Libraries that slow down a SSD

[kitekrazy]

  Sonokinetic phrase based loads low.  I will best testing more as I have put all of my Kontakt and Komplete on SSD.

   

 

[bitflipper]

Just guessing, but it could be that phrase libraries load bigger chunks into memory than single-note samples.

By default, Kontakt preloads only the start of each sample so it's ready to go when called upon. If the note played is short, no further disk access will be required. If you have lots of RAM, you can increase the preload cache, resulting in better performance during playback at the cost of longer load times. With phrase libraries, you're less likely to only play the first 100 milliseconds of a sample and more likely to play the whole phrase.

[AFTERTHOUGHT]

Take a look at the preload buffer size for those phrase libraries and compare them to conventional libraries. The library author sets the default preload cache size, which can then be overridden by the user. I'd be curious to know if the default cache is larger for phrase libraries (I don't have any here to look at).

[Jim Roseberry]

FWIW, No libraries actually slow down a SSD.  

• SATA SSD sustains ~540MB/Sec.
• PCIe 3.0 M.2 SSDs sustain ~3500MB/Sec
• PCIe 4.0 M.2 SSDs can sustain up to 7000MB/Sec

These speeds don't change as the drive gets full... or because any specific library.

 

Some libraries load faster than others.

A sample-playback plugin like Kontakt has to buffer the initial transient of every sample.

Otherwise, disk-streaming would be subject to additional latency.

If the library has say 40 thousand samples, each of those has to be buffered (the transient) using RAM.

 

IE: Keyscape's C7 Grand loads *really* slow.

Put it on a PCIe 4.0 M.2 SSD... and it loads... well... "less slow".

[Tim Smith]

An argument possibly in favor of physically modeled instruments?

I have both and can't tell the difference in sound. Sometimes physically modeled sounds better to my ears. 

Libraries OTOH will probably not be the best candidates for physical modeling. Seems more complex. If we take each instrument from a group to physically model and then combine them, but maybe not as bad as I'm thinking.

As to the present ordeal, I have run across it depending on the instrument and what other demands were being placed on the computer at the time. Freezing tracks always helped in most of those cases.

I have heard one singular SSD .vs multiple SSD really doesn't affect things much. I don't see how it wouldn't. 

[craigb]

40 minutes ago, Jim Roseberry said:

FWIW, No libraries actually slow down a SSD.  

With the possible exception of loading all of Bapu's plugins... 😆

[Jim Roseberry]

40 minutes ago, Tim Smith said:

An argument possibly in favor of physically modeled instruments?

I have both and can't tell the difference in sound. Sometimes physically modeled sounds better to my ears. 

Libraries OTOH will probably not be the best candidates for physical modeling. Seems more complex. If we take each instrument from a group to physically model and then combine them, but maybe not as bad as I'm thinking.

As to the present ordeal, I have run across it depending on the instrument and what other demands were being placed on the computer at the time. Freezing tracks always helped in most of those cases.

I have heard one singular SSD .vs multiple SSD really doesn't affect things much. I don't see how it wouldn't. 

Pianoteq is physically modeled... and sound great to my ears.

Not sure the technology/techniques are there yet... at least not for all instruments.

Physically modeled guitar sounds (that I've heard) sound more like a harpsichord.  Thinking of the Kronos 

[bitflipper]

On 7/27/2022 at 8:21 AM, Tim Smith said:

An argument possibly in favor of physically modeled instruments?

Yes, it is. They take up less disk space, use less memory, load waaay faster, and sometimes are capable of things their sampled equivalents are not.

The problem is that modelled instruments just haven't achieved the same level of realism yet. While lots of users say they can't distinguish between Pianoteq and a sampled piano, anyone who plays piano as their primary instrument can easily tell the difference - but only when soloed or way up front in a mix. So for most people and most applications, the modelled alternative is workable.

Personally, I'll stick with Keyscape and endure its excruciating load times because it sounds incredible. It's slow because they didn't use any of the usual tricks for minimizing memory usage, e.g. every note is sampled, so no stretching.

I have the full suite of modelled instruments from Audio Modeling. They load in the blink of an eye, take up a tiny fraction of the memory and disk space that a Kontakt library would, are expressive and can sound quite good in the right context. But naked, they often sound a bit "synthy". I still like them because they can do things a sampled instrument cannot, such as programmable glissando and vibrato speeds. 

But as Jim notes above, the speed of the drive itself is consistent, regardless of the library. Perceived slowness is a function of how much data is being loaded into memory on startup. That's why I suggested that phrase libraries might be inherently slower to load because the size of their individual samples is larger, or maybe it always loads the complete sample set. I cannot test this hypothesis myself, as I have no phrase libraries here to look at. 

