded, but on bypass, the page faults are consistently 30000-32000 for each run.

@Brian: These numbers look a lot more like what I've seen here in the past - little difference between different latencies.

Also at latency 512 there's less difference between 6.10 and 6.20 - that's because among others the pre-ringing filter that seems to be the cause of the big increase since BETA109 is not used at low latencies.

So these numbers are much closer to what I would expect than your previous measurements.

Let's make sure we're not having an issue with larger numbers appearing, well, larger...and forgetting when I made the previous estimation.

I measured 16% with your preferred way. I estimated 20-30% via my Task Manager way. The 20-30 was estimated before there were optimizations. If I were to estimate now via the Task Manager way, I'd be estimating 20-25, so either way, it'd be pretty close...

Poe-tay-toe

Poe-tah-toe

Please watch what I'm doing with Process Explorer, and the Page Fault and I/O Read data. I'll continue after I have a sleep...

Let me ask something
Why when StereoTool plays let`s say FLAC or OGG the screen updates rate goes a real mess ....?