Preload settings in Kontakt with SSD
- Thread starter lux
- Start date
babylonwaves
Darth Fader
mine is on 12k
EvilDragon
KSP Wizard
Since I have mixed SSD and HDD, mine is at 24k. Otherwise I'd put it on 6k if all my Kontakt libs were SSD only.
Hopefully they'll add a per-drive DFD buffer setting sometime...
Hopefully they'll add a per-drive DFD buffer setting sometime...
OP
OP
lux
Senior Member
Thanks for the replies babylonwaves and Evildragon! so basically I can set it very low for SSD's. I as well happen to have a couple instruments (mostly non orchestral) on HDD's, perhaps I could set the load buffer on a single instrument level and leave a slightly higher load on HDD's instruments. Or should I set it in the general options only?
Polkasound
Senior Member
I have all of my libraries on SSD. Setting the preload to 6k dramatically reduced the VST performance meter spiking in Cubase.
Guys - I don't know how you pull this off with a buffer as low as 6k - or even reliably much less than 60k.
I have all of my Kontakt libs on Samsung Pro 850 1TB ssds (8 of them). These are mounted in two external SAS enclosures, which essentially provides eight discrete 6G Esata paths to the Mac via a pcie card (which should be no different than mounting the drives internally). The drives are properly formatted and have barely even had any re-writes - I organize my libraries on a separate drive before moving them over to the SSD's - and so libraries are added "cleanly" and permanently and sequentially.
Also, as an alternate scenario, my mobile rig has the same SSD's mounted in ThunderBolt 2 bays connected to a Mac trashcan.
What I find is that, even though the super low buffers seemingly work fine without Kontakt glitching at all, with various samples/instruments that are sustaining and have long-ish loop crossfades (i.e. Spitfire strings or even a custom ambient pad that's only one sample, but a long crossfade) these instruments will intermittently stop looping properly and cut out after the first loop when trying to sustain. Something in my memory says this is a known issue ( I think I remember Spitfire gents posting that this was a problem with super short buffer settings?) but the upshot is that, with a large template and a multitude of Kontakt instruments loaded, this rears it's head somewhere in the session and often requires reloading - until it happens ultimately again. With one/a few instruments loaded, it might not be a problem - but in any real workload, I always end up raising the buffer back up. So I've mostly given up on this as an effective solution to reduce ram usage.
If you guys know the secret I'm missing, I'd love to know it!
I have all of my Kontakt libs on Samsung Pro 850 1TB ssds (8 of them). These are mounted in two external SAS enclosures, which essentially provides eight discrete 6G Esata paths to the Mac via a pcie card (which should be no different than mounting the drives internally). The drives are properly formatted and have barely even had any re-writes - I organize my libraries on a separate drive before moving them over to the SSD's - and so libraries are added "cleanly" and permanently and sequentially.
Also, as an alternate scenario, my mobile rig has the same SSD's mounted in ThunderBolt 2 bays connected to a Mac trashcan.
What I find is that, even though the super low buffers seemingly work fine without Kontakt glitching at all, with various samples/instruments that are sustaining and have long-ish loop crossfades (i.e. Spitfire strings or even a custom ambient pad that's only one sample, but a long crossfade) these instruments will intermittently stop looping properly and cut out after the first loop when trying to sustain. Something in my memory says this is a known issue ( I think I remember Spitfire gents posting that this was a problem with super short buffer settings?) but the upshot is that, with a large template and a multitude of Kontakt instruments loaded, this rears it's head somewhere in the session and often requires reloading - until it happens ultimately again. With one/a few instruments loaded, it might not be a problem - but in any real workload, I always end up raising the buffer back up. So I've mostly given up on this as an effective solution to reduce ram usage.
If you guys know the secret I'm missing, I'd love to know it!
EvilDragon
KSP Wizard
That's actually a bug in Kontakt regarding looping and certain other conditions, like loop length, sample rate of the samples, loop crossfade, etc. basically I think DFD buffer gets overloaded or something. But it might also be a typo somewhere in Kontakt's looping code (it doesn't happen JUST with lowest buffer settings, it can happen with buffer set to 24 KB or even 60, just harder to get it to happen on those)... it's really not easy to get a 100% surefire repro case, because it rears its head randomly. It does eventually happen, but to be able to fix it, NI would need 100% repro cases, and they are hard to come by. i.e. you manage to get that bug with a particular NKI, you save it, you send it to me, and it doesn't happen on my computer. It's a really nasty gremlin.
Pietro
Senior Member
This is something I really appreciate in PLAY. It had this feature since version 5, I think. Kontakt still lacks in that field and it had a lot of time to catch up.
Hopefully, next update, though I wouldn't bet my money on it
.- Piotr
babylonwaves
Darth Fader
make sure you use the right PCIe slots, in many macs they have different speeds depending on which slot you use. on top, i doubt that you'll get a good performance with so many drives on one slot. AFAIK you don't even get 6G in total through one slot. blackmagic has a free speed test app which you can download through the app store.
