Kontakt CPU spikes

[NeoDavinci]

Hi guys,
I\'m still having issues with having to raise the latency to unacceptable levels (about 40 ms) to get Kontakt to play without popping and clicking. I noticed if I check the performance tab in Windows, my CPU spikes to 100% about every two seconds. I\'m running a 2.4Ghz P4, 756 MB of RAM (which isn\'t even close to maxed out since I haven\'t been running that many instruments), and I actually replaced my motherboard to get have an Intel chipset instead of a Via. Any ideas?
Mark

[Haydn]

Do you have hyper-threading enabled? Kontakt isn\'t hyper-threaded enabled and this can cause higher CPU usage as well as pops and clicks. What sequencer are you using? Sonar has a hyper-threading option that should be disabled with Kontakt as many have found that you get worse CPU usage and more pops and clicks.

[galvedro]

Just in case you haven\'t done this already:

I would check for the last driver updates for your motherboard, ie. chipset, and may be some ports, for example if you use USB or firewire interfaces (as I do). Then update your operating system, and apply all the patches that are available.

And last: if you use Windows XP or 2000, be careful with Sasser, and Blaster worms. One of the cues to find out you have one of them in your system, is performance degradation. If you happen to have one of this, bad news. Although patches can be found that remove the worms from your system, it will remain damaged, so, you will need to format your system disk, and ensure to activate a firewall when updating your system.

Hope this helps,

Anton

[NeoDavinci]

[ QUOTE ]
Do you have hyper-threading enabled? Kontakt isn\'t hyper-threaded enabled and this can cause higher CPU usage as well as pops and clicks. What sequencer are you using? Sonar has a hyper-threading option that should be disabled with Kontakt as many have found that you get worse CPU usage and more pops and clicks.

[/ QUOTE ]

When you say \"hyper-threading enabled\", do you mean only within a sequencer, or is that specifically on my motherboard? I\'m not sure how to check for that.
As for Kontakt, all I have to do is fire up GPO studio to watch it spike every couple of seconds. I don\'t have a sequencer yet. I typically use GPO with Sibelius, but Sonar is on the way. I just checked online and my board definitely does support hyper-threading.

My drivers are up to date, and I don\'t have a virus. /ubbthreads/images/graemlins/smile.gif

[Haydn]

You need to go into the BIOS setup program when your computer first starts before Windows loads. This is usually the Delete, F1 or F2 key depending on the motherboard. It will usually tell you which key to hit when first booting up. Once your in the setup program, you\'ll need to look around for this option. Set it to disabled. Each of the BIOS setup programs are different. Make sure you save your change when you exit the setup program. Reboot and test your system to see if it makes a difference.

[NeoDavinci]

Thanks, I\'ll try that when I get home tonight. I just had a thought too. Does GPO Studio require higher latency than VST instruments in Sonar would? Is it possible I\'m expecting too much out of the program?

[Hardy Heern]

Mark,

I certainly don\'t want to teach you how to suck eggs but Hyperthreading requires a CPU AND a compatible motherboard and you\'ve only mentioned that you have a Hyperthreading motherboard. Just check that your CPU is a Hyperthreading version.

For interest\'s sake, Cubase is designed for Hyperthreading and tests show it to benefit speeds by around 10%. I have Cubase and Kontakt so it sounds as if it\'s going to be difficult to satisfy both from what I read here! /ubbthreads/images/graemlins/smile.gif

Frank

[NeoDavinci]

Hmm...I\'m not sure about that. And to think I built this machine myself.
Anyway, the processor is a P4. I checked my original purchase receipt (from 2002) and it said: Intel P4 2.4Ghz 512k Cache,533FSB SocN 478
I\'m not sure if that allows hyperthreading or not, but I would guess no?
Mark

[Haydn]

That would be a non hyper-threaded model. Only the 800 MHz FSB CPU\'s are hyper-threaded. Your motherboard should have hyper-threading disabled then. Next item would be to set your buffer size larger.

[NeoDavinci]

How does raising the buffer relate to performance/latency? If I set the buffer at 128, but 10ms of latency, how is that different than if I set the buffer at 256 and 10 ms of latency?

[Haydn]

As you raise the buffer settings you get higher latency. Going 128 to 256 will double your latency.

[nexus]

Problems like this are why I went with an AMD proc (Barton XP3000+). The Nforce chipset is very nice for audio work.

HT seems to be a clever marketing gimmick by Intel and so few programs make use of it.

Basically the whole P4 series has a math problem of some sort. This is the basis for the famous \'de-normalization\' bug.

[NeoDavinci]

So if it tells me I\'m getting 10ms of latency and I\'ve got the buffer doubled, it\'s actually 20ms? Or is it trying to compensate to get the latency I tell it I want when I double the buffer?

[galvedro]

Latency is caused by buffer size. So doublig buffer size, you are doubling your latency.

[NeoDavinci]

When I go into my ASIO settings, there is both a buffer setting and a latency setting though. How can I tell what my actual latency is if the buffer setting is the determining factor?

[galvedro]

I\'m confused. They must be coupled in some way, this is mandatory. The options I can think about, right at this moment are these:

a) Both controls afect two different buffers, may be one of them in the hardware side, and the other inside your driver.

b) Both controls are joined by software, so you can chose different buffer sizes for granting one latency value. You can easely check this if the choosable range on one control varies when you change the other.

c) One of the controls take precedence of the other, then if you chose 20ms latency, then the specified buffer is allocated but only a small part of it is really used.

I can\'t tell you much more. I supose your audio card manual doesn\'t say anything useful at this respect, right?

What I can tell you is that if you can setup a buffer of 2048 samples, and a latency of 10 ms, then there is something wrong, because a 2048 sized buffer means:

2048 [samples] / 44100 [samples/second] = 46.44 ms

and this is the MINIMUM amount of latency you will get. There are additional factors that could be taken into account. But, as you can see, buffer size is the critical factor that determines your latency.