New test version

All about the development and use of SynthFont2
Admin
Site Admin
Posts: 671
Joined: Tue Mar 17, 2015 9:22 am

Re: New test version

Post by Admin »

But thanks to you I again found a flaw in the SFZ code: for a parameter like "cutoff" SynthFont did expect an integer and didn't know what to do with a floating point value (like e.g. 500.1) ! How silly of me. I have fixed that, and will have to go through all other parameters as well.

Rain is pouring down here in Finland so I will have some time today for fixing bugs.
KaleidonKep99
Posts: 14
Joined: Tue Mar 17, 2015 7:42 pm

Re: New test version

Post by KaleidonKep99 »

An export with the new version of SynthFont! The audio is awesome!

Keppy's Steinway Piano 3.0 demo on Reach for the Moon ~ Immortal Smoke 5.5 Million notes (Revisioned version of the SFZ, thanks to Kenneth): http://puu.sh/j5HkT/0dbcafc37b.rar

The filters are working great now! Thanks again! :D
ray890
Posts: 40
Joined: Tue Mar 17, 2015 4:27 pm
Contact:

Re: New test version

Post by ray890 »

There's a bug introduced in 2.0.2.0 that I noticed but haven't focused much on until someone else in the community discovered and brought the same issue up to my attention. Basically, once the program is in "expired trial" state, you press "x" to cancel, which is normally supposed to close the program. There is a "error report" message that pops up for half a second then closes right before SynthFont finishes closing before you have a chance to press send, and in some machines this could also trigger a appcrash along with an access violation.


As a sub-issue, ever since the newer version the user cannot enter and accept a registration code if it is manually typed, as opposed to pasting the contents from the email, which personally I see this as an annoyance, but if this happens to be intentional I'd understand that too.
ray890
Posts: 40
Joined: Tue Mar 17, 2015 4:27 pm
Contact:

Re: New test version

Post by ray890 »

Oh another thing, for a while, I've suspected there has been some performance regression with the new versions when it came to varying sample rates. I've finally got around to doing my testing, rendering with my i7-4810MQ Windows 7 laptop .wav using this sample midi and GMGSx.sf2, let these results speak for itself:

Version 2.0.2.2:
• 44100hz = ~13 seconds
• 8000hz = ~19 seconds
• 500hz = ~79 seconds

Version 2.0.1.1:
• 44100hz: ~3 seconds
• 8000hz = ~1 second

Version 2.0.0.10:
• 44100hz = ~10 seconds
• 8001hz = ~3 seconds
• 501hz = ~0.5 seconds

To be fair, I'm specifically talking about the fact that in the new version lower sample rates seem to lower performance; the opposite effect of the older versions which seem to increase performance with the lower rates.. NOT the fact that rendering time average is longer, as I think I'd blame that on the increased accuracy of 2.0.2.0+ (I could be wrong though).
Admin
Site Admin
Posts: 671
Joined: Tue Mar 17, 2015 9:22 am

Re: New test version

Post by Admin »

I cannot reproduce this. This is what I have got, on my rather basic dual core, Windows 7 PC:
The "RenderExec time = xx" (seconds) is displayed in the DBWIN32 window when playback is over. You can try it yourself.

For version 2.0.1.1
Playing to file
44100HZ: RenderExec time = 17.30
8000Hz: RenderExec time = 4.48

Playing to test (no file, no output);
44100HZ: RenderExec time = 15.60
8000Hz: RenderExec time = 3.70
NOTE: Maximum voice count is ~188

For version 2.0.2.2
Playing to file, no Voice limit
44100HZ: RenderExec time = 75.21
8000Hz: RenderExec time = 13.75

Playing to file, Voice limit=200+Cutter
44100HZ: RenderExec time = 23.17
8000Hz: RenderExec time = 6.94
NOTE: Maximum voice count stops at ~197 (even with the Cutter)

So, yes, version 2.0.2.2 seems to be somewhat slower than version 2.0.1.1, but not much. This is something I will have to look at some time later.

Putting the Voice limit to 188 (YOU can't do it, but I can) and playing to test gives this:
44100HZ: RenderExec time = 16.76
8000Hz: RenderExec time = 4.60

I did not do any tests at 500Hz this time. That would be of academic interest only.

The results are exactly as you would expect - you get faster results at 8000Hz than at 44100Hz.

The other issue you mention is something I could fix easily. I remember that you mentioned it some time ago but I forgot all about it. It was a piece of cake. I have also considered adding two new buttons to the registration form, one for pasting the name only and the other for pasting the code.
Cos
Posts: 14
Joined: Wed Mar 18, 2015 2:32 am

Re: New test version

Post by Cos »

Hello everyone. I see Synthfont is evolving in the right direction and almost there, which makes me happy, and maybe Kenneth is happier than everyone else because of that. :lol: I've been using Synthfont quite a lot as of late. I've found a couple of bugs.