 

[AndyB01]

I have Pianoteq Stage and I love its sound - which is crisp and clear, it really cuts through a mix if you're trying to showcase the piano.

However, Synthogy Ivory sounds more like an actual piano to my ears at least.

Horses for courses I guess and I use both in equal measure, but when I'm practising or just riffing for fun - Pianoteq standalone mode is my go-to.

Andy

[Tim Smith]

Not to derail a good thread, but I was looking at some of the SWAM products. I think they sound quite good. They don't emphasize what I think is a key selling point, I'm guessing because they want to present their instruments as equals and not sell it only based on size.

After looking through the manual in specifications I eventually found this:

Required space after installation: 16 MB per single plugin format + 15 MB for shared resources and assets, 240 MB for the complete Solo Strings bundle (all plugin formats). RAM occupancy: about 15 MB for each instrument instance

Wait....let me wipe my glasses. 240 MB!

While I haven't heard physically modeled guitars, I have heard Pianoteq and it is VERY convincing. Any variable in it can be adjusted. Same company now has a nice pipe organ. SWAM instruments are pretty amazing too. Can't vouch for physically modeled guitars. Maybe that tech is still catching up.

 

[bitflipper]

I reviewed the SWAM instruments a couple years ago for SoundBytes. That was the angle I went with - small size and fast load times. I've since become very fond of the sax and the other reed instruments, but prefer samples for most string instruments, often layering sampled strings with the SWAM violin to bring out a melody.

Convincing physical modeling for guitars is still a ways off, being a far more complex challenge than piano.

[Starship Krupa]

On 7/29/2022 at 10:10 AM, Tim Smith said:

I haven't heard physically modeled guitars

To get a taste of physically modeled guitar (and a ton of other useful sounds), go to A|A|S and download Swatches. It's their 560-patch fully-functional demo of all of the patch sets ("soundpacks") for their excellent line of physically modeled instruments.

Their guitar product is Strum, and there's a lite version of it out there that can sometimes be found for anywhere from $10 down to free, depending on promotions.

On 7/27/2022 at 8:21 AM, Tim Smith said:

An argument possibly in favor of physically modeled instruments?

Mmmmm, yeahhh, but there's a tradeoff, which is that physical modeling, especially as it gets more realistic and/or complex, is more "expensive" in terms of processing resources than sample playback is. (surprised Dave didn't bring this up) In most cases, I'd rather have to wait a second or two for a library to load than have the audio engine pack it in as my project gets bigger. I know I can freeze, but that doesn't suit how I create (I swap instruments and FX a lot while putting together the sound of the project).

As much as I LOVE the sound of A|A|S' instruments, I've resisted purchasing the full version of Chromaphone even at 50% off because some of the patches bog down my laptop. Admittedly it's kinda old, but it does have an i7 in it, and it runs most other things pretty well. This doesn't happen with SONIVOX Companions or Orchestools, both of which are sample-based.

Of course, if you're on a rocket sled, processing resources are less of an issue.

As Jim pointed out, NVMe M.2 drives haul booty compared to SATA SSD's, and they have a comparable price:storage ratio. If your mamaboard is out of M.2 slots, you can get PCIe adapters that will allow you to install multiple NVMe drives. I hacked the BIOS in my Dell Optiplex 7010 to allow it to boot from such a drive. Fortunately, new #1 system has an M.2 NVMe slot built in. Oh is it quick.

Still, even on the new system, loading from the onboard NVMe drive, MSoundFactory's Monastery Grand takes a while to load.

[Tim Smith]

21 hours ago, Starship Krupa said:

To get a taste of physically modeled guitar (and a ton of other useful sounds), go to A|A|S and download Swatches. It's their 560-patch fully-functional demo of all of the patch sets ("soundpacks") for their excellent line of physically modeled instruments.

Oh I have that soundpack but wasn't aware the guitars were physically modeled. Had no idea A/A/S stuff was physically modeled.

I believe this technology is worth paying a little extra if it can reproduce an original instrument with a minimum load on memory. Pianoteq is a good example of a winner. It's probably more expensive than the average sample library.

In many cases I think the way the samples are played makes a huge difference using either sample libraries or physically modeled. Guitar in particular. 

