Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards New incoming Yamaha CFX vst from VSL

Hello angmyu and welcome to PW!
The demo song is the beginning of Kapustin's Sonatina op. 100.

You can find the MIDI file here (you might have to scroll to find Howard Na performances). That is the midi version of this I believe.

Re: is the full version worth it? Well, it depends - hard to say!

I think that with the standard version you have most important mics for playing - with the full version you add another close mic, mixable with close1 (but you have to use the delay carefully to avoid strange interactions), which opens good possibilities IME. If I could have only one, I'd say I like close1 tone better though.

Then you have the surround mics if you use surround or for general ambience which are nice and sweet sounding. And another version of the mid and close mics which don't add much I believe - but I haven't used them a lot yet.

There is also a post from Karvala about this that is worth reading.

Hello, Erard. Yes, that is my first reply wrote as soon as I create an account! Thank you for sharing the information about the song and the source of the MIDI file.

First, in the case of Garritan, for example, there is Lite version that produces a defective sound in some ways, and Full version supplements this and serves as a complete piano sound. But VSL CFX is not the case. The Standard version gives a fairly complete piano sound, and Full provides additional options for the sound. In short, I was already satisfied with the 'standard' sound of Standard version. The problem is whether the additional options are worth the price.

Right now I have to rely on the demo samples you uploaded and Karvala's advice, but I'll don't know if they're worth the price until I buy it myself because this is especially a matter of taste and self-satisfaction. One thing that is clear is that the mood of full version is quite sweet. There is something about it that stimulates hearing although the use range of high & surround mics is limited and the difference in sound between mid1 & mid2 is small.

Will I pay € 165 for the space offered at various depths or will I be satisfied at present. It's time for choice.
Thank you for helping me.

I'm not even sure what surround sound really is. Generally I know is means more speakers. If I purchase the upgrade, can I use those surround presets/mics with only two speakers, or is that not advisable? Thanks.

God Bless,
David

Yes, you can use the surround channels with two speakers.
Give them a little bit of delay ( look at the factory presets for some examples ) for more air in the resulting ambience.

I also have dilemma about full version, I think I will rather keep the upgrade money and invest in Steinway D by Production voices. This huge piano vst is around the corner so maybe it is better to wait. I love standard version of VSL CFX but that Close 2 mic raise G.A.S. in me.

Just to give an update from my earlier comments, I've been experimenting a bit more with different mic settings, and see a bit more value in the additional mics. My current default setup uses five mics, of which four (Close2, Surround, High, High Surround) are from the full upgrade. I find that the combination of Close1 and Close2 is definitely preferable; Close1 on its own has too many distracting flaws and Close2 on its own, which I tried to use for a while, is not well defined enough. Put them together, and you get a somewhat less flawed close mic with better depth; I find it works quite well. The three ambient mics are there to get a room reverb in place of the algorithmic reverb; their necessity is more questionable but they give do better control and a more natural sound than the builtin reverb.

The other update is that crossfade between the main tone sample and the release sample, which has been problematic, is being revamped and it's possible a control will be added for this (I certainly hope so). It also seems as though as a control for the release samples themselves is also in development, so there may be a way to bring that room resonance under control.

It's still a borderline decision, but for what it's worth I'm happy I got the full upgrade. For me it's become the difference between using it regularly and probably putting it aside for occasional use. It may not be that way for everyone, though.

Thanks for taking to post your updated expereinces karvala.

Whilst the 30% introductory savings is somewhat tempting, this will be a more compelling VI once VSL sorts the niggles and adds some new features.

Full price and well sorted is a better value to me than 30% off with the potential for sorting, as I have enough unused software on my laptop. . .

Thank you for the update. Would you be willing to private message me your current default mic settings? Even though I can't run the full upgrade when I get it (hard drive space), eventually I will be able to run the full CFX and I'd like to have a custom preset to try out. You impress me as someone who understands the fine intricacies of tones. Thanks for the consideration.



I'm so happy right now with the Standard version I just can't neglect the opportunity to have even better or different sounds with the full library. For me the VSL CFX actually plays better and sounds more realistic than the internal Kawai sounds on my MP11SE and the other two VST's I have (True Keys/Ravenscroft and Pianoteq).

