Vienna Symphonic Library Forum
Forum Statistics

200,845 users have contributed to 43,214 threads and 259,138 posts.

In the past 24 hours, we have 2 new thread(s), 6 new post(s) and 63 new user(s).

  1. Home
  2. Sample Libraries
  3. Vienna Instruments Libraries (legacy)
  4. Computers Needed for Real-Time Full Orchestra?

  • Computers Needed for Real-Time Full Orchestra?

    A
    A
    I have a similar setup to the Mac below, except that I use DP5.11; Dual 2.7; 6.5 gigs of RAM and internal 7200 HD

    [i]TEST COMPUTER 1

    Processor: G5, 2 x 2.5 GHz,
    Ram: 4 GB
    OS 10.4.2
    Data storage: Firewire 800
    Host application: Logic 7.1.1
    Soundcard: RME HDSP MADI with 648 MADI Interface

    Latency at 512 Samples (12ms) / 2.5 GB Ram usage
    Maximum Vienna Instrument instances: 24 (CPU 90%)
    Maximum Samples loaded: 45000
    Maximum polyphony: 300 stereo voices / (CPU 70%[/i]

    So question is how many additional Macs (or PCs) will I need to be able to load a full orchestra for example in Universal Mode? I don't need access to every art but enough that I can compose freely for a full orchestra in real time and then substitute/tweak things later. Would one more computer do it?
  • J
    J
    It's sort of hard to guess because a lot depends upon the amount of MIDI data being called upon and the rate of speed required-- ie: faster notes make your system work harder.

    But based upon the stats you listed of loaded VIs, voices, and CPU load, it would seem that you could, at the very least, double that amount on a 2.5 and 2.7. The .2Ghgz difference between the two CPUs may be less of an issue as the 2.5GB RAM bump on the 2.7. This may result in another variable which adds to the difficulty of guesswork.

    The question really is what specific instruments/samples are loaded with the amount of voices and instances you posted.

    In a nutshell, to run a full orchestra on both machines you would need to run (only as an example) winds and brass on one machine with percussion and strings on the other.

    Just taking your stat of 24 instances on one machine, a good place to start would be to see if the winds you need would fit into 12 of those instruments and brass would fit into the other 12. Of course, 12 + 12 could be adjusted to any two numbers equaling 24 instances as the project requires. A similar approach would be taken for 24 additional instances on the second machine.

    Also noteworthy: which two orchestral sections on any one machine would be entirely at the discretion of the user: winds/strings + perc/brass, strings/brass + winds/perc, etc, etc., etc.

    The final questions pertain to whether what can be loaded as such would be enough or practical for your purposes where fx and some CPU headroom would be needed for mixing.

    Answer #1: Yes, you should be able to get a lot done with these two machines with a full understanding of the limits of available resources.

    Answer #2: No, with the understanding that many users are putting 3 or more machines (sometimes 8 or more) to work to run all that they need.

    Important phrase: "all that they need".

    Sort out more specifically what *you* need in the way of instruments and weigh that against your resources. Only then, reconcile your needs with your available resources. At that point, it will be easier to decide how a third machine would serve your needs.

    By this, I mean that you should load up your ideal string section on one machine and with the remaining resources load up a wind or brass section. Delete or add samples as needed to make the most what the machine can reasonably handle. For example-- you may opt not to use muted strings on a project to save resources where muted brass would be more important in your template.
  • J
    J
    last edited
    last edited

    @JWL said:

    Important phrase: "all that they need".

    Sort out more specifically what *you* need in the way of instruments and weigh that against your resources. Only then, reconcile your needs with your available resources. At that point, it will be easier to decide how a third machine would serve your needs.


    I completely agree. First, decide what you really want, and then measure it against what you can afford.

    My experience is that running 10-15 instances on a Dual 2.5Ghz G5 requires to use a 1024 buffer and thus gives a lot of latency.

    Jerome
  • A
    A
    Well in terms of needs, I am looking to run a full orchestra sketch pad (i.e., basic arts and perhaps legato/rep files for each solo instrument and ensemble at a buffer of 256), and am trying to figure out how much hardware I would need to run it.

