I finally secured a little production contract and can no longer go w/o a VST host.
I frequently read Anand S.'s site for hardware analysis and came to the conclusion that Conroe CPU/Chipset will be my choice, for the A/V benchmarks merit this. Intel seems to be back on top with it's performance again, and it's stability has kept me in the money for a year now.
Since I also play live I need to build a studio box seperate from my live rig, but be able to disconnect it for my nightime endeavors in live performance. I don't trust XTC mode with it's added latency, besides why would I want to turn Scope into VST when I could just add or remove a second box. So my current 2 x 15 DSP Scope DP box will be integrated into my live rig. Which leads me to the questions at hand.
#1) Can I assign VSTi's and VST FX to one core, and Nuendo to another core? Nuendo is my choice for it's media transferring abilities, and maturity as VST host.
# 2 ) Will this powerful CPU allow VST programmers to make more authentic sounding VSTi's? I have heard them all, and being an analog convert I can assure you that VSTi's are sonically inferior to DSP's, and that DSP's still don't have the true mono capabilities of my Oberheims, or Studio Electronic synths. But I think that marketing and ROI had to do with the tradeoffs made to programmers. For I have heard the Lounge Lizzard modeling electro-mechanical instruments, and though it is a hog on cycles, it sounds damn good for a VST. But it was an acception IMHO. You'd think that programmers will have more cyclic resources now,eh?
I need to build this by August, and have it spitting out audio by September. So I have to master Solaris 4.0 by then. You computer guys have really got me working my ass off now.
To Us, And Those Like Us,
Conroe w/Scope + Giga 3 Orchestra
To answer #1:
Nuendo (and SX) will load-balance across multiple cpu's and cores automatically. In the Device Setup > ASIO section (whatever its called now) under the ADVANCED button you'll find a checkbox 'enable SMP support' or something similar. You'll also see 'audio priority' (this is the control that sets the executable priority as seen in task manager). Going back to SMP support, if you disable it Nuendo should load all the MIDI and user interface tasks onto one cpu (or core) and all audio tasks on the other. Now I have to add a bit of a caveat here, because I never upgraded past 1.6 for Nuendo. Here's the caveat:
In Nuendo 1.5x, Steinberg officially added VSTi support. However at that time my SMP support went to s***! What basically happened wasthis. Previous to VSTi support being added, if Cubase's processing meter showed 100%, then I would look and see it consuming 100% of a single cpu and a minor amount of the other cpu. This was the audio engine running on a single cpu, and the MIDI/UI code running on the other. Then, with the SMP Support enabled, you'd see the load magically spread between both for the audio engine AND the Midi/UI (according to the techs on the nuendo forums at the time). However with 1.53 you could start with the same setup, no SMP support enabled and a pure audio project (no VSTi's) and watch the cpu do the same thing. Start adding VSTi's and they actually started running on the OTHER cpu. So VSTi's ran in their own thread. Now enabled SMP Support and you'll notice that instead of both cpu's now being loaded 60% or more, the first cpu stays roughly 50% and the second cpu shows the VSTi load still (proportional to the first cpu). Ie, if you had loaded up another 20% of VSTi on the 2nd cpu, cpu0 now shows 50% and cpu1 shows 70%. This also resulted in a lot of glitching because the VSTi thread was obviously not load-balancing. Nuendo v1.6 smoothed out the glitching but still ran a lot more reliably with SMP support disabled.
The short of the story is, I'm not sure but I would guess they've solved the SMP issues caused by improperly load balancing the VSTi thread in Nuendo/SX 3. HOWEVER do NOT believe that you're going to see 200% gain with 2 cores, or even 150% gain. Audio is not as granularized and scalable as other tasks are for multiple processors (tasks like SETI or 3d rendering). Even SONAR (which excels at SMP for audio on PC) is able to achieve loads of 130-140% from what I've seen.
The new core designs coming from intel promise improvements in other ways, since they're departing from the p4 philosophy of deep fast pipelines now. This should bring performance back to levels more comparable with AMD clockspeed for clockspeed, but beware that AMD's integrated ram controller still brings them quite an edge, especially when paired with more than one cpu core.