That's one misconception with  a nice guitar sampled library. Keyboardists think they can now play guitar to sound just as well as a real guitarists. Unlike some random synth patch creation, you have a single instrument with characteristics that need to be perfectly emulated. To do that with a few track passes free hand not using a robotic helper like STRUM and get all the articulations right takes some know how beyond just loading a patch and thinking you can play it. I recently had a similar experience with a harp library. Harps are played a certain way or they don't sound real.

To me, something like strum is taking too much of the control out of my hands. I would much rather program my own which takes a lot more effort, but I think sounds much better.

 

Edited August 1, 2022 by Tim Smith

[Starship Krupa]

28 minutes ago, Tim Smith said:

wasn't aware the guitars were physically modeled. Had no idea A/A/S stuff was physically modeled

That's their thing, physical modeling. Objeq Delay even involves physical modeling somehow although I'm not sure exactly how. I think it's as if you're sending your signal through a physically modeled pipe or drum or whatever, then it's delayed and mixed back in. I kinda get the concept but it's hard to describe.

I suspect that all of their instruments use the same engine, which is how they can do Swatches/Player. The fact that Player can play soundpacks from any of their instruments suggests this. Those aren't samples, they're the actual patches being played by their engine, which is why sometimes A|A|S Player can bog my laptop down. A|A|S Player trick: limit the number of simultaneous voices to 8. They often have very long tails.

37 minutes ago, Tim Smith said:

To me, something like strum is taking too much of the control out of my hands. I would much rather program my own which takes a lot more effort, but I think sounds much better.

Strum (and the Strum-based soundpacks in Swatches/Player) has 3 different modes of operation. Loops, where you trigger strummed loops, Chords, where you trigger chords, and finally Notes, where you trigger individual notes. Chords and Notes modes include articulations like damping, squeaks, etc.

So to do what you want, you should try it in Notes mode. If you're using Swatches/Player, each Strum soundpack lists the different modes and you just select which one you want. The articulations use the same key triggers as Strum and Strum Session, so if you look at a manual for one of those, you can figure out what keys do what in Player. At first I wondered why anyone who didn't own Strum would want a Strum soundpack that only played via Player, but then I sat down and figured out how to do it.

Their programming for the Loop and Chords mode is pretty good, probably better than I could do manually. I think their stuff sounds more authentic than the actual sampled guitar libraries I've heard, but I'm no expert. I don't know how, but maybe their methods allow for more flexibility than just playing back a recording of someone playing a note or chord. They can have their engine respond to velocity more accurately, maybe. Once you start throwing in the articulations (damps, scrapes, etc.) it sounds pretty cool. Even though I'm a guitar player, and my pride probably wouldn't allow it, if all I needed were some funk chirps or the occasional ambient nylon string chord or something, Strum would do a fine job.

[57Gregy]

So, when you read SSD, do you think ess ess dee or solid-state drive?

[craigb]

14 minutes ago, 57Gregy said:

So, when you read SSD, do you think ess ess dee or solid-state drive?

Studies have shown that people usually "translate" WTF and LMAO, but tend to say LOL as the three letters that make up the acronym.

(YMMV! 😁 )

[bitflipper]

wtf, who studies that sh*t, lmao? I'm guessing it's the same grant-scammers who studied why snow-shovel usage declines in July.

btw, it's pronounced "SDuh".

[kitekrazy]

On 7/27/2022 at 9:05 AM, Jim Roseberry said:

Pianoteq is physically modeled... and sound great to my ears.

Not sure the technology/techniques are there yet... at least not for all instruments.

Physically modeled guitar sounds (that I've heard) sound more like a harpsichord.  Thinking of the Kronos 

 When most people are familiar with playing an instrument you realize how fake a sampled library can sound.  There's this worry the average listener might notice this. 

[Jim Roseberry]

18 hours ago, Paul Young said:

When most people are familiar with playing an instrument you realize how fake a sampled library can sound.  There's this worry the average listener might notice this. 

Sure... as great as advanced sample libraries can be, they're still a "snap-shot" of the real thing.

Witness frowns/groans that often accompany playing Sax samples.  🤪

Physical Modeling has the potential to sound more realistic (especially in dynamic changing/evolving facets). 

Both are bound by current techniques and available DSP resources.

 

When Physical Modeling offers an experience close to 100% , (just like a real violin/sax/etc) it'll take many hours to learn to play it musically.

That will truly be "VAST" synthesis.  

 

[Tezza]

Modelled pianos like Pianoteq also have better velocity response, Pianoteq responds to all 127 velocities, sampled pianos have usually up to about 16 velocity layers, also sampled instruments sometimes tend to knock off that all important first part of the note, really noticeable on guitar sampled libraries but also noticeable on piano libraries, some do better than others.