God Bless,
David

Yes, happy to share mic settings; I'll do it publicly here in case anyone else is curious or wants to share: file here. Just drop that file in whatever folder you are using for your user presets, and it should show up.

I ordered a new hard drive and it should be here Monday. Therefore, I'll be able to run the full VSL CFX. I don't know how my little MacBook air will handle some of the larger presets (>5 mics). I'm running an i7 2 GHz Dual Core, 8GB ram. My host for the CFX is MainStatge. I know I might have to tweak my settings. My goal is to run the VSL CFX with the least amount of latency as possible. Currently, I don't have any problems at 128 but that is only with 4 mics. Can someone please help me understand a few things?

1. In the Synchron Piano Player (in settings) I have the option to change both the "streaming threads" and the "loading threads." The values range from 1-16. Both are currently on 4 as the default. What does this mean in relationship to my processor?

2. Also in settings: what is the "pan law."

3. Also in settings: Is it better to have a higher preload default size (I know it uses more ram) if my processor is underpowered?

4. Finally, when I open up the CPU meter in the Synchron piano player, it gives the CPU% in Blue (upper left corner) and Peak % in Red (upper right corner). How can I intelligently use this to understand what's going on?

I know it will take sometime to answer these questions. Therefore, I want to let you know how thankful I am for the time you spend doing this. I'd gladly take you out for a meal if you lived close by.

God Bless,
David

I'll give it a shot.


It's the number of concurrend (in this case read) operations from the disc, or queue depth. With a recent, large (>= 500LGb) SSD i would use from 8 to 16, depending on the quality of the disc. I guess "loading threads" equals the number of files open concurrently during the loading of a mic perspective, and "streaming threads" the number of files open for read concurrently during playing.
Synchron software will "load" the initial portion of samples during the loading phase and then "stream" remaining part as needed while you play.


I'm not a mixing expert, but I think the pan law is the difference in volume when when panning a stereo signal from center to left or right. By panning both signals on one side, the volume would go up for that particular channel ( by 6dB for power which equals 3dB for the volume level). I guess a -3dB setting will compensate for the 3dB increase when panning hard on one side.


The more you can load initially the better - depending on ram size. That helps a lot with the disk where the samples are located, and it helps also a little with the processor.


I personally would use the CPU meter to find out if anything else in the computer is stealing cpu time from the audio driver, for example. But I think it is of limited use.

Wow, the edit time is really short now... I can't change the post anymore after a few minutes...
I wanted to add that I would gladly accept your invitation!

Again...
On the pan law, in the previous post I mixed the dB count. 3dB is double the power, 6dB is double the level. So 3dB change in power is 6dB change in level!

Thank you for taking the time, Erard.



So does loading and streaming threads have anything to do with the processor core number? Currently, I have a dual core but hope to eventually get a quad core. Should the loading and streaming values be the same?



The default is -3. So if I'm not panning I should not touch it?



Do you know what the peak % on the right represents. I'm at 100% a lot of the time while the CPU % is less.

Thanks again. What type of cuisine do you prefer?

God Bless,
David

About loading threads : quite all will be waiting for the disk to provide samples. Then we shouldn’t limit the number of thread by the number of core. The main effect will be probably that with more thread, more note samples reads will be interlaced, and with 1 thread, all reads are in sequence. With a plain hard disk, the nearest block from the previous one will be prioritised, then, with a defragmented disk, I expect a whole note to be read at once with 1 thread as with many. With an SSD, I don’t think some reads are more prioritised, then any active thread will probably read samples at the same rate. Also many threads won’t accelerate the disk which is the bottleneck. The default value could be adequate.

I didn’t manage to hear an effect but f the pan law : with centered setting and with full left mixer : 0dB and -6dB seem to produce the same levels.

Without commenting on the theoretical basis, I would just observe that I did some experiments (this with a decent 500GB 850 EVO SSD drive and an i7-4770k), and 16 load threads was substantially faster than 8 (not quite twice, but well over 50%), which itself was substantially faster than 4, so if you're using an SSD drive I'd suggest simply increasing the load threads to the maximum 16 to reduce the initial loading time (which is otherwise painfully slow if you're loading all mic perspectives as I tend to do).