    The stats I quoted re "TEST COMPUTER 1" are from Herb's info thread at the top of the VI topics. I am therefore a bit concerned to learn that this performance is not necessarily playing out in the field. Might I need 3+ machines to suit my needs? I Like the idea of several Mac minis (small, quiet and relatively cheap).
  • J
    J
    last edited
    last edited

    @aplanchard said:

    Well in terms of needs, I am looking to run a full orchestra sketch pad (i.e., basic arts and perhaps legato/rep files for each solo instrument and ensemble at a buffer of 256), and am trying to figure out how much hardware I would need to run it.


    Indeed. But what constitutes a sketchpad varies radically from one user to the next.

    You have a 2.7 dual with 6.5 GB RAM. Do you have *any* VSL VIs at the moment? If so, start loading it up with winds using Universal Artics and work your way down the score template to see how far you can get on that one machine.

    Multiply the instances by 2 and subtract about 15-20%. That should be about what you could get on the two machines you mentioned.

    Problem is, there's no way of knowing if that alone will be enough for you.

    Two things:

    1. I'll try to do a MB count on my G5 2.5 dual using only Universal matrices and post them for you-- that is, if you do not have any VSL VIs of your own. You'll have to let me know before I do it, though.

    2. This is ENTIRELY personal and may go against your concept-- but...

    I use Finale a lot and found that dealing with VSL VI was more trouble than I had time to deal with. When working with VSL VI, I'm always distracted by the need to get a good mix and the need to put in lots of expression-- instead of getting the basic orchestration entered and proofread. I use DP for fast note entry and port the project into Finale as an SMF.

    I realized that I needed an orchestra library in addition to VSL VI for the specific purpose of just doing first drafts and sketchwork. I found that Miroslav Philharmonik works well on a single machine for this purpose. I never present audio demos with my Miroslav sequences, but neither am I distracted from matters of orchestration as I am with VSL VI.

    I can work much faster in VSL VI if my score is completed first. That doesn't always mean that I port my Miroslav sequence into a VSL VI sequence (because the templates are as different as the way the two libraries work). What it does mean is that if I have 24 hours to deliver a score I can do it by removing whatever slows me down-- and in this case it would be VSL VI. I'm just not fast at it.

    What takes me hours with Miroslav takes me days with VSL VI. The thing that works in my favor is that I don't have to ever play the Miroslav sequence for anyone.

    With that said, perhaps VSL SE would have been the better choice, except it wasn't available two years ago when I got Miroslav. Come to think of it, I was still using VSL Pro Edition in Logic at the time (and failing miserably) in advance of the release of VSL VI Cube.

    All of this is due to the way my poor brain works. I can't really focus on the finer points of technology, coming from the world of pencil and paper, while making orchestrations work for a REAL orchestra. This is something I find to be quite different than just making a virtual orchestra mix work. For me, it's two different end goals requiring two different approaches and (in my case) two different orchestra libraries.

    I've also found Miroslav many times more portable with my laptop than the entire VI Cube.

    Whew.

    So, I'll try loading some instances on my 2.5 Dual and give you the numbers if you want...
  • A
    A
    The only VSL I have is what came with Kontakt2, which I am trying to get a full template to load, but not having luck with DFD settings, etc. I would certainly appreciate you loading some VIs to give me an idea of what is possible. Thanks

    I too came from a pencil and paper background. Writing was certainly more efficient then, but the lure of being to hear one's ideas rendered immediately with the potential of a convincing recording (if one master's programming skills -- BIG IF) is enticing. Onward I guess. I look forward to your feedback
  • J
    J
    last edited
    last edited

    @aplanchard said:

    The only VSL I have is what came with Kontakt2, which I am trying to get a full template to load, but not having luck with DFD settings, etc. I would certainly appreciate you loading some VIs to give me an idea of what is possible. Thanks

    I too came from a pencil and paper background. Writing was certainly more efficient then, but the lure of being to hear one's ideas rendered immediately with the potential of a convincing recording (if one master's programming skills -- BIG IF) is enticing. Onward I guess. I look forward to your feedback


    Indeed-- and to the hear them coming back at you with VSL is an added treat.

