1. IMPORTANT:
    We launched a new online community and this space is now closed. This community will be available as a read-only resources until further notice.
    JOIN US HERE

Kontakt 2 v.s EXS bench mark on a Quad G5

Discussion in 'KONTAKT' started by Cyril Blanc, Nov 23, 2005.

Thread Status:
Not open for further replies.
  1. Cyril Blanc

    Cyril Blanc NI Product Owner

    Messages:
    670
    Hello


    I did some testing on my Quad G5 2.5 8 GB :

    This test play the same 60 half notes on each midi channel, tempo is set to 40, 120, 240 and 360
    Each instance of 60 notes are looped on 11 bars

    Running EXS :
    1) With VM
    - I can play 480 simultaneous notes : 8 instances of 60
    2) Without VM
    - I can play 660 simultaneous notes : 11 instances of 60

    Running K2 :
    1) With VM
    - To be reliable I can play one instance of 60 notes at tempo 360
    2) Without VM
    - To be reliable I can play one instance of 60 notes at tempo 360

    At tempo 40 I can play 4 instances of 60 notes but K2 is not reliable, some times the 240 notes are playing sometimes no sound is coming from K2


    Remarks :
    - I have put only one multi for K2 and up to 11 instances of the EXS
    - Putting the tempo to 120 it does not play with K2 ; the tempo has no incidence with the EXS I went to 360 tempo
    - K2 seem to forget to switch off some note as the counter of notes are increasing
    - Logic reports load only on one processor but Activity Monitor show load on the 4 processors ( I thought that K2 was multi processor !) (I have check that Multiproc was on)
    - The first bar in the loop does not play with K2
    - If I want the first bar to play I have to put K2 down to 60 notes
    - The number of note played does not correspond to the number of voice that are displayed in the instrument

    I also did a few test on K 1.53
    - it report to play 120 note when you play 60 notes
    - it is a little bit better than K2

    Conclusions :

    NI has all my tests under there hands ; I hope they are going to come with a good Christmass present where K2.1 will be reliable and powerfull

    Best

    Cyril
     
  2. bigjobs

    bigjobs Forum Member

    Messages:
    153
    are you using stereo samples?

    does the exs count 1 stereo sample as 2 voices or 1?

    How much RAM does the exs use to stream these voice counts?

    Which precise pre loads and voice buffer settings are you using?

    What is the bandwidth of the samples?

    I have seen voice counts of over 620 stereo 16 bit 44k voices with K2. (2 hard disks 7200rpm, 1 instance of K2)

    And 512 simultanious voices from 1 hard disk, sustaining.

    P4 3ghz

    Thats 1240 mono voices.

    K2 displays 1 stereo sample as 1 voice.
     
  3. Cyril Blanc

    Cyril Blanc NI Product Owner

    Messages:
    670
    all samples used are stereo, i used the piano sound to be sure that I could play 60 notes at a time on one midi channel

    I have test with and without DFD/VM
     
  4. 63strat

    63strat NI Product Owner

    Messages:
    38
    Cyril,
    I noticed in your email that you're using a RAID setup. For streaming, JBOD (just a bunch of disks) will yield better performance than a RAID setup, as the latter retrieves data sequentially. Throughput on a RAID is great for video streams and the like, but for sample streaming, where there are hundreds of samples to be accessed at once, JBOD is better. I suspect you'll get more out of your system if you reconfigure to JBOD.
     
  5. Cyril Blanc

    Cyril Blanc NI Product Owner

    Messages:
    670
    Hello
    Thanks for your comment
    I have been talking to a lot of people about raid and JBOD

    What does the raid : spread across your drives your files at the segment level
    If you have a JBOD you will have to do that manually !
    if you put the brass on one disk the string on another ....
    If in one piece you have more strings than brass in another more brass than string.... it's a non ended story

    I have run Xbench 1.1.3* on a G5 2x2 3 GB here are the marks given by Xbench on different configuration (higher is better)

    86 with Lacie 160 GB FW800
    97 with Lacie 200 GB FW800
    103 with Lacie 250 GB FW800
    109 with my internal Sata
    122 with a Maxtor 250 GB
    156 with Western Digital Raptor 36 GB
    158 with a raid 0 of 2 Sata using Apple raid software
    160 with a raid 0 of 2 Sata using Softraid
    184 with the Rocket raid card in slot 3 and 4 Raptor 10 000 (with 1024 KB segments)
    198 with the Rocket raid card in slot 3 and 4 Raptor 10 000 (with 64 KB segments)
    205 with the Rocket raid card in slot 3 and 4 Raptor 10 000 with Softraid driver optimize for digital audio
    227 with the Rocket raid card in slot 4 and 2 Raptor 10 000 with Softraid driver optimize for digital audio
    237 with the Rocket raid card in slot 4 and 4 Raptor 10 000 with Softraid driver optimize for digital audio
    268 with the Rocket raid card in slot 4 and 6 Raptor 10 000 with Softraid driver optimize for digital audio
    ------------------


    I dont know if I will move my raid to the quad I will have to make more test.

