my personal experience is that i dont like to kill my CPU haha so im using 512 or 1024 buffer size for the most part sound doesn't feel compromised at this settings to me and i don't want a latency thats is so huge its impossible to automate at the same time
what about nebula reverbs guys ?
how low can you go with buffer size before it doesn't sound right ? is there a limitation at all or is it only the cpu who get hit by a lightening lol, on my mac mini 2012 i7 quad i had to go for insane huge buffer settings for reverbs
now that im on a 5820k i think i will lower my buffer size for reverbs a little
what do you guys use for reverb latency settings ?
firstname.lastname@example.org wrote:my personal experience is that i dont like to kill my CPU haha so im using 512 or 1024 buffer size for the most part sound doesn't feel compromised at this settings to me
what do you guys use for reverb latency settings ?
I am able to get away with 128 buffer setting using the same exact cpu you have (5820); I don't over clock either.
However, this only works on certain lighter CPU presets and some will of course crap out and will still load but sound like a garbled mess due to heavy cpu usage..
And the presets that work at 128 buffer settings, I leave Nebula at 128, and I turn my interface buffer to 64.. and often it will still work well, but with slightly less monitoring latency (live tracking purposes).
There still is a tiny big of latency from when the vocals hit and nebula processes the sound.. but it is a compromise, and I just consider it 'predelay'
Not cheap to create either Created an original piece of hardware just to replicate the real sound of the original tubes. Then gather a collection of the finest example of the rarest tubes in the world based not just on scarcity but on their desired sound.
I am always more than happy to do a custom combo for sure! $20? Probably not, but I'm certainly happy to help make it easier especially bundled with something else- just email me and I'll make it work.
RJHollins wrote:Them tubes ain't cheap
I recently re-tubed my Dynaudio ST-70 amp for ~$200 [and I got a good deal].
Just wanted to chime in on this. Did the same zero-invert tests mentioned in this thread. Did you test the SAME buffer settings against each other as well? What you are experiencing is the positive effect of the high quality of reproduction of the original hardware- it always has subtle changes from track to track from the same program. If you zero 128 to 128 or 1024 to 1024 from multiple instances you will find the same variation results.
Now yes, things like sample rate, levels that samples are recorded, bit depth, and editing the engine all change the way it is played back, but the buffer is just that- a buffer.
Shibata wrote:If you want magic on lows, try to AlexB Thermionic Phoenix Mix Compressor with Siemens tubes. Copy preset from 3 to 5 times or by taste. But not compress. Important thing - nebula internal buffer, must be 128 and youre audio buffer the same or lower.
I've found that for colour and for that hardware sound, work best not pre's but full chain eq, built in the position, when all bands with ND go first and the last band is HQ. Especially nice sound in that chain the M****y massive passive from Alex and all his eq as well. Very good alternative to pre is Tim's Gyrator, because its dynamic and there famous Marinair transistors, which was on old N**e 1073. The most profit from that sound is 96k version ( even with SR ) and 128 nebula buffer. But is very heavy on CPU.
Why would a shorter Nebula buffer increase sound quality?
Because the engine render it with more resolution. And this can be seen by how render in reaper the waveform. At 8192 it do fast and at 128 it do slowest and of course it use more CPU. Higher buffer lacks the transient because of latency of the buffer and the latency compensate not work due to nebula engine. And it is easy to prove.
http://i.imgur.com/xBUTcbW.jpg?1 You see three files. One rendered with 8192 internal buffer and 128 audio card buffer. Second with 128/128. And the last is phase inverse difference between them and than normalized. You may here how much signal is gone at the 8192 buffer. And now imagine, how much you lose at mixing stage with lot of tracks.
I use 128 buffer on every tipe of nebula with the exception of reverbs and delays, which is better to put 8192 due to long kernels.
Yes the 128 eating much more CPU, but the sound worth it. Following this logic, if nebula had the lower buffer than 128, let us say 32, like on RME cards, this was even more precise emulation, but killer on CPU. Even so I would like to have than kind of opportunity.
There is variation within the program itself, including how it plays back the layered dynamic changes to frequency and harmonic distortion. The same material rendered over and over will produce subtle changes similar to the analog. Reverb instance left at long kernel length and high rate cnvt will give consistently positive results. There can be advantages to intentionally creating a preset, especially compressors, with or without different timing elements intentionally built in because it is then the engine making intentional changes to the signal as opposed to memory- cpu based changes. CPU load is not a reflection of working harder because a better signal, it is working harder because of that particular system's bottle neck. Since the size of programs with more information- especially harmonic distortion kernels, are a ridiculous pull on resources, there isn't much better to test cpu than loading some NebPro reverb instances up! Thank goodness for offline render and batch.
be a gorilla wrote:interesting!!
hey shibata, do you so the Same dsp Buffer Procedere for aquas? i'm a Little bit confused about Sound Quality dependent on dsp Buffer Settings ? giancarlo, can you confirm?
No i am not testing aqua, i don't like how the library sounds on it, though this is the same nebula in GUI and N4 features. Maybe it's a bad choice of sampled units or bad signal path ( converters, hands) _) duno, but i always return to 3rd party libraries. If I'm not mistaken, the default buffer in aqua is 1024, and you can change it by xml file. However, there are Z markered preset which means zero latency, but i never use it and can not say anything about them.
The buffer by itself not downgrade the sound quality. It's like a discrete processing. Because RTE in kernel page is larger when the buffer is higher. Look at this: http://i.imgur.com/kycxvJQ.jpg
During the rendering it miss part of the signal due to that difference between 2.902 and 23.21 and i don't know why latency compensate not work in this situation. Perhaps the case in the engine architecture.
By the way. Remember old holywar on forum between which engine sound better: timed or freqd? So here, if you set the buffer 8192 with 23.21 ms RTE ( don't forget save and reload preset after change ) and than switch all kernels to timed engine you'll get even faster RTE 1.995 ms. http://i.imgur.com/fRigtYM.jpg
So possibly its the answer why timed is better. But on CPU its crazy. So 2.902 ms not bad choise, though 1.995ms theoretically it render all transients better, especially in a mix stage where many nebula instance and cumulative effect more evident.