CM or Dietz (or anyone at VSL) :).

Do you know or have direct knowledge if say an 8-core AMD chip, can outperm a 4 core Intel i7 chip running at the same core speed?  Specifically with use to MIR Pro/VE PRO 5.

I am just wondering if the extra cores really help on this application, versus fewer cores with a possibly higher core speed.

Thanks,

Maestro2be

@cgernaey said:

WOW!!  That will be a beast lol.  That has got to have one enormous price tag lol.

I was thinking of something.  I could also take my Dimension Sections, and put them on seperate drives since I have 4 SSD's.  I could put Violins on one drive, Violas on another etc..  This might actually work (but I haven't tested it yet).

Maestro2be

Putting them on seperate drive will not help as the instrument are loaded sequentially (confirmed lattely by VSL support)

it's better to build a raid 0

Awesome, thanks Cyril.  Glad to hear that is confirmed actually.  RAID 0 adds enormous speed benefits in many cases but I have usually not done it due to the downtime when the drive fails.  However, in this situation, I would be down ANYWAY if my 1 drive failed.

I think I will order a RAID controller and then a second identical drive and move them over to that and see what happens.  Even if it doesn't completely solve my issue, I could use it in my new system when I built it.

Exactly what model RAID controller are you using and what kind of speed/throughput are you getting from the read access on it?  I know you said it was a rocket raid, but exactly what model and brand are you using?

Maestro2be

There is very little chance that the raid fail as you are going to read it 99,99 % of the time

You just need to backup your lib on HD(s) in case of !

 I use the Rocket Raid sata III card  640 made for MAC, HighPoint tech has card for PC and MAC

@Another User said:

256 GB ssd (read = 269 mb/s)

Raid of 4 x 64 GB ssd (read = 711 mb/s)

@Another User said:

Thanks Cyril,

I think I will upgrade my processor to the 6-core extreme and then add a raid controller and RAID 0 these drives.  I just haven't decided if I want to raid all 4 x 512GB drives and just put every sample library company I own on that RAID 0, or to have separate RAID 0's.  It's still much cheaper than building a complete new system and see how far I can take this system.

Maestro2be

Dont forget that more drive you have, faster the Raid 0 will be, of course limited by the speed the channel 

It will be great if VSL could do like the EXS24, have a counter of "samples not loaded in time "

So I have beyond a doubt figured out this is due to the processor and not hard drives.  I decided to reduce Studio One down from 6 threads, to 2 threads and it was still working at 19% CPU which is fantastic.

I then took my VE PRO 5/MIR PRO session and upped it from 2 threads to 4 and it only was hiting 75% CPU now with all strings playing.  So I am definitely going for the maximum CPU for this motherboard and calling it a day.  I can't really work with doing freeze tracks because I use y Bricasti all the time and it takes hours to freeze a song (because it freezes one track at a time, in real time).  If this all works out with the CPU upgrade though I will probably just use Hybrid Reverb instead of the Bricasti during the working process and then after all tracks are bounced do the final mix of audio files to the Bricasti instead of the Hybrid reverb.

As a side note, I upped the thread from 2 to 4 and got over 25% gain in performance.  When I did from 2 to 6 thread It was only 30% so the returns are greatly diminished after 4 it seems.

Maestro2be

This has been a great discussion.

---

I basically had the same question as Maestro2b:  I've waited for years to upgrade to a set up  that can handle a full Vienna orchestra, Digital Performer, and MIR on one machine.  (My old setup requires three machines, and the master computer has given up the ghost!)  

---

I'd like to buy one of those amazing new "trashcans" which use Xeon E5 Ivy Bridge processors.  I am trying to map what was discussed here to the right number of cores on the new system, using Geekbench 3.0 scores to see how they compare.

---

Andyjh:  Your software setup sounds very close to my goal: full orchestra, MIR, additional software (in my case... my DAW Digital Performer). You said you have an i7 (960) 8 core, but everything I read says 960 is a 4-core.  Wondered if you have a dual processor machine.  In any case, I used the Geekbench 3 browser and the only Geekbench 3 scores I see are for 4 core.  Have you tried running Geekbench?  Or any other info on where it might stack up?  (http://browser.primatelabs.com/)  I may just not understand how to read the numbers in your case.

---

Dietz:   Your machine, with an i7 (3930K, Sandy Bridge) 6 core - looks to have a Geekbench score around 21516.  That seems to place it close to a New MP with Xeon 6 core.

I just ran Geekbench in tryout mode (which is 32-bit only, so I skipped the memory benchmarks). These were the results:

--------------------------

Windows 7:

Integer Performance:

Single-core 3111

Multi-core 21635

Floating Point Performance:

Single-core 3139

Multi-core 22453

-------------------

OSX 10.8.4:  

Integer Performance:

Single-core 2985

Multi-core 21018

Floating Point Performance:

Single-core 2974

Multi-core 21443

----------------

HTH,

My 6 core 980X came in the mail so I will be testing it hopefully this weekend to see the difference in speed and power.

Stephen,

Keep in mind I have always run on one single machine.  It is only with the introduction of Dimension Strings and Brass that I hit a limitation now, due to so many individual tracks versus the old style which were entire sections in one track (appassionata, chamber etc.).

One machine is very easy to do with MIR Pro and all those things (vienna suite), Cubase 7.5, Studio One v2.6 etc.  It just depends on how many actually tracks you intend to use without "freezing" any of them.  I setup this week individual group tracks and audio tracks for all my sections so that I can freeze them into audio tracks as I go to free up the remaining power.  So whatever results I get this weekend, I will be making work for me as I have decided I won't buy a new machine as it's not necessary.

