Crashing in live osx and can't authorize - SOLVED
-
- Banned
- Topic Starter
- 22457 posts since 5 Sep, 2001
[DELETED]
-
MeldaProduction MeldaProduction https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=176122 - KVRAF
- 14325 posts since 15 Mar, 2008 from Czech republic
Oukey doukey, I'll see what can be donettoz wrote: the reverb is ALMOST fixed..... it now just makes a small pop (like a sibilant) where before it used to play the rest of the tail.. unfortunately that pop is still going to be present in a bounce, so it still makes it hard to use. Rob Papen RP verb had this problem worse than any other i tried and they fixed it 100% so at least we know it can be done.
Well, you can still use the lower quality upsampling. But generally upsampling is only an extreme solution. You should do all processing at 96kHz and upsampling usually is not necessary in that case.ttoz wrote: The dynamics seems ok no pops - however an entire core for 4x oversample still seems too much for me. Could you let me know how much 4x oversampling uses on a windows box? It would be interesting to compare.
Anyway on my PC (2.5GHz AMD Phenom 64 quadcore in 32-bit mode) it takes 4.4% with 4x upsampling and 0.9% in LQ mode. Used 44kHz and latency of 256 samples. Note that the core in Windows and Mac is basically the same, both optimized for SSE2, which should be everywhere now.
I wouldn't be surprised if this would be caused by Logic + AU. From what I know about AU internals, I believe it is very very slow. I have also found that Logic doesn't even use basic optimizations (e.g. dispatched processing). And actually when I compare my PC and Mac, the Mac seems extremely lazy compared to it. But I hope you don't feel offended (I know many Apple folks don't like this, but it's just the truth, at least my truth
).For comparison, on my Mac MDynamics 4x upsampling takes 42%, but since it is double-core, it is in-face 24%, but still 6x slower. But it really isn't the best out there, so I guess it could be better.
Well, on my PC it takes 7.5% with 4x upsampling and default settingsttoz wrote: the linear phase auto is ok, no more sound/pops
the spectral dynamics now uses an entire core in logic for one instance.... with no upsamplingno buffer problem though
.My mac takes 12% (converted from double-core) with no upsampling.
Well, I'll check the reverb. The computation speed is probably matter of CPU, OS, host and interface I'm afraid.
-
- Banned
- Topic Starter
- 22457 posts since 5 Sep, 2001
[DELETED]
-
- Banned
- Topic Starter
- 22457 posts since 5 Sep, 2001
[DELETED]
-
MeldaProduction MeldaProduction https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=176122
- KVRAF
- 14325 posts since 15 Mar, 2008 from Czech republic
Well, I can see you feel offended, so I'm sorry. But I'm afraid I know most of how it works and this is just how it is.
However:
MDynamics is a little more optimized on Windows (measured by 1.5x using SSE, which has not been ported to Mac, this doesn't involve the upsampling core). Also I found than MSpectralDynamics & MAutoEqualizer cause some denormals when processing silence, so they get slower, probably only on Intel CPUs, I'll check.
But:
I just checked Logic vs. Reaper using MSpectralDynamics - Logic took about 20%, Reaper 16%, when using VST Reaper took 15.5%, so not a big difference, but still (measured overall, between 2 cores). Another test - MDynamics 2x upsampling takes 20% in Reaper and 25% in Logic. Both Logic measurements are "about", because of the digital meter, in Reaper there is a percentual view. And for comparison on PC it takes 5.2%, but yes, it is a faster processor.
I know you cannot accept the possibility of PC's being faster than Macs. There is now the same hardware, so why should it be? The thing is, this is an extremely complex topic and we cannot even state that something is faster, because it depends on the usage. For example one of the reasons, why my PC is faster is because it is based on AMD processor, which in certain circumstances is faster, but sometimes it is slower than Intels.
Well, if you want to discuss it, I'll be happy to give you several technical arguments, why I think the situation is the way it is, but let's do it somewhere else. I'd like to focus on the problems related to our software here.
1) MSpectralDynamics upsampling on PC
I made a little confusion here
- I didn't realize that with increasing sampling rate MSpectralDynamics actually gets faster. This is because of the "resolution" parameter, which ensures stable time-domain response over different sampling rates.
Therefore when I increased upsampling, the upsampling algorithm itself made it slower, but the actual spectral processing algorithm got faster, so it kinda compensated each other. I feel like the processing algorithm was too fast then, but well, who knows, I'll rather check though.
So the example with MSpectralDynamics wasn't really the best one
.
2) Upsampling performance
The thing is, there are many ways to do upsampling - when you turn of the HQ upsampling, you will see. The HQ filter is much much slower, because it tries to be as good as possible. Now give me a number of percents (at least one ) you want the upsampling to take and I'll give you a filter like that , but the quality will be reciprocal.
Frankly the HQ algorithm can be improved, I just didn't get to it, because it really isn't the main priority. If you feel you really need it to be as fast as possible, I'll try to rush it. But honestly, you can still use the LQ algorithm.
