Garritan CFX lag problems

Gamma1734

Hi folks,

unfortunately I have to report a problem with the Garritan CFX, that starts to become unbearable. First off, this problem does NOT occur for my other vst's, which is important to know. My CFX is parked on a pretty expensive nvme2 ssd, and the disk program says the ssd is fine. (maybe I should add that I recently made a 3 hour long memory test in the bios that the RAM is also without error!)

I can play without pops and cracks down up to 8 samples of delay on my PC. However, independent of that setting, the problem is that some notes feel like their audio is slightly delayed - it's not a lag for all notes, it's like with a chord I play in my right hand that it's random in which order they arrived my ear, although with other vst's it sounds like a proper chord, all at once.

Another description is that when I play a scale in my right hand or a melody, the second or third note might be "on point" but the fourth note has that slight, almost unnoticeable delay, so that I get confused because the sound arrives my ears later and then I try to compensate that by playing the fifth note earlier, which completely messes up my brain and the whole rhythm of everything. It's honestly pretty much unplayable like that. The mean thing is that that delay is so small, that I can't really pin down what exactly goes wrong, but when I play a chord with Garritan and VSL in comparison, it's extremely noticeable that yes, vsl outputs me a chord as a chord should sound, and Garritan outputs it delayed "ding-ding-ding" instead of "ding".

All these problems are also not consistent! So when I sit down and play a chord, it might be just fine, but within a piece suddenly it might work or not.

Then I noticed the problem seems much less with the compact versions, for some reason, but even there, after a while of playing, I get this problem. Then at some point I thought I found the solution but turning the reverb and sympathetic resonance and added sustain down, which also seemed to work, but I'm not sure what the correct correlation is! I've hesitated to write but it's been months since I try to solve this! I even tried to program the gui.xml thing file to add more pre cache settings options (up to 2^14kb lol) and choose them but no….

Thanks!

n-player

What has changed on your system?

Some people preferred the NTFS rather than exFAT here
https://forum.pianoworld.com/ubbthreads.php/topics/3275751/3.html

Also, on the PianoWorld forums, I remember some people found better performance with certain Samsung external SSDs (which is baffling).

Frédéric L

I don't know if it can help, but I have put a Microsoft Defender exception on the sample directory of Garritan CFX. I makes the initial loading much faster.

Jose EB5AGV

Yes, if you can, check it on an external SSD and let us know if it works differently… I don't know why but after lots of hours of tweaks and head scratching, I found the external Samsung T7 SSD was lots better than any internal SSD on my music devoted PC, so now all my VSTs samples are located on it

dore_m

Not a big expert here or anything - but I'd think that it could be the SSD/storage speed, or it could also be the ARIA player itself. maybe try another sample library in ARIA player (perhaps one of mine) and see if the problem persists. If you still see the problem, then ARIA player has something wrong.

Gamma1734

so many great replies thank you! now I need some time to hopefully try all out.

vagfilm

Did you try changing the value of the "Inst disk pre caching"? I have seen reports that small values (32kb, eg) is preferable to larger values (like Gb). Worth a try.

CeDur

Are you using standalone CFX player or plugin loaded in a DAW? I have similar issues with Garritan when loaded in Reaper and GUI is visible. The moment I hide GUI the problem goes away. It's the same on Win7, 10 and 11 and different machines.

Gamma1734

Alright, I don't have an external SSD right now, but I tried the other things you said. I changed the pre cache size to a low value (which I already tried before) and hid the GUI. It seems to work without that problem now again. I can't believe it, hope it stays like that now. For the video today I could play butter smooth again on it, it's such a difference. Thanks guys!

HZPiano

Gamma1734 it's such a difference

Glad to hear!

Cheers and… the happiest playing,

David Lai

CeDur I always hide my GUI on Reaper, only doing tweaking when needed, then I can press F on the selected track to bring the window up, then F6 to switch to the GUI interface if needed.

Gamma1734

Here's a follow up question: Modern PC's can handle very low latencies, I think I can push it down even to as little as 8 samples (with triple buffering though). I read somewhere, that 64 samples might be the "golden" setting.