EvilDragon
KSP Wizard
PCIe x4 is enough for running regular SATAIII SSDs through it... I have a bunch on one such card, no streaming issues whatsoever.
babylonwaves
Darth Fader
from what i recollect, it depends on what Mac Pro it is. older models have several PCIe 1.0 slots and only one or two with 2.0 features which are usually occupied by the graphic card. but all that is like 10 years ago, so maybe i'm simply wrong. the MP 5.1 2008 I had for a long while never managed to get the best out of an SSD on an PCIe card. I had two graphic cards, probably both on the faster slots.
EvilDragon
KSP Wizard
Yeah, probably other PCIe ports left were just x1 then.
NYC Composer
Senior Member
I think your 2008 is a 3.1. I may be wrong, but my early 2008 Mac Pro is a 3.1 and has two PCIe x4 slots, and two PCIe 2.0 x16 slots (one free with one graphics card installed.) I have three 1TB SSDs in mine, internally so they're just running SATA II, but I don't run huge sessions and I have a Mini slave. Cubase is always on the bleeding ASIO edge when I do orchestral stuff, but thank god for VEP.
However, I've never had much luck running at a 6k Kontakt buffer. My system seems more or less comfortable at around 18 or so, 24 even better. It's all voodoo and system specific (or so it seems to me.)
Yes... right - that's exactly what it appears like to me. In the end, it's enough of a consistent/intermittent problem to prevent me from reliably being able to use these lower buffers, no matter how many variances I've tried. Which is why I really do wonder... if anyone claiming such low buffer settings is actually able to use these settings in a large real-world full-blown project? Perhaps everyone is willing/used to reloading their projects when these glitches happen; here that's just too much of a flow killer for me (and there's always a terrible shortage of any extra time for these sorts of "pauses" around here as well.
) Buffers back up at more forgiving settings, and I seldom see this issue. Except...FWIW, I can tell you one scenario that will make it happen reliably (at least here) at near any buffer setting. If you load say, four time-machine pro-based instruments with long samples (like beat-mapped transition effects that play over four bars) on the same midi channel in a single K5 instance, and trigger them as a layered patch, Kontakt will almost always exhibit this drop-out behavior. I can get away with three layers usually, but not four. I've got bunches of these custom effects instruments that we've made here, and with two+ instruments set up like this, it's really easy to quickly browse these "modules" and create a custom layered transition element that lands exactly where and how you want it to.
I don't have any issues with speed readings from the SSDs via BMD speedtest, or any other benchmark tools I've used; all report streaming speeds as near ideal for these drives. Also, this happens on both the drives-via-TB enclosure on the MAC trashcan, and the slightly older cheese grater 12c - so I don't think it has anything to do with PCIe arrangement (but yes, we've confirmed that the the SAS card is in the best slot as well in the cheese grater).
I changed my preload buffer to 24kb and my template went mental (disk meter's red spike flicker despite being at 0 at all times). I tested it with a Spitfire-only template residing in an EVO SSD 2tb drive. I wonder why I can't set it to lower than 60kb. Is it because I'm changing the preload buffer in an existing template? I'm gutted. I'm using a slave btw. 46gb of 64gb of ram is being used. I'd like to bring it down though.
UPDATE: The dropouts are now occurring at 60kb buffer too, most pronouncedly in Spitfire's Chamber Strings - 'Core Techniques' patches.
UPDATE: The dropouts are now occurring at 60kb buffer too, most pronouncedly in Spitfire's Chamber Strings - 'Core Techniques' patches.
Last edited:
I often have the same issue, specially now with my recent acquisition of SCS. My disk is at 0% yet it spikes red somehow. Changing DFD does nothing for the problem. The library is still playing fine most of the time even with the spikes...
Simon Ravn
Senior Member
Mine is at around 40-50k. I tried lowering it - it saved a little RAM (surprisingly little actually, considering my template is now loaded at more than 50GB - I guess most of the RAM is actually used to allocate for voices/loading the actual patches/scripts/GUI etc) but I got clicks and pops pretty often. When my template is smaller, I can lower the buffer to a much lower setting it seems. But when I start getting above xxGB loaded, I need to raise it to avoid clicks. Doesn't really make sense but that's my experience here on a Mac Pro, everything hosted in Vienna Ensemble Pro. Your mileage may vary...
TonalDynamics
Noodler Extraordinaire
Hey dude sorry for the necro, but did you ever find a solution for this? I'm also using a hybrid setup (several 860 EVOs + 6TB Barracuda Pro) and have found 48k to be a suitable compromise for the libraries I have.
I've also read about being able to set preload sizes on a 'per-instrument' basis, but have never found such a setting and haven't looked super deep into it yet...
Cheers,
TD
EvilDragon
KSP Wizard
No solution for per-drive DFD buffer yet.
Per-instrument, first you need to of course disable global DFD buffer size override, then in Instrument Options->DFD of each instrument you can set up the buffer size. But this is really tedious.
Per-instrument, first you need to of course disable global DFD buffer size override, then in Instrument Options->DFD of each instrument you can set up the buffer size. But this is really tedious.