    Okay- let me give this a try. I need a break from doing my taxes anyway!! [:D]

    be back within the hour....
  • J
    J
    G5 2.5 Dual PPC, 8GB RAM OS 10.4.8, latest versions of Syncrosoft and VSL installers (as of 5 days ago)

    Storage: internal SATA, eSATA II

    DP 5.11, buffers @512

    17 instances loaded with L1 Articulation Combi's

    1. picc: 55MB, 893 samples
    2. fl1: 119 MB, 1919 samples
    3. fl2: 87MB, 1404 samples
    4. ob1: 84MB, 1346 samples
    5. ob2: 78MB, 1258 samples
    6. eh: 78MB, 1252 samples
    7. cl e-flat: 107MB, 1714 samples
    8. cl b-flat: 123MB, 1978 samples
    9. bcl: 128MB, 2060 samples
    10. bsn: 101MB, 1631 samples
    11. cbsn: 59MB, 946 samples
    12. fh (vienna): 101MB, 1628 samples
    13. fh (triple): 84MB, 1351 samples
    14. trumpet C: 94MB, 1511 samples
    15. tenor tbn: 81MB, 1307 samples
    16. bs tbn: 93MB, 1493 samples
    17. tuba: 100MB, 1582 samples

    Total:

    964MB
    25,273 samples

    Initial license scan time, without samples: 1m 20 sec

    Reload time, with samples (without reboot): 1m 45 sec

    CPU-- hovering @50% (occasional spikes during standstill)

    Activity Monitor:

    Wired: 1.95GB
    Active: 162.24MB
    Inactive: 2.25GB
    Used: 4.36GB
    Virtual Memory: 11.78GB

    I suspect that a lower buffer could be used in Logic, but the other numbers are likely to be the similar.

    The question remains as to how usable this template is with only Artic. Combis loaded. There are no perf leg matrices included, which is where the library shines.

    My guess that 6-8 instances with L1 Artic Combi and an L1 Perf Leg matrices (2 matrices per instance) would be more usable and fairly close to the same CPU/RAM load. 6-8 instances appear to be consistent with what I've been able to load up reasonably well.

    Pops and clicks can occur easily at 512 buffers, so 1024 often has to be used at the risk of latency, as Jerome pointed out.

    One thought: using the generic combination string patch will help save on instance and matrix counts in a sketchpad setup.

    All for now.
  • A
    A
    Thanks so much JWL. That is very kind of you to take the time to do that. Gives me real world info to chew on. Got to do some ciphering now [[;)]]
  • J
    J
    last edited
    last edited

    @aplanchard said:

    Thanks so much JWL. That is very kind of you to take the time to do that. Gives me real world info to chew on. Got to do some ciphering now [[;)]]


    No prob, AP.

    I sort of wanted to know myself as well. I'd only loaded up various blends of patches into custom matrices and never did a test quite as basic as this.

    I'm only starting out in a MacPro, but aleady I can feel a big difference especially where load times are concerned. At some point I'll try the same test on the new machine, but I'd already made a template of empty instances-- 40 instances so far. Not terribly usable by any means, but it was fun to look at!!
  • H
    H
    Some great information here.
    Just wanted to add that i knock up the score first, and use a self built soundbank with Sib 3 (or even just write out the score 'by hand' depending on requirments.), then export the midi to sequencer into a matching template. I then carry out the second part of the process, that of tweaking the 'higher quality' sounds, then bounce to audio for the third part of my working process, mix.

    It may well be that VSL SE could substitute for the soundbank, but given the current playback facility in notation programs, that means an extra bit of software to host the VI. I'm not sure it has to be that complicated.

    For laying out a score, and if required, a playback reference as one goes, the simple sample sound set (Sf2=130MB with 4 basic artics for each string section, and two each for the others) is sufficient to get an idea. My particular soundbanks come in two varieties, small orch and large orch, and i've removed all non essential instruments and sound effects. Works a treat when speedy input and at least a coarse reference is required.
    I will add to this, that are occasions when i simply lay out the score in SIB, then print it, and play into the sequencer with the score in front of me in hard copy. It's not really that much more work, and when i sequence input, i can concentrate on getting the sound right, blending, etc.....