To answer #2:
If people were still content to use 16 tracks of audio with 4 insert slots, probably the modern cpu's would bring real gains in audio quality. However as cpu speeds have increased, so has trackcounts, the number of stacked effects, the number of send effects, User Interface bloat, application bloat in general and even OS bloat.
Then of course there's the fact that increase cpu speeds aren't necessarily used by developers who hack away in assemblers and integrate hand-coded routines into their custom C libs. More typically the higher cpu speeds allow them to use higher-level tools to bring products 'to market' faster, resulting in tools that may have a lot more features but not necessarily HIGHLY improved sound quality, and which probably wouldn't have even run on systems a few years ago.
I'm going to go out on a limb here and use implicate Native Instruments as an example of a typical modern audio-software house (I'm NOT intending to start a flame war with fans of NI). Case in point: Generator 1.5x ran great on my p2-300. I know that they have implemented oversampling on their oscillators & filters now, and they're doing bandwidth limiting techniques to reduce aliasing, but given that I can now buy a cpu that's 3Ghz or faster you'd think they really would show major improvements over the products from 8 years ago. There are certainly improvements in the number of modules you can use, the UI itself (and their new lower-level math bits) but overall the improvement in sound quality has been really slow in comparison to the improvements in computing power behind it. None of this so far should be alien to people familiar with these forums
Then there's also the fact that NI's corporate size has increased rather dramatically (especially the executives and marketing dept.), and yet most of what they've done has been to repackage the backend reaktor/generator code into newer applications with more restricted UI's (kontakt, impakt, etc etc etc). In fact I noticed that NI would often bring 5-6 new products to market in the time it took to release an update or two that fixed *critical* bugs. Oh, Absynth & Fm7 were not coded in-house btw. Poor FM7 desparately needs an update to its filters and effects too.
I think that this is representative of many of the larger companies out there, and its small shops like Audio Damage are able to bring products to market that really DO sound better than their plugin predecessors, they do it quickly and make products that are worth supporting imo. The hard part is determining the good shops out there like the AD guys from just another bloke armed with Synthedit or a similar tool.
Hm I've written way too much here. Hope it was at least somewhat useful.
Nuendo (and SX) will load-balance across multiple cpu's and cores automatically. In the Device Setup > ASIO section (whatever its called now) under the ADVANCED button you'll find a checkbox 'enable SMP support' or something similar. You'll also see 'audio priority' (this is the control that sets the executable priority as seen in task manager). Going back to SMP support, if you disable it Nuendo should load all the MIDI and user interface tasks onto one cpu (or core) and all audio tasks on the other. Now I have to add a bit of a caveat here, because I never upgraded past 1.6 for Nuendo. Here's the caveat:
In Nuendo 1.5x, Steinberg officially added VSTi support. However at that time my SMP support went to s***! What basically happened wasthis. Previous to VSTi support being added, if Cubase's processing meter showed 100%, then I would look and see it consuming 100% of a single cpu and a minor amount of the other cpu. This was the audio engine running on a single cpu, and the MIDI/UI code running on the other. Then, with the SMP Support enabled, you'd see the load magically spread between both for the audio engine AND the Midi/UI (according to the techs on the nuendo forums at the time). However with 1.53 you could start with the same setup, no SMP support enabled and a pure audio project (no VSTi's) and watch the cpu do the same thing. Start adding VSTi's and they actually started running on the OTHER cpu. So VSTi's ran in their own thread. Now enabled SMP Support and you'll notice that instead of both cpu's now being loaded 60% or more, the first cpu stays roughly 50% and the second cpu shows the VSTi load still (proportional to the first cpu). Ie, if you had loaded up another 20% of VSTi on the 2nd cpu, cpu0 now shows 50% and cpu1 shows 70%. This also resulted in a lot of glitching because the VSTi thread was obviously not load-balancing. Nuendo v1.6 smoothed out the glitching but still ran a lot more reliably with SMP support disabled.
The short of the story is, I'm not sure but I would guess they've solved the SMP issues caused by improperly load balancing the VSTi thread in Nuendo/SX 3. HOWEVER do NOT believe that you're going to see 200% gain with 2 cores, or even 150% gain. Audio is not as granularized and scalable as other tasks are for multiple processors (tasks like SETI or 3d rendering). Even SONAR (which excels at SMP for audio on PC) is able to achieve loads of 130-140% from what I've seen.
