CPU and Omnisphere

[Gustar]

Yes, I applied all available updates and deleted everything in my /windows/temp folder. Turning off disk streaming doesn't change anything. Actually the built in Cubase SX3 cpu meter shows 80% with some spikes to 100. And that's with two patches of Omnisphere and one drumloop. Seems that my cpu is too weak for several patches at once, taskmanager shows cpu spikes as well. Ah well, maybe spectrasonics will improve the performance; if not I probably need to start saving for a quadcore cpu...

That's odd. The performance sounds worse than on my P4 3.2 GHz. What soundcard and ASIO buffer size are you using?

EDIT: Is that with the GUI for Omnisphere displaying or with it hidden?

I am using my good old Audiophile 24/96 with 128 samples buffer size. During the test the Omnisphere Display was open but not in the foreground - not visible.

That's the same sound card I tested with. 128 sample buffer performance does not do well for most intense plug-ins (Omnisphere included) that I use with that CPU and sound card combo. Try 256 or 384 samples and see if it becomes workable. Not really a fix, just a workaround I find I have to use until I upgrade my CPU.

Yes, 384 samples eases the cpu load a bit, but it's still above 90 % most of the time...

[Per Lichtman]

Hmm. Well they just posted a new update earlier today (software 1.0.1d and patches 1.0.2. with the Atmosphere patches included). Let us know if the situation improves at all with that.

[gnu23]

Hmm. Well they just posted a new update earlier today (software 1.0.1d and patches 1.0.2. with the Atmosphere patches included). Let us know if the situation improves at all with that.

I noticed the CPU meter in energyXT (v1.x) would spike on some (not all) Atmosphere (1.0.2) patch loads and Omnisphere would go unresponsive for a few seconds in the load process, usually at the 88% or 92% point of a patch load. Once I hit a note on the keyboard, eXT's meter would return to low levels and Omnisphere was just fine. It was something I hadn't noticed prior to 1.0.2, keeping in mind I am assuming eXT's CPU read is accurate and assuming eXT v1.x's VST "shell" (for lack of a better phrase) isn't part of the problem with a VST 2.4 product. I don't see it as a major issue, just something that wasn't there before 1.0.2. I haven't tried reproducing the behavior in Sonar v7 PE.

[Per Lichtman]

I used to have that happen more with Omnisphere originally, before I cleaned the Temp folder, but I haven't noticed it getting worse with the new update. It still does happen, though, both 1.0.1 and 1.0.2. I haven't looked into it as much since some of the other issues had seemed higher priority.

Well, shall we both put in our e-mails to info@spectrasonics.net about this one? I don't have enough systems around to check whether this is "normal" or not.

[Gustar]

Hmm. Well they just posted a new update earlier today (software 1.0.1d and patches 1.0.2. with the Atmosphere patches included). Let us know if the situation improves at all with that.

No change there... I wrote an email to their tech support and íncluded a link to this threaad , but they seem rather busy right now. I guess I 'll just be patient, after all it's only the 1.02 release and usually things improve over time.

I wonder why there are so big differences in CPU usage between the multis, some stack patches don't even need 30 % and are perfectly playable...

[klagga]

Quad for Omnisphere - should I buy it?

Getting a new computer, I'm looking at an Intel quad 9300 2.5 Ghz processor.

1. What performance should I expect with this processor together with Omnisphere and Sonar 7 or 8?

2. Is it a good idea to get a quad with Omnisphere, considering some people have reported problems with that?

Thanks

PS I'd rather have this a thread of its own, but I'll try getting answers here first.

[Per Lichtman]

Gustar:
Well given how much the individual patches varied in CPU usage, it makes sense that the multis would multiply that variance by several times (especially when some of them use more patches than others).

Klagga:
Sonar 8 performance is an unknown quantity, though the PR says that it will improve greatly at low latency or "high track counts". A few months ago I did some testing with single core plug-in stacking, so here is what you might be able to expect if the core balancing in Sonar 8 can utilize all the cores well.

See my earlier P4 benchmarks. Divide each % value by 2.4. That is how much CPU drain each patch will show on the core it runs on, assuming you are using the same OS, sound card and buffer size.

In other words, even at playable latencies (e.g. 256 samples), you should be able to run several instances of Omnisphere at once with multiple patches per instance. I'm being pretty conservative here because there are a lot of variables involved, but I'd say it would likely work well.