    So for me (generally) there are three distinct stages.
    Write.
    Perform.
    Mix.

    Seperating the three may just give a clearer idea of system requirements to play everything 'live', and if you've already made the articulation decisions before loading the sample 'big guns', then you may save a lot of resource before you even begin. I think a lot of the overload challenges come from writing as one goes (nothing wrong with that) instead of writing first, then performing/recording. Let's face it, if you've already written the score, and know where you going musically, then the only instances you need to open at early stage is those you're working on,with maybe a few others for balance, then hit the 'purification' button in the VI and add the rest.
    I think the VI is a great opportunity, but the introduction of this technology also means we may have to re-examine our working methods, and tinker with them here and there.

    Two roubles worth.

    Alex.
  • J
    J
    Hey Alex-- for a long time I really felt badly for not being able to do everything at once quickly enough. It's nice to know that others think in "phases".

    I really want to be as fluent with this stuff as I am with a pencil and paper, but the other hurdle for me is Syncrosoft ^&%@*!!! It's the single most abysmal, unstable piece of tripe I've ever seen. I do fine for a while, but Syncrosoft will wait until two days before a project is due to glitch. I've often had to revert to another orchestra library, working from scratch to meet deadlines. I'll get it sorted out, but I lose my momentum and motivation.

    I'm hoping that it works better on my MacPro-- while I was doing the test for aplanchard, Syncrosoft decided not to authorize-- so the test took 45 minutes instead of 15 as it should have. Three restarts!!

    If it weren't for VSL's amazing sounds, I'd have given up long ago.
  • H
    H
    last edited
    last edited

    @JWL said:

    Hey Alex-- for a long time I really felt badly for not being able to do everything at once quickly enough. It's nice to know that others think in "phases".

    I really want to be as fluent with this stuff as I am with a pencil and paper, but the other hurdle for me is Syncrosoft ^&%@*!!! It's the single most abysmal, unstable piece of tripe I've ever seen. I do fine for a while, but Syncrosoft will wait until two days before a project is due to glitch. I've often had to revert to another orchestra library, working from scratch to meet deadlines. I'll get it sorted out, but I lose my momentum and motivation.

    I'm hoping that it works better on my MacPro-- while I was doing the test for aplanchard, Syncrosoft decided not to authorize-- so the test took 45 minutes instead of 15 as it should have. Three restarts!!

    If it weren't for VSL's amazing sounds, I'd have given up long ago.


    JW, Thinking in phrases (a good description) has been the most consistent part of my working method through many years of experimentation with efficiency, and speed. Like you i can write fairly quickly, and i found that trying to write digitally, as i go, often slowed down the flow, and gave me sense of frustration.
    I guess the psychological effect of quickly drafting out with a small soundbank first, is the 'sense of achievement' i get when the draft is finished, quickly, without getting slowed down by my inability to input digitally as fast as i write. Not everything is right, or sounds near enough, but the overall objective is achieved at the first step.

    Reading your comments it strikes me that my next stage avoids that Synchro frustration as well. (Learnt from dealing with the Cubase Redbrick dongle many dusty years ago)
    That is, once everything is written, and i'm in the next stage of 'performing', as each small section is finished, i bounce to audio. This may seem contrary to the idea of working with a VI, but i think the two complement each other, as i view VI's as a fast means of inputting, not neccessarily a multi instanced 'end to finish playback'. I can appreciate others will want everything playing at once, with the possible reasoning being the chance to tweak and adjust instantly, but most of my audio chunks are 1 or 2 bars long, and with practise, it doesn't take long to create another chunk. (In Logic for example, there is the opportunity to mute regions, so once a 1st Violin tremolo section has been 'bounced', for example, the midi region can be muted, instead of erased, and further adjustments can be made by unmuting, and re-bouncing.)