The new core designs coming from intel promise improvements in other ways, since they're departing from the p4 philosophy of deep fast pipelines now. This should bring performance back to levels more comparable with AMD clockspeed for clockspeed, but beware that AMD's integrated ram controller still brings them quite an edge, especially when paired with more than one cpu core.
To answer #2:
If people were still content to use 16 tracks of audio with 4 insert slots, probably the modern cpu's would bring real gains in audio quality. However as cpu speeds have increased, so has trackcounts, the number of stacked effects, the number of send effects, User Interface bloat, application bloat in general and even OS bloat.
Then of course there's the fact that increase cpu speeds aren't necessarily used by developers who hack away in assemblers and integrate hand-coded routines into their custom C libs. More typically the higher cpu speeds allow them to use higher-level tools to bring products 'to market' faster, resulting in tools that may have a lot more features but not necessarily HIGHLY improved sound quality, and which probably wouldn't have even run on systems a few years ago.
I'm going to go out on a limb here and use implicate Native Instruments as an example of a typical modern audio-software house (I'm NOT intending to start a flame war with fans of NI). Case in point: Generator 1.5x ran great on my p2-300. I know that they have implemented oversampling on their oscillators & filters now, and they're doing bandwidth limiting techniques to reduce aliasing, but given that I can now buy a cpu that's 3Ghz or faster you'd think they really would show major improvements over the products from 8 years ago. There are certainly improvements in the number of modules you can use, the UI itself (and their new lower-level math bits) but overall the improvement in sound quality has been really slow in comparison to the improvements in computing power behind it. None of this so far should be alien to people familiar with these forums
Then there's also the fact that NI's corporate size has increased rather dramatically (especially the executives and marketing dept.), and yet most of what they've done has been to repackage the backend reaktor/generator code into newer applications with more restricted UI's (kontakt, impakt, etc etc etc). In fact I noticed that NI would often bring 5-6 new products to market in the time it took to release an update or two that fixed *critical* bugs. Oh, Absynth & Fm7 were not coded in-house btw. Poor FM7 desparately needs an update to its filters and effects too.
I think that this is representative of many of the larger companies out there, and its small shops like Audio Damage are able to bring products to market that really DO sound better than their plugin predecessors, they do it quickly and make products that are worth supporting imo. The hard part is determining the good shops out there like the AD guys from just another bloke armed with Synthedit or a similar tool.
Hm I've written way too much here. Hope it was at least somewhat useful.
Hi,
@S4L: currently, the Conroe with P5WD2E seems to be the next big deal ... for Q4!
The i975X can give access to 4 full Gigabytes (without sharing memory space with IO/devices memory) and better bandwidth.
@VALIS: Good points!
I would add also that VSTi granularity might help load balancing. I mean that a huge VSTi (like multi channel/ high polyphony samplers ... Halion,etc) might not help for spreading the load across CPU. I am not sure that these plugins are internally designed to support multithreading (as one thread per channel) which might lessen the performance gain of the overall system. But this still have to be verified!
cheers
@S4L: currently, the Conroe with P5WD2E seems to be the next big deal ... for Q4!
The i975X can give access to 4 full Gigabytes (without sharing memory space with IO/devices memory) and better bandwidth.
@VALIS: Good points!
I would add also that VSTi granularity might help load balancing. I mean that a huge VSTi (like multi channel/ high polyphony samplers ... Halion,etc) might not help for spreading the load across CPU. I am not sure that these plugins are internally designed to support multithreading (as one thread per channel) which might lessen the performance gain of the overall system. But this still have to be verified!