Now, when I do livestreams I have to use the internal ASIO from my DAW, and that produces -way- more latency, which at first glance looks bad. However, I noticed that playing for a long time on that setting, the instrument feels different, and I even thought it felt a little bit more like playing on a real grand, because I had the feeling the keys had more "weight" to them that way. (I'm also aware of the explanation that the "latency" on a real grand is bigger than we might actually think.) So, in your experience, what's your favorite setting for the buffer?

David Lai

Gamma1734 Great question, Gama. For me, using an RME Babyface Pro FS, I love 64 samples with Garritan. In fact, I think 64 buffer size is the golden setting for me. All my VSTs run without pops and cracks.

johanibraaten

Gamma1734 That's an interesting question that I've been thinking about too as I don't find the least latency possible always feeling most natural. For me the sweetspot is 128 samples at 48KhZ I think. Sometimes I play at 64 samples and in VSL CFX that feels quite good but Pianoteq can feel almost "to snappy" at 64 and 128 or even 256 can feel better with the action feeling more substantial and a more overall "acoustic" feeling. I noticed this when I was teaching online and have to increase the buffer size just to be able to run other applications such as google meet or microsoft teams.

n-player

Gamma1734 Modern PC's can handle very low latencies, I think I can push it down even to as little as 8 samples (with triple buffering though). I read somewhere, that 64 samples might be the "golden" setting.

My RME BabyFace Pro is limited to 48 samples and that generally works OK on my old Dell laptop with Garritan CFX. A desktop machine will run better with fewer dropouts than a laptop principally for the energy management I think.

I think most of the studio RME units were limited to 32 samples. But maybe there was one top of the line unit that went lower.

If you look at the pro audio forums (gearspace), engineers don't seem to go below 64 samples; they are running specialised computers and interfaces, lots of tracks, and generally less concerned with latency. I have see a couple of guys post that they can run 32 samples, and one guy post that he got his system to work ok at 16 samples.

All that said, the ecosytem latency is really what we care about (key-to-eardrum). Reducing the sample settings becomes increasingly inmaterial in the whole scheme of things (e.g. 512-256 is a much bigger change than 64-32 samples).

The easy reduction in latency is to change from speakers to headphones, which could buy say 3ms, depending on one's setup.

MacMacMac

As you suggest, engineers are running lots of tracks ... which is taxing on the system in ways that my one-track piano is not.

So I'm having not problems using a basic Presonus Audiobox (over a decade old) at 32 samples.

Gamma1734

As a wrapup, I've never encountered the lag problem since we talked about it. What a cool community… Thank you again! I'm going to stick with either 32 or 64 samples I guess. The next "bottleneck" is certainly the velocity curve, not the buffer. I like how someone said, it's like an exponential effect, and that is true. The difference gets tinier and tinier the smaller the buffer. For the Garritan and the VPC1 I certainly like to use a softly exponential increasing curve, because I just like the softer piano sound better, I guess that is just personal preference. Nowadays I choose another point in the velocity curve for around x=80 or x=90 and let it increase rather linearly to x=110 or x=115, because that is the highest velocity the keyboard can input (although most of the times I don't really need that high velocities at all). The instrument starts to get quite spiky in its attacks for those high velocities, so I'll always adjust the velocities in the end a bit , because it feels quite impossible to be that accurate while playing. Anyways, see you next time

Gamma1734

I have to make another note on this topic, because it might help people who read this thread and it's a very general problem with a painfully easy solution.

I have a Limiter plugin on the master which introduces some latency. Deactivating this made the playability much better. I can't believe I overlooked that!

So, if you use any plugins in the effects chain, make sure they don't introduce unwanted latencies.

SouthPark

As usual - the people that make that software should be responsible. Responsible for making sure they put the most important stuff on the front page of their instruction manual in big writing. Like - the more heavy processing needed - such as limiter plugin -- the more chance of glitches or unwanted situations.

Johnstaf

I don't know how the Garritan and VSL pianos work differently, but the Samsug T5 was better than the T7 when I tried to find the streaming limits of each one. The T7 is faster for streaming large files, but the T5 is faster for non-sequential access. Even though the VSL files are large, the samples within the file are accessed non-sequentially.

I did this test in 2020, and I don't know if anything has changed since then. IIRC my findings were the same as those of other VSL users.