I would imagine maximising streaming threads is also advisable for similar reasons, but I have to admit I haven't properly tested that and the relationship between streaming threads and CPU usage is unclear to me (the are arguments for both directions), and CPU usage is the limiting performance factor with my rather old CPU.

Just a few words about cores and threads (units of work on the CPU): even if you turn on your computer and leave it there doing (seemingly) nothing, you will have around a thousand threads working just to keep things running.
How can a CPU do that? It would seem that if I have, say, four cores then I can have only four threads (although with some tricks they can be doubled) at any given time. And guess what, this is also true!
The secret lies in slicing in time the main four threads (hardware threads) to change task every few milliseconds (spawning many software threads).
In this way you can have four hardware threads in the CPU, one per core, and thousands of actual software threads running many different tasks - in our much slower reality, all working at the same time.

Going back to disk In/Out, one thing to keep in mind is that the CPU is much, much, much faster that the disk.
A good example would be a clerk (hw thread) in a city delivery company. Let's say the clerk needs a certain number of parcels from a few places around the city to be then redelivered to customers. He will print an order, give it to the driver and then go about to work on other tasks inside the company.
After, say, two hours the driver comes back, someone unloads the goods in the warehouse and only then the clerk is called with the information that the goods have arrived. In the two and a half hours the clerk will have worked just a few minutes for this task, being free to do other things the rest of the time.

The ckerk here is a main CPU hardware thread, and issuing the request to the driver is a very small slice of time on the same thread. The rest of the workers in this task (the driver, the people who load and unload around the city and in the warehouse) are threads on a completely different CPU which is contained in the ssd controller.
So the CPU just requests the data, and the controller (which is a small computer in his own right, also multicore) in the ssd does all the rest, including loading the data in RAM directly. When it's done, the controller issues an interrupt to the CPU which then knows the data is available. In the meantime the CPU, in our case, is busy rendering piano sounds.
It's a simplified picture, but pretty close.

So, when you maximize the threads for streaming the CPU has more work to do not because of the I/O operation, but because you can have more data that the CPU needs to render as piano sound concurrently.
In general, you might want to maximize the streaming threads according to the speed of the disk (when the CPU requests data, the data is there faster), and then maybe give more time to the CPU for rendering (by increasing the size of the buffer, for example, or by rendering fewer mic perspectives, by stopping other unrelated tasks etc.) if that's needed - I don't think that "starving" the CPU on the I/O side is the right way to avoid overloading it.

Your entire last post has given me the greatest clarity on this subject so far. Thanks. However, there is one piece missing for me.

What does "I/O" mean in this context? You said:

"So, when you maximize the threads for streaming the CPU has more work to do not because of the I/O operation, but because you can have more data that the CPU needs to render as piano sound concurrently."

And

"I don't think that "starving" the CPU on the I/O side is the right way to avoid overloading it."

God Bless,
David

Sorry for the jargon - I/O means In/Out - in this context it indicates the system that reads/writes to/from the disk (SSD or otherwise).

Actually, the first sentence you mention assumes a lot and is not very clear: I wanted to change it, but the edit time is so short now!

The last part can be put this way (sorry, English is not my first language!):
If you maximize the number of reading threads (leaving everything else unchanged) the CPU doesn't work more, because reading from disc doesn't use much CPU time (compared to rendering).
In any case if the CPU is overwhelmed, you should address the problem by lowering the quantity of data (less mic perspectives etc.) or by giving more time to the CPU (larger buffer)
I don't think that lowering the number of threads reading from the disk is the correct way to lower the CPU load.

I don't notice any difference when I increase my "streaming threads." The only think that seems to work when I go >4 mics is to increase the latency in MainStage to 256.

I love Karvala's default preset. I think it uses 6 mics. My little computer can handle it at 256 latency. So I'm quite happy for now with a replacement hard drive (OWC 480GB) and the full CFX library. However, I do plan on getting a dedicated machine to run the VSL CFX and whatever they offer in the future. Hopefully a Synchron Piano Steinway.

God Bless,
David