The key thing to keep in mind is that most VSTi plug-ins will only use one core per instance. In other words, if you are mainly concerned with creating huge multis to use for live work, instead of being able to run a lot of instances at once, you might be better off with dual core clocked faster vs. a quad core clocked slower.

I advise that whatever system you are planning on getting, you compare it against the ones at www.studiocat.com and www.adkproaudio.com as I've heard nothing but good things from Sonar users about both.

[krraqk]

See my earlier P4 benchmarks. Divide each % value by 2.4. That is how much CPU drain each patch will show on the core it runs on, assuming you are using the same OS, sound card and buffer size.

What you mean with it? that one Quad 2,5 ghz equals 2,4 times a P4 3,2 ghz?

Compare cpus is a very difficult thing, because in every area the performance can be different, but reading for example the cpu chart in tomshardware it seems that the quad cpu at 2,5 can be 4 o 5 times faster than the P4 3,2 ghz.

http://www.tomshardware.com/charts/cpu- ... ks,16.html

At the extreme in 3Dmax a Intel Core 2 Extreme QX977 Yorkfield 3200/400,X38,800 is about 8 times faster than a Intel Pentium 4 "E" 540, Prescott,3200/200,P35,533. Not saying that it would translate to the daw world linearly.


* ,

[aciddose]

it's only X times faster because it has X times as many cores. the newer processors have larger cache and higher fsb bandwidth, but in order to get real results out of them the test must be limited to a single core, and many tests must be taken to measure random memory access, caching, floating point performance, simd performance, integer performance and so on.

most of the benchmark software currently out there is total BS.

[dreamkeeper]

Quad for Omnisphere - should I buy it?

I too would like to know how well Omnisphere deals with quad-cores. Like I said here, TruePianos as plug-in makes use of all cores, so it's possible obviously. I also know that it's problematic with NI Kontakt. So what about Omni?

Maybe anyone with a quad could make a simple test? Load 1 instance of Omni with a "heavy" patch, play and watch the activity of the CPU cores in task manager. That'd be great. Or maybe we can get a definitive statement from Spectrasonics?

The advice to use multiple instances instead of a multi doesn't exactly fly in my book, looking at those interesting layering/stacking options in Omnisphere. Plus you can't route the mix through the same global on-board FX, filters, etc... which can be important to get a certain sound - esp. with modulated effects.

werner

[Roel de Witt]

Hi,

Multi CPU/Core implementations in plugins are tricky. There are no defined standards on how to implement it. While it works in TruePianos within many hosts under certain conditions there are also cases where it causes problems (thus this option being disabled by default). Current Quad core CPUs are unlikely to have a Core being maxed out by one instance of TruePianos so this is more of an option for feeling safe while doing live performances or when you have a slower multi-CPU/Core setup when you can't live with the option of the sound breaking up.

The main problem lies in that most hosts these days have multi-CPU/core scheduling implemented which can interfere with TruePianos' implementation. E.g. if the host assigns several plugins to one core and one of such cores gets maxed out TruePianos' sound will break up, no matter whether the other cores aren't doing anything. Usually there is an easy fix for this by disabling multi-cpu/core support in the host software but I doubt people would prefer this over disabling this option inside TruePianos.

While it makes a lot of sense to have this in a standalone application it should always be optional in a plugin, whether it works fine or not for most people. With increasing speeds of CPU, system requirements of plugins will grow when they are doing things that wasn't possible in realtime on older computers. This is one area where it makes a lot of sense to at least offer the option since it allows large groups of (potential) customers to use such software on lower-spec systems while the otherwise wouldn't be able to do so without limiting options in such plugins. This is the main reason why we implemented it. When TruePianos was released it worked fine at default settings on 1.0-1.3 Ghz CPUs but when wanting to go full out a large part of the potential customer base didn't have machines with specifications that would allow them to do so. Thus the option to go multi-core.

I guess that plugin developers will wait till one day there is an official method to implement this. Till that time it's likely that they'll for the most part will stay away from this.

---
Roel / 4Front Technologies
http://www.truepianos.com/

Quad for Omnisphere - should I buy it?

I too would like to know how well Omnisphere deals with quad-cores. Like I said here, TruePianos as plug-in makes use of all cores, so it's possible obviously. I also know that it's problematic with NI Kontakt. So what about Omni?

Maybe anyone with a quad could make a simple test? Load 1 instance of Omni with a "heavy" patch, play and watch the activity of the CPU cores in task manager. That'd be great. Or maybe we can get a definitive statement from Spectrasonics?