cheers
Thank You,
Conroe sounds like it will work for me. Load balancing the threads is exactly what I want to do. I'm just doing jingles at first, nothing fancy. But Nuendo will come in handy with its transferring abilities, and summing quality. I'll be using some very expensive mic pres and a 480L, however VST FX don't seem to suffer the same fate as VSTi's do. I have heard some really good high end VST FX like Waves,etc. I have used Lexicon, and Ursa Major for so long that I can get those to work for me. All others I will buy and learn as I go along, for I have too much advice from all of my VST buddies here in local studios, I personally think they will never know what sounds good to me, so I will thank them and move along. For my own recordings of my music, I will suffer with VDAT for I have heard the 32bit integer stuff, and it sounds as good as Pro Tools in the summing aspects, it is just a pain to learn. But for speed and production turn around, I will use the VST host approach. Afterall, it is commercial production, they only want the low bidder, and fastest turnaround. The marketing people will all agree that it is a classic, and pat themselves on the back as I steal their cash. For they are not worth the effort of syncing vocals w/ VDAT and using the MC-500MkII, I will make so much easy money from the first 6 months to pay for the box, and even buy me more DSP's. Since prostitution is legal here in Nevada, I may as well be a musical whore. As long as I jam at night, my soul goes for cheap in the daytime.
Thanks again for confirming my suspicions.
Strength And Honor,
Conroe sounds like it will work for me. Load balancing the threads is exactly what I want to do. I'm just doing jingles at first, nothing fancy. But Nuendo will come in handy with its transferring abilities, and summing quality. I'll be using some very expensive mic pres and a 480L, however VST FX don't seem to suffer the same fate as VSTi's do. I have heard some really good high end VST FX like Waves,etc. I have used Lexicon, and Ursa Major for so long that I can get those to work for me. All others I will buy and learn as I go along, for I have too much advice from all of my VST buddies here in local studios, I personally think they will never know what sounds good to me, so I will thank them and move along. For my own recordings of my music, I will suffer with VDAT for I have heard the 32bit integer stuff, and it sounds as good as Pro Tools in the summing aspects, it is just a pain to learn. But for speed and production turn around, I will use the VST host approach. Afterall, it is commercial production, they only want the low bidder, and fastest turnaround. The marketing people will all agree that it is a classic, and pat themselves on the back as I steal their cash. For they are not worth the effort of syncing vocals w/ VDAT and using the MC-500MkII, I will make so much easy money from the first 6 months to pay for the box, and even buy me more DSP's. Since prostitution is legal here in Nevada, I may as well be a musical whore. As long as I jam at night, my soul goes for cheap in the daytime.
Thanks again for confirming my suspicions.
Strength And Honor,
Yes,
I wanted the 975XBX, but only 2 x PCI. Since I believe that CWA is going to go with external solutions, and occasionally throw us a bone,i.e. Dynatube,etc. Thw Asus P5WD2E has 3 PCI, and I'm sure that Asus has included it's great OC'ing BIOS. Any news On early release, or ES samples I will pass along.
_________________
Jimmy V.
<font size=-1>[ This Message was edited by: scope4live on 2006-04-18 05:35 ]</font>
I wanted the 975XBX, but only 2 x PCI. Since I believe that CWA is going to go with external solutions, and occasionally throw us a bone,i.e. Dynatube,etc. Thw Asus P5WD2E has 3 PCI, and I'm sure that Asus has included it's great OC'ing BIOS. Any news On early release, or ES samples I will pass along.
_________________
Jimmy V.
<font size=-1>[ This Message was edited by: scope4live on 2006-04-18 05:35 ]</font>
-
tripstation
- Posts: 176
- Joined: Tue Oct 12, 2004 4:00 pm
Hi Guys,
I have tried dual cpu's with Nuendo 3 and two scope 15dsp cards and it's a bit unstable.
The system seems to crash in the end.
Thought I'd warn you.
I now run my scope cards on a seperate single cpu machine.
regards,
Jimi
I have tried dual cpu's with Nuendo 3 and two scope 15dsp cards and it's a bit unstable.
The system seems to crash in the end.
Thought I'd warn you.
I now run my scope cards on a seperate single cpu machine.
regards,
Jimi
Asus PC-DL delux
Winxp Pro sp1
Nuendo 3 / mLAN
Gigastudio3
Native Instuments Komplete 2
2*xeon 3.06ghz
2gig ram
wd raptor 74gig system
wd raptor 74gig audio
Scope pro 30dsp (stand alone pc Asus P4GPL-X, Win XP sp1)
<BR
Winxp Pro sp1
Nuendo 3 / mLAN
Gigastudio3
Native Instuments Komplete 2
2*xeon 3.06ghz
2gig ram
wd raptor 74gig system
wd raptor 74gig audio
Scope pro 30dsp (stand alone pc Asus P4GPL-X, Win XP sp1)
<BR