I found two bugs:

- Copying and pasting files from the program don't work like before, or don't work at all. Synthfont navigator is one of my favourite features of the program to isolate my preferred songs and stuff like that. Only way I made it work was dragging and dropping a file from Synthfont to a folder and then when you select Copy in the Synthfont's "explorer" it seemed to work again. But then I tried it tonight again and this method didn't work this time around. Copy seems to be ignored in the program, but if you had a file in the clipboard from a previous Copy operation that file is copied instead of the file you want to copy from Synthfont's own explorer.

- Setup and Options settings aren't saved between sessions in my Synthfont 2 version (I have the latest one). "GM2" and "Yes" for SysEx messages aren't saved between sessions, nor your settings for the device of choice, nor the checkbox of "Use selected output instead of ANY Soundfont or VST instrument when no Arrangement exists"...

Have a good night. Keep up the great work, as always, Kenneth. No better program for actual and correct use of MIDI patches, I don't know how it works, but it works like any other program out there for me. A "Rain" instrument is a Rain instrument whether you use GS or GM2, and that's unique.

Take care of you and stay safe people
ray890
Posts: 40
Joined: Tue Mar 17, 2015 4:27 pm
Contact:

Re: New test version

Post by ray890 »

I've just gotten around to being finished my performance times on two of my machine, first one being my main laptop with an Intel i7-4810MQ, and the second one being a desktop running on an Intel Core2Duo E6700, both are running Windows 7 x64:

Version 2.0.1.1:
Rendering to file:
Machine #1:
44100hz = 4.03
8000hz = 1.59

Machine #2:
44100hz = 6.16
8000hz = 1.51

Playback:
Machine #1
44100hz = 79.58
8000hz = 76.21

Machine #2
44100hz = 76.56
8000hz = 76.21

Version 2.0.2.2:
Rendering to file:
Machine #1
44100hz = 13.51
8000hz = 45.16

Machine #2
44100hz = 22.49
8000hz = 4.01

Playback:
Machine #1
44100hz = 80.34
8000hz = 98.17

Machine #2
44100hz = 81.24
8000hz = 80.28

Hmm.. Seems like it's my laptop in particular that's having this issue, I'll investigate this and try to see if the culprit is with my laptop, or SynthFont itself.

Edit: BINGO! I've figured out that this is entirely caused by an apparent issue in this new version where every additional CPU core (after the second) decreases overall rendering performance since 2.0.2.0. I knew something felt a little off when CPU usage seemed rather low lots of the time. Currently this issue seems to disappears if either "use multiple cores when available" is disabled, or the Synthfont processes affinity is only set to 1 CPU
Last edited by ray890 on Mon Jul 27, 2015 4:08 pm, edited 2 times in total.
Admin
Site Admin
Posts: 671
Joined: Tue Mar 17, 2015 9:22 am

Re: New test version

Post by Admin »

How many cores does your laptop have? My test PC has two and doesn't behave like yours. Using both cores is faster than just one. When I'm back I can of course prepare a version where you can choose the number of cores to use.
ray890
Posts: 40
Joined: Tue Mar 17, 2015 4:27 pm
Contact:

Re: New test version

Post by ray890 »

It's an Intel i7-4810MQ, so it's a quad-core, hyperthreaded, SynthFont detects this as 8 CPU cores.

- At 8000hz, I started off with only one CPU core enabled in task manager (CPU0), CPU percentage sticked to 13% which is completely normal and healthy, I was given a render time of ranging between 4-6 seconds on average (seems to vary each run).
- First and second core (CPU0+2), CPU usage percentage ranged between 12-17%, I think I got slightly faster results that time.
- Adding hyperthread of the first core (CPU0+1+2), CPU percentage ranged between 13-20%, and got around the same results.
- Added second core's hyperthread (CPU0+1+2+3), got CPU usage between 13-18%, about the same results.
- Third core introduced (CPU0+1+2+3+4), and things started to get interesting, CPU usage was between 8-17%, and I got 9-12 second render time
- Third core's hyperthread added (CPU0+1+2+3+4+5), no noticeable difference
- Fourth core introduced, got between 7-21%, 16-18 second render time
- Hyperthread of the 4th added, no real difference

While 2.0.1.1 seems to have not have much trouble at all with speed with all of my cores/threads enabled, using all of my cores minus the hyperthreads (most of the time); ranging between 13-60% cpu usage.


Long story short, it seems to only really kick in after the third real core is introduced in the mix and gets worse with the 4th, meanwhile hyperthreadded cores don't seem to really do much to this bug. You should get a hold of a quad-core to test this, trying various process affinities, disabling and re-enabling single core mode, and watching the CPU usage percentage on the process.
Last edited by ray890 on Sat Aug 15, 2015 4:02 pm, edited 2 times in total.
KaleidonKep99
Posts: 14
Joined: Tue Mar 17, 2015 7:42 pm

Re: New test version

Post by KaleidonKep99 »

Well, I'm still waiting for new betas to try! xD
Post Reply