The advice to use multiple instances instead of a multi doesn't exactly fly in my book, looking at those interesting layering/stacking options in Omnisphere. Plus you can't route the mix through the same global on-board FX, filters, etc... which can be important to get a certain sound - esp. with modulated effects.

werner

[Per Lichtman]

See my earlier P4 benchmarks. Divide each % value by 2.4. That is how much CPU drain each patch will show on the core it runs on, assuming you are using the same OS, sound card and buffer size.

What you mean with it? that one Quad 2,5 ghz equals 2,4 times a P4 3,2 ghz?

Compare cpus is a very difficult thing, because in every area the performance can be different, but reading for example the cpu chart in tomshardware it seems that the quad cpu at 2,5 can be 4 o 5 times faster than the P4 3,2 ghz.

http://www.tomshardware.com/charts/cpu- ... ks,16.html

At the extreme in 3Dmax a Intel Core 2 Extreme QX977 Yorkfield 3200/400,X38,800 is about 8 times faster than a Intel Pentium 4 "E" 540, Prescott,3200/200,P35,533. Not saying that it would translate to the daw world linearly.


* ,

Hi krraqk,

My testing that I did was with various plug-ins on the P4 3.2GHz vs. a Q6700 2.66GHz. The plug-ins or hosts I tested were set to use only one core and then I re-tested setting processor affinity to only utilize one core. This insured that I could compare single core performance. Each test was performed multiple times with multiple boots, etc. In general, the performance of a single core of the Q6700 was 260% in single core benchmarks.

In other words, one core of the quad got 2.6 times the performance of the P4. If you could somehow max out all 4 cores to 100% and still be able to work (not very realistic at the moment) then the Q6700 would run a little over 10 times as fast (ca. 10.4 times) as the P4 3.2GHz.

[klagga]

Quad for Omnisphere - should I buy it?

I too would like to know how well Omnisphere deals with quad-cores. Like I said here, TruePianos as plug-in makes use of all cores, so it's possible obviously. I also know that it's problematic with NI Kontakt. So what about Omni?

Maybe anyone with a quad could make a simple test? Load 1 instance of Omni with a "heavy" patch, play and watch the activity of the CPU cores in task manager. That'd be great. Or maybe we can get a definitive statement from Spectrasonics?

The advice to use multiple instances instead of a multi doesn't exactly fly in my book, looking at those interesting layering/stacking options in Omnisphere. Plus you can't route the mix through the same global on-board FX, filters, etc... which can be important to get a certain sound - esp. with modulated effects.

werner

Good points. Some more actual tests with the Omni using a quad would be great.

I'm a bit concerned that it may not matter paying more for a quad, if the Omni only makes use of one core at a time anyway.

Since there seems to be a bit speculation about this going on, it would be nice with some real facts about this from the ones that really should know, ie Spectrasonics.

- How well does the Omni go with quadprocessors? Does the Omni make use of all the cores at the same time?
- What are the requirements to really take advantage of quad processors using Omni?
- Does it matter what host I use? If so, what hosts really gives me maximum CPU-power with Omnisphere?

[krraqk]

Good points. Some more actual tests with the Omni using a quad would be great.

I'm a bit concerned that it may not matter paying more for a quad, if the Omni only makes use of one core at a time anyway.

Since there seems to be a bit speculation about this going on, it would be nice with some real facts about this from the ones that really should know, ie Spectrasonics.

- How well does the Omni go with quadprocessors? Does the Omni make use of all the cores at the same time?
- What are the requirements to really take advantage of quad processors using Omni?
- Does it matter what host I use? If so, what hosts really gives me maximum CPU-power with Omnisphere?

I think it has been said already that Omni only uses one core (just like practically all of the vsti that I know, more if they are working as a plugin inside a host) Its the host that you use which will do the management distributing the processes between cores.

According to the numbers posted (with a P4) its seems that a quadcore would manage to do easily Omni multis with one core and you can run many other vstis and also other Omni instances , so it seems the best option. A 2,5 ghz quad is cheap today and you can only gain 500 or 600 mhz with the fastest duocore, so there is not much room to improve that 2,5 ghz quad mentioned.

Anyway in the end you would not use also the pc to compose music with multiple vstis and effects?

[klagga]

Thanks for the input. Got some interesting answers here too, http://omnitalk.lightbb.com/the-omni-cr ... it-t29.htm