    Buying the Quad i have decide to load a maximum in RAM.

    Today there are big problems with Kontakt and this configuration.
    Lately I have been doing many test that I have post to the person that is responsable for Kontakt at NI. They are working on K 2.1 and I just hope they are going to fix all the problems I have reported for Christmass.

    This is there last chance !! if they dont fix the problems they are going to have BIG problems !

    Best

    Cyril
     
  6. bigjobs

    bigjobs Forum Member

    Messages:
    153

    you have not said weather or not the exs displays
    a stereo sample as 1 voice or two.

    'the piano sound' - do you mean a K2 specific piano patch?
    if so, is this a monolithic patch? (Have you tested to see if monolithic files perform as well as standard audio files in folders? Does the exs have a similar monolithic format to check with?)

    If so I dont see how you could have used the same patch with the exs?

    To perform an acurate test between both formats you ideally need to use the same samples in both sample players.

    You have not stated the bandwidth of the samples you are playing back in Kontakt or the exs.

    24 bit samples require more bandwidth than 16 bit etc...
    48 k samples require more bandwidth than 44.1 etc...

    VM - Virtual Memory?


    What are the precise pre load and buffer settings you used for K2 and what are the comparable settings for the exs?

    When checking the resources on your computer during playback of the samples, what is displayed in relation
    to RAM usage and CPU usage for each sample player respectively?

    You are not providing enough concise information to justify your claim that the exs performs better than K2.

    It could be that the exs is using more RAM than K2.

    You need to thoroughly examine the system resources used by each sample player when loaded and in action.
     
  7. 63strat

    63strat NI Product Owner

    Messages:
    38
    Hi Cyril,
    Yes, I agree, that is one of the problems with JBOD -- you have to manually distribute your libraries and hope for the best. In my setup, I have 4 SATA drives, and so far things are working very well with the way I distributed the samples. I also did a ton of research on this too and was advised by many drive gurus that JBOD is the way to go. I'm not 100% sure this is true. Putting performance issues aside, RAID is a little risky in that when a single disk goes down, you lose everything (depending on which RAID configuration you have). Would like to get some data from NI on this issue, if they have run any tests with JBOD vs. RAID. It's so hard to get valid info on a subject that should have been covered a long time ago by software creators like NI, or at least by magazines like Keyboard or Virtual Instruments magazine.
     
  8. Cyril Blanc

    Cyril Blanc NI Product Owner

    Messages:
    670
    to be really fare with the test I have used the Logic EXS bank instruments so Kontakt and the EXS are playing the same samples

    1st test :

    Condition
    G5 Quad 2.5 8 GB
    Logic pro 7.1
    Setting : NO DFD
    A loop of 13 Mesure of 60 half notes for each midi tracks
    Tempo 360
    One output
    No effect added
    EXS instrument used :
    - Midi channel 1 = the grand piano on stage
    - Midi channel 2 = grand piano punchy
    - Midi channel 3 = Harpsichord

    I have manage to play 2 midi tracks of 60 notes . : the grand piano on stage and grand piano punchy

    Test done with the EXS show you could play 11 midi tracks of 60 notes

    Remarks :

    K 2 plays 2 notes for one played ! so instead of showing 60 voices used you get 120 voices used

    I did the test to add 12 more notes of the "Harpsichord" , K2 is starting to play them the number of voices increase to maximum that I have set (220 voices) then you get crackels)

    ------
    Second test :

    Condition
    G5 Quad 2.5 8 GB
    Logic pro 7.1
    Setting : NO DFD
    A loop of 13 Mesure of 60 eighth notes x 8 for each midi tracks
    Tempo 120
    One output
    No effect added
    EXS instrument used :
    - Midi channel 1 = the grand piano on stage
    - Midi channel 2 = grand piano punchy
    - Midi channel 3 = Harpsichord


    I manage to play 1 midi track of 60 note

    K2 goes crazy ; display up 360 voices uses instead of 60

    If I add another midi track of a loop of 13 Mesure of 60 half notes some note are not played you get

    If I add another midi track of a loop of 13 Mesure of 60 eighth notes x 8 ; in the Engine window note event used display 7/ 8192 note events used and only a few notes are playing


    So this test give the same conclusion as the previous test : Kontakt 2 is very slow and not reliable when it get too busy

    I hope NI will be able to fix this for Christmass

    best

    Cyril
     
  9. Cyril Blanc

    Cyril Blanc NI Product Owner

    Messages:
    670
    I use a Raid 0 just for my libs (Kontakt 2 VSL and QLSO Platinum) and I have a backup on a FW800 disk !

    it is very difficult to do a benchmark of JBOD v.s. RAID because it will depends of the product you will use !

    Best

    Cyril
     
  10. Sypher

    Sypher NI Product Owner

    Messages:
    216
    I completly agree !!
    And appreciate all the work you went through testing this.
     
  11. chriscaouette

    chriscaouette NI Product Owner

    Messages:
    42
    In EXS 1 stereo sample = 1 voice.

    Chris
     
Thread Status:
Not open for further replies.