Btw. I hope you are not using upsampling on a daily order. As one of our customers stated "upsampling is a fake cure for everything these days". The worst thing you can do is to use upsampling in multiple plugins in a chain. Each of these plugins would increase a sampling rate, filter it, process, filter it down, decrease a sampling rate. Do you know how much computations is that? Besides the fact, that it is crazy to do that and it takes most of your CPU without any real income, it modifies the audio signal, A LOT!
So I hope you understand that upsampling is an extreme solution, when you HEAR something you don't like, so you can try upsampling to remove it. If you want the best quality, use higher sampling rates. This way you spare your CPU and don't overprocess the audio.
3) The main problem
So it seems working in Live now, right?
However:
MDynamics is a little more optimized on Windows (measured by 1.5x using SSE, which has not been ported to Mac, this doesn't involve the upsampling core). Also I found than MSpectralDynamics & MAutoEqualizer cause some denormals when processing silence, so they get slower, probably only on Intel CPUs, I'll check.
But:
I just checked Logic vs. Reaper using MSpectralDynamics - Logic took about 20%, Reaper 16%, when using VST Reaper took 15.5%, so not a big difference, but still (measured overall, between 2 cores). Another test - MDynamics 2x upsampling takes 20% in Reaper and 25% in Logic. Both Logic measurements are "about", because of the digital meter, in Reaper there is a percentual view. And for comparison on PC it takes 5.2%, but yes, it is a faster processor.
I know you cannot accept the possibility of PC's being faster than Macs. There is now the same hardware, so why should it be? The thing is, this is an extremely complex topic and we cannot even state that something is faster, because it depends on the usage. For example one of the reasons, why my PC is faster is because it is based on AMD processor, which in certain circumstances is faster, but sometimes it is slower than Intels.
Well, if you want to discuss it, I'll be happy to give you several technical arguments, why I think the situation is the way it is, but let's do it somewhere else. I'd like to focus on the problems related to our software here.
1) MSpectralDynamics upsampling on PC
I made a little confusion here
- I didn't realize that with increasing sampling rate MSpectralDynamics actually gets faster. This is because of the "resolution" parameter, which ensures stable time-domain response over different sampling rates.Therefore when I increased upsampling, the upsampling algorithm itself made it slower, but the actual spectral processing algorithm got faster, so it kinda compensated each other. I feel like the processing algorithm was too fast then, but well, who knows, I'll rather check though.
So the example with MSpectralDynamics wasn't really the best one
.2) Upsampling performance
The thing is, there are many ways to do upsampling - when you turn of the HQ upsampling, you will see. The HQ filter is much much slower, because it tries to be as good as possible. Now give me a number of percents (at least one
) you want the upsampling to take and I'll give you a filter like that , but the quality will be reciprocal.Frankly the HQ algorithm can be improved, I just didn't get to it, because it really isn't the main priority. If you feel you really need it to be as fast as possible, I'll try to rush it. But honestly, you can still use the LQ algorithm.
Btw. I hope you are not using upsampling on a daily order. As one of our customers stated "upsampling is a fake cure for everything these days". The worst thing you can do is to use upsampling in multiple plugins in a chain. Each of these plugins would increase a sampling rate, filter it, process, filter it down, decrease a sampling rate. Do you know how much computations is that? Besides the fact, that it is crazy to do that and it takes most of your CPU without any real income, it modifies the audio signal, A LOT!
So I hope you understand that upsampling is an extreme solution, when you HEAR something you don't like, so you can try upsampling to remove it. If you want the best quality, use higher sampling rates. This way you spare your CPU and don't overprocess the audio.
3) The main problem
So it seems working in Live now, right?
-
- Banned
- Topic Starter
- 22457 posts since 5 Sep, 2001
[DELETED]
-
MeldaProduction MeldaProduction https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=176122
- KVRAF
- 14325 posts since 15 Mar, 2008 from Czech republic
Ok very well,
so I'll try to summarize the current issues:
1) cpu usage - this generally means upsampling, right?
2) stability in logic - now I don't recall any problems. Can you please point me out?
3) AU reset issue - now it remains with the click in MReverb, right? (only from the plugins I've sent you of course)
Is that right?
so I'll try to summarize the current issues:
1) cpu usage - this generally means upsampling, right?
2) stability in logic - now I don't recall any problems. Can you please point me out?
3) AU reset issue - now it remains with the click in MReverb, right? (only from the plugins I've sent you of course)
Is that right?
-
- Banned
- Topic Starter
- 22457 posts since 5 Sep, 2001
[DELETED]
-
MeldaProduction MeldaProduction https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=176122
- KVRAF
- 14325 posts since 15 Mar, 2008 from Czech republic
Very well. I'll check and will inform you. About the spectral dynamics - there is the problem with denormals, but otherwise I'm afraid there isn't big chance to lower the CPU requirements, because the algorithm is just too complex. On the other hand you can easily decrease the CPU usage by increasing the resolution parameter (basically from 10ms to 20ms it should almost halve the CPU usage).