When I am not playing all the dimension products my machine is sitting at about 10-20% with 40 instruments playing and MIR Pro, with full vienna suite presets on instruments.  And with a complete FX chain on the master output and a Bricasti M7 send.  This machine is doing more than I did before with 3-4 machines and I am not ready to give it up yet.

I do have a 4930K motherboard system I purchased to actually use for internet and gaming ;), that if all else fails, I could buy memory for that and it would add an enormous upgrade as well.  I just haven't talked myself into this yet.  Let me receive pops and clicks all weekend long this weekend and I might be singing another song on Monday :).

Maestro2be

Thanks for the input!  

Maestro2be: Good luck and have fun getting the new system running!  Eager to hear how it goes.

Dietz:  Thanks for running the numbers.  So, your machine is faster than I had estimated.  The dual processor score seems to put it mid-way between a 6- and an 8-core NMP.  (A seven core? ha!)

So the results weren't as good as I were hoping, but certainly what I expected to see.  VE PRO/MIR PRO didn't really respond very much to the additional threads.  I moved it from 4 threads to 8 and only seen about a 10% CPU reduction when adding 2 more cores/4 threads to it.  I sure was hoping for more than that.

I am able to now work at 512 samples latency but I am hitting about 85-95% CPU so I won't push my machine that hard as it will run to hot.  I haven't fully decided what I will do yet but I am not entirely happy with the results.

One option I have is to build the second machine I have that has a 4930K processor and use it to process MIR PRO.  I just can't stand all the latency.

Maestro2be

@cgernaey said:

[...] I just can't stand all the latency.

 Why don't you just bypass the one instance of MIR Pro which is used for the instrument you plan to play in real-time?

Hi Dietz,

I will give that a try Dietz.

I have found for me that playing them in realtime, within the room itself gives me a much more realistic performance than when I don't.  If I take it out of the room etc. then I tend to get either a to aggressive performance, or one that isn't enough.  If I play it back in it's pretty much end result sound, I find that I naturally just place it with the right dynamics the first time, versus going back and having to tweak so much after.

I even thought about making some cheap Vienna Suite Reverb Busses to just use for initial recording, and then do the actually mixing/final in MIR PRO.  I just have to try this out and see if it gives me good results in the end, with no added time.  I could simply put a few instances on the busses of VE PRO 5 and then disable them, and re-activate the MIR PRO for each instrument and see how that works.

It certainly would reduce my CPU usage by enormous ammounts.

Thanks,

Maestro2be

@cgernaey said:

I just can't stand all the latency.

Assuming a sample rate of 48K, then 512 / 48000 = ~1ms latency. So why is that such a problem? 1ms should be un-noticeable. (Or is there something I'm missing?)

PS. If I was you, I'd definitely build that 4930 machine for a 50% (in theory) improvement. (Trouble is, the reality rarely lives up to the theory).

Hi Phil,

The latency is much more than that simple math formula.  There is the latency buffers that are more than doubled by adding VE PRO and MIR into the mix.  If you use VE PRO with MIR PRO, you would see that it has it's own latency settings for MIR PRO and also a buffer multiplier on the front connection page which multiplies by 0-4 times the latency settings.  Mine is set for 2 times latency already.

But as far as a single 512 latency assuming you are using nothing but an instrument track with no reverb or effects, and running it locally in your DAW there is still an amount of latency that I can clearly detect.  It is certainly much less, but I am a pianist and it's very noticeable when I perform on it first, and then move to my DAW and try to use Vienna Imperial.  It's also very noticeable when working with drums that tend to be much faster responding and instant type instruments.  Strings etc. are already a little sloppy so it's ok for those.  Some people don't have an issue with the latency and others do :).  I am one of the people who does :).

I tell you what, I am certainly interested in the idea of building the 4930k and upgrading the memory etc. to see the results difference, but it gets very old wasting so much money to not end up where you need to be lol.  The other issue is that I would need to buy an RME PCIe card.

What I am going to test with it is putting just MIR PRO on it, and let it handle processing MIR PRO.  This combination would work great as long as I can get some really low latency on the gigabit connection.

Maestro2be

Take a look at the specs in my signature. I just had these machines built. They were relatively cost effective. I can run *almost* anything I want + MIR Pro at 256 samples/48k/Cubase 7.5. I'm sure I could get that number even better if I replaced my MOTU with an RME MadiFX card or something. All the slave does is run VSL + MIR Pro. I don't add any many mix plugins on the master computer until I'm nearly done programming. Using 2 480 GB Samsung EVO SSDs for the samples on the slave.

Thanks for the information.  Yea those specs are almost exactly what my Slave would be, except I have a 4930k instead of the 4960k.  This just further pushes me to finish the build and then test having it just process MIR PRO for me.  I am sure it will solve my reaching 100% CPU issue, I will just have to workout a workflow to remove the latency while I am doing certain things.

I don't think this setup would be to much to work with each day until I decide one day to build another super machine (I am waiting on the new Xeon V2 15 core processors).

Maestro2be

I am still curious what specs would be needed to run it all in a single box.  For example, if you are using a 6-core master and 6-core slave, would a 12-core of the same processor family do the trick?

(I am envisioning a workflow of running a DAW, in my case Digital Performer, plus VSL with a full orchestra, and minimal effects while originally laying down MIDI.  Then fine-tuning and mixing using a MIR stage.  I want to minimize theneed to bounce to audio and freeze tracks while fine-tuning the arrangement and the mix.)   My old set up was a Master (now dead machine) + 2 slaves, and I'd really like to downsize the physical footprint if it can be done.

I don't think anytime soon that a single machine will be able to give low latencies. The tradeoff has always been more cores for lower clock speeds.