    It seems like a lot of extra work to do all this twice, (Write then Perform) but it's proved to be fairly quick and a lot less taxing on the nerves! As far as feeling bad about not being as fast digitally, personally, i'd rather be acknowledged for excellence compared to speed. If it takes a few hours more to get something done, then the time is worth it (IMHO). We write this stuff for ourselves, not as a test of workflow for others to condemn or judge.

    But then i'm one of those chaps who worries over every note! [H]
    A big thanks to you JW, for conducting this test, as I consider it thoroughly worthwhile and valuable information, and while give your results serious thought.

    And i'm also one of those fellows that thinks from parchment and quill first, and it's slower for me as well with the digital stuff.
    So you're not on your own!

    Regards,

    Alex.
  • A
    A
    last edited
    last edited

    @aplanchard said:

    So question is how many additional Macs (or PCs) will I need to be able to load a full orchestra for example in Universal Mode?


    what method do you use to actually "add" more computers?
    Can you access each VI on the main DAW computer or is it more of a "sequencer + external sample players" approach? How do you render the whole project?

    I'm on Nuendo3 (Mac).
  • A
    A
    To keep it simple, I use (and will add) external computers as sample players (K2 or Gigastudio) with audio piped into my main Mac comp's audio interface via ADAT and via midi interfaces, although many have had success with midi over ethernet.
  • A
    A
    ipMidi from nerds.de is excellent. Rock solid and much easier to configure then MOL (which requires a lot of networking knowledge).

    This is how I use a dedicated GS computer with my Mac, but all this computer switching is not fun at all, really hope for a real network processing solution some day though.

    How do you host VIs on the external machines?
  • J
    J
    last edited
    last edited

    @Abel said:

    ipMidi from nerds.de is excellent. Rock solid and much easier to configure then MOL (which requires a lot of networking knowledge).

    This is how I use a dedicated GS computer with my Mac, but all this computer switching is not fun at all, really hope for a real network processing solution some day though.

    How do you host VIs on the external machines?


    Abel:

    First of all, thanks of the tip on ipmidi. I'm going to try it out in the next day or two.

    You are using a dedicated GS computer with your Mac already, so your question of hosting VIs on an external machine strikes me as funny. It seems you're already doing it! [:P]

    The external machines are setup pretty much the same as they would when using one computer. Independent MIDI and audio interfaces are connected in-to-out, out-to-in. Both computers are aware of each other.

    External machine is loaded with VIs and routed from its own outputs to the inputs on the master machine.

    The master machine "sees" the external machine's audio and midi, and all sequencing is done on the master machine. In my setup, the external machine just sits there-- receiving midi from and returning audio to the master. There is no transport synchronization between the two machines.

    All devices are also connected via a central word clock hub.
  • A
    A
    Unlike GS3, which has a multi channel and multi bus engine,
    VI Standalone is a non multi-timbral stereo output instrument
    and you can run only one VI Standalone at a time.
    So, I understand that you use multiple VST VI instances instead.
    My question is, what software do you use to host them.
    (Cubase, V-Stack?)
  • M
    M
    Abel,
    I've been using V-Stack and it's been fine so far. I believe you can only open up to 16 instances in V-Stack, but that's far more than I need per computer. As they say, you'll run out of RAM before needing more instances.

    I wish I could get more stability out of FXTeleport, but something is amuck in Denmark with my system. I know other people run FXTeleport very well, though. Problem is for me, I have very little networking knowledge to draw on when things go wrong, so I waste enormous amounts of time trying to figure it out. I eventually had to give up and go back with the old midi interfaces.

    Best,
    Mahlon
  • J
    J
    last edited
    last edited

    @Abel said:

    Unlike GS3, which has a multi channel and multi bus engine,
    VI Standalone is a non multi-timbral stereo output instrument
    and you can run only one VI Standalone at a time.
    So, I understand that you use multiple VST VI instances instead.
    My question is, what software do you use to host them.
    (Cubase, V-Stack?)


    Ah, now I understand.

    Yes-- almost any virtual instrument DAW will work as a host, but the lower its profile, the more resources there will be for the instruments.

    Currently, I have Digital Performer on both machines with multiple instances of VI Cube loaded. But there is AULab among the OSX developer apps that appears to be a great low-profile host for AU plugins.