Stereo Tool

Since my attempts at discussing the CPU load for older systems have been very unfruitful, I have a question for you, and Hans, Bojcha, and anyone else that wants to comment.

I downloaded your new preset just a few minutes ago. I was highly confused by the low CPU load, where "low" means on my system it fluctuates between 45 and 65 percent versus the 95-100% I was expecting. I went looking to see why, and noticed that you had reduced the quality slider under configuration to 50%. Bumping that up to 100% ended up with Task Manager reported CPU load of 97-99%. No stuttering, but I'm sure it was probably on the edge of stuttering, and would've if I had been trying to do anything else on my computer.

Question for everyone, although mainly for Hans: Is there a diminishing return point for this setting, similar to the "Strictness (CPU)" setting in Advanced Clipper? If so, what do you think that point is, and can it be established as a default?

About 2 weeks back, I checked out the new multi-band processor, tried to tame it and failed.
Results were far from satisfactory. To add to this injury, I ended with calloused wrists after a prolonged period of playing with the new UI. This was then.

Now, the sound is marvelous, AGC is better than ever, MB section churns out an exceptionally clean mix.

One thing which has not changed is how the parameters change with mouse. I expect my rant to be taken with a pinch of salt.
In previous UI I could use keyboard for continuously toggling/scaling up and down the values. Not anymore with the new UI.
Plus, the change with mouse pointer is either way too coarse or way too fine. With mouse scroll it's way too fine always.

One other thing, the old UI used to display precise dB values corresponding to the multiplier, eg: +6.00 dB for 2.00/ +9dB for 2.828427124746190097603... and so on.
This is not the case now. Not sure if the inaccuracy is merely at the UI level or it even affects the calculations in audio internals.

_________________
visit website

It is true that the fine tuning is complicated, but in almost every slider you can use the keyboard, which in many cases do not respond the same way. In some must be left down a long time to change the value and other changes can flchas faster with up and down and more finely with the right and left.
I think the behavior should be unified, and in some cases remove decimal.

_________________
by GAP
"Less is More" (Bob Katz)

On my system I must go down to 50% the Quality Slider, otherwise I feel the same as you. I have an AMD dual core and suffering enough!

_________________
by GAP
"Less is More" (Bob Katz)

Really? What AMD exactly? I have one AMD dual core here .. and cpu load is good, especially after multicoe optimizations.

This is a slightly smaller, AMD Athlon Neo X2 L325 1.5 GHz Dual Core.
Same with the optimization goes a little better.

_________________
by GAP
"Less is More" (Bob Katz)

Could you be more precise on what you feel is "good"?

I have a feeling that since that appears to be a K8 core CPU, and you're saying things are "good", that could lead to my observations being discounted / ignored, as it is highly obvious that Hans listens to you more than anyone else.

Just to let you know: The new MB is partially optimized (as I think you confirmed at some point), but the things that have been added since have not really been optimized yet. Basically there are now 2 implementations: A simple one (default settings, no burst protection) and a more elaborate one (also non-default settings (not accessible in the MB), combined with burst protection). Also, adding flat frequency response adds extra CPU load.

I will probably either remove those non-standard things later on (if no-one needs them), or add an extra code path with burst protection and default settings.
You never used this thing?! Ow.... Well, this depends on the latency. At latency 512 or 1024, you should really stay as close as possible to 100%. Basically, you can see it like this: Quality 50% at latency 4096 still outperforms quality 100% at latency 2048. I'm not lowering the defaults because I don't want to compromise the quality if it's not necessary, but I would say that at latency 4096 you can safely drop the quality slider to 50-70%. (In fact, in version 5.00 70% was the default, but as PC's got faster and the total audio quality improved I changed it to 100%).

I read your post a while ago but I didn't notice (or check yet) how bad it was until I was trying to precisely set the limiter values today. Had to hold the cursor-down key for over 20 seconds sometimes to reach the target level. Will look into it.
Somehow, there's an offset in the log() calculation. I don't understand this one either. I can tell you this though: If it says x2.00 (+6.02 dB), it's really x2 so 6 dB, not 6.02. (Not that small differences like this should ever be noticeable).

I knew it existed, but, without an explanation like what you provided above, I hadn't used it.

I suppose I'm wondering what the setting does behind the scenes? Quantization errors? It's been quite a while since Calculus, but I would've thought the samples would influence the accuracy. For example, with Integrals, the more samples / larger domain, the better the approximation of the area.

I split your range and set to 60. It makes things bearable again (reduction of 15-20 points). Hopefully this will allow me to start experimenting with the new stuff.


...but bear in mind I still truly believe there are performance improvements possible for systems with 1MB L2 cache.