Does anyone know the nitty gritty details on how VE Pro handles threads, multiple instances and audio ports? I have some issues with CPU spikes on my slave, and I'm curious about how to go about easing the load. Does the number of audio outputs affect CPU load? How about the number of instances VS the number of instruments loaded inside? Is it better to run at higher or lower sample rates? High buffer on soundcard and low buffer on VE, or vice versa? Is it easier to use the "None" buffer settings on local host instances because there is no data going over the ethernet?
-
VE PRO CPU use
-
A lower samplerate, higher soundcard buffer size and more buffers of latency in the VE Pro Server Interface plugin will usually give you the most dramatic performance increase. VE Pro spreads audio processing over multiple cores in an ideal fashion, but for best performance, the total amount of threads should be set to the amount of CPU cores. Keep in mind that the Threads preference in the VE Pro Server is a per instance setting. So if you set it to 4 threads and you have 2 instances, this will actually mean 4*2=8 threads. So for 4 cores, a setting of 2 threads would be more ideal in this case. Also, if you're running the VE Pro Server locally, it might be a good idea to reserve a core for the master host as well. In general it's easiest to keep this set to the amount of cores and lower it only in case you experience performance issues. Lowering the amount of audio ports shouldn't make a significant difference in performance, but it could help.
-
So, for a slave (1 x Quad Core Xeon) with 11 VE Pro instances (8 x Vienna, 3 x Kontakt 4), is it best to set the threads to "1"? Also would you advise to disable multiprocessor support in Kontakt?
At the moment I am only getting a glitch free operation with a 512 sample buffer on the host (Logic 9), but this feels just too slow when playing in parts. All the VE Pro plugins in Logic are set to 2 x buffer. There are many variables, and it's hard to find out where the bottleneck is. The problem doesn't seem to be CPU power on either host or slave. During playback of sessions, the CPU meters on either machine hardly go beyond 10%, except core 8 in Logic which seems busy with the Lexicon reverbs, but even if those are disabled, it's not possible to get lower latency operation with VE Pro.
-
@Karel said:
So if you set it to 4 threads and you have 2 instances, this will actually mean 4*2=8 threads. So for 4 cores, a setting of 2 threads would be more ideal in this case.Karel, could you please define 'instance' for me? I am still confused whether an 'instance' means a single Vienna Pro 'instance' (the mixer with any number of instrument tracks, buses and input tracks), or whether it refers to each individual virtual instrument that is used (i.e. Kontakt, Vienna Instrument, Stylus RMS, Atmosphere, etc., etc.....). It seems as though each is referred to as an 'instance'...
Thanks for clarifying.
- Greg
-
@Dom said:
So, for a slave (1 x Quad Core Xeon) with 11 VE Pro instances (8 x Vienna, 3 x Kontakt 4), is it best to set the threads to "1"? Also would you advise to disable multiprocessor support in Kontakt?
I searched for an answer to this question exactly. Is Dom right about this? I have 6 local instances of VEP, some more intensely used than others. And I, too, have the 8th HT core on my Quad Xeon always more busy than others in Logic.
-
The biggest issue for me, unless I have some bug on my system, is that VE Pro seems to keep using CPU up - some kind of legacy usage or something - even when you power off an instance. For example if you have instruments, plug ins etc, and you power it off, it should (that instance) drop to zero cpu usage, but I find that if one has say four or five instances open at once, and you power them all off, there is still a residual drain on the cpu. It would be great if this didn't happen!