Hi,

Just built a system for a composer and I'm testing it, and wondering what kind of performance should I expect, and if there are any ways to improve it.

Here is the system:

Windows 7 64 Pro

i7 3.4Ghz 2600 Sandybridge with Corsiar CH50 watercooler

Gigabyte 67A-UD3H motherboard

16 gig ram

2 Velociraptor 10,000 drives for samples

2 Hatachi 7200 drives for samples

Corsair 256 gig SSD for system

2 SIIG Sata 6 PCI e cards for better Sata performance

The samples are spread out over the 4 main drives for as balanced a load as I can get. The master computer is a current 12 core 3.2 Ghz Mac Pro running Logic.

Questions:

1) Looking at this setup, what kind of performance can I expect? Does anyone have any hard ideas of data rates required for streaming a single voice, or any technical information I could look at to get a "this is a best case/worst case scenario for performance with this system while composing for a movie?

2) Is there any suggestions to make things work better, like sample placement across drives, ram amounts, I know SSD drives would help a bit, but how much really? On my "lets see how much I can do to make the audio break up" stress test, I was able to get about 54 voices going before I heard the audio breaking up, but this template has quite a few "layering" and EQ/reverb tricks to make it sound larger, so I suspect it has double or triple the 54 voice estimate I can make.

3) Is there a way to slow down the cpu meters numbers? I can see it's getting high, but I can't be sure if it actually topped out.

4) On a 4 core system like this, how does the thread instances really get used, since 4 of the 8 cores aren't real? Are the multiprocessing thread numbers divided between projects, or is it per instance of the server app?

It seems to be difficult to get and find this information on the internet, so I thought this would be the best place to ask.

Thanks for your time,

George Leger III

Depends on what you or the composer wants to do.

UIess you aim for state of the art high end including MIR and brute force arrangements, I see no problems at all with this setup.

I mean, look at the configuration : 2x Velociraptor + 256GB SSD.

What else do you want ?

The only improvement i could imagine would be a Raid setup for speed optimization ( like Raid 0 ) or additional SSDs. On the other hand the Velociraptors are already top notch drives.

In my opinion you should absolutely be ready to go.

Which audiointerface do you use and what is the desired latency?

In the end we still talk about music production, not deconvolving satellite pictures for military applications.....

Greetz

kh

Hi,

We have been testing for the last couple of days. Here are a few things people might like to know about VSL, VEPro, and the new i7 2600 CPU...

1) There is a problem with the intel 67 motherboard chip designs. We had to get 2 sata cards to resolve it. IF you decide to try this design, please be aware of this issue.

2) We have a system with 6 drives: 1 SSD, 2 Velociraptors, and 3 7200 rpm drives, and all the data spread between them. We have about 100 meg a second of data coming off the drives, which are all formatted to a 64k data size (sorry, I'm so tired I can't remember what the name of it is right now)... so it's about 600 meg/sec of drive access speed when we tested using Sisoftware Sandra Lite.

3) We have created a massive project, with all 5 string groups spread across the keyboard (think Epic Orchestras string ensemble, but with a number of libraries layered). 1 note triggers 48 mono voices when everything is taken into account (number of libs, layers, crossfades, etc). We actually think it is a bit more, because there are a few other libraries added for texture (LASS, Symphobia, SAM Classic Brass)

4) We turned off hyperthreading and turboboost, and have no reverb being triggered in VEPro, our host, streaming via gigabit ethernet to a Mac Pro from a newly built PC that cost less than $3,200 to have custom built by me.

We are able to play about 80 notes doing 1/8 notes with a 1/16th note rest between to trigger the release samples.

So: 80 (notes) X 48 voices per note (3840) X 2 (release samples that overlap with the samples when the keys are released, for a grand total of...

7680 mono voices at 1 time, no crackles, dropouts, or distortion of any kind. At this point, we do run out of CPU.

We were able to get this up to about 100 notes playing slow lines, or about 9600 voices, before we would get consistent noise.

A point of clarification: We are thinking that, when you play a note (2 voices) with 4 velocity crossfade layers, you would be playing 8 voices to achieve a single note. If someone from VSL could confirm this, we would appreciate it.

And the We, for the record, is myself, George Leger III, and the person who actually programmed and created for my client in VEPro a massive keyswitched project that emulated the Epic Orchestra's "Appassionata Strings" Matrix, using VSL Appassionata String, Chamber Strings, and Solo Strings, as well as a few additional 3d party Kontakt libraries. Her name is Vanessa Garde, and she has been invaluable to myself and my client. IF you need a person to help you get your VSL templates going, I'd HIGHLY recommend her.

I really hope this information will help someone get a realistic idea of computer performance with data streaming using VSL, VEPro, and Kontakt, on a custom computer with 16 gig ram, W7 Pro, and a batch of drives... 

Good luck,

George Leger III

Hello George

Very interesting report, thanks 

I have a few question for you :

How much memory left on you W7 computer when you have loaded everything ?

Do you just run Logic on your Mac ?

Do you have SE plus on your W7 computer ? 

Best

Cyril

Hi Cyril,

We were running about 1/2 or so of the 16 gig for one instance of the massive string project on the VEPro PC. I can't be 100% sure, I'll check tomorrow (We are going to add Brass and WW projects to the Strings, making a huge Orchestral Metaframe).

The Mac Pro is a current 12 core with 24 gig of ram, and we have a 32 bit VEPro server with 3 projects in it, a 64 bit server with 3 projects in it, and a bunch of EXS and Kontakt libraries running in Logic, so it's actually pretty busy. The template also have all the FX return busses (64) being used, and live, a few reverbs, and we are using 1/2 of a core when Logic is sitting still.

The VEPro PC only have the big libraries, no SE editions at all.

It's actually a system I designed and built for a local LA composer. Between shows we add and tweak the system to get it working better, and hopefully making things easier (once they actually learn and get comfortable with the new work methods). it's going well, but I have to be honest, getting real info about this type of system is none existent. I get general comments, but no real answers. So, I finally spent a day testing and getting a "real world" idea of what I could expect. Maybe I can start the dialog and others will add their experience, so people can get some kind of an idea of what to expect before investing their $$$ into a system that might or might not work.

In a few weeks, I'll have a good idea of how well this went, the movie will be done, and I'll get feedback on how well it went, if it saved or cost money while the score was being created, and if it made a real difference in how the music was received by the director, or if it added to the realism for the composer (who actually has a majority of his work performed for the score by a professional orchestra anyways, but not all the time).

George