Things I hate about... - Cubase 7

[Compyfox]

I am trying to determine if C7 is really such a problem for you. When cubase 7 is using 20% extra, is asio guard ON and individually ticked for those fx?

FX is automatically ticked for FX, and automatically un-ticked for VSTi. Cubase seems to be that intelligent at least.

Then again - no real difference if guard is on or off.




Now on to my... special... guest... who still thinks(?) and treats me like a total PC and host newbie.

This is because Reaper uses a system very similar to Asio Guard. They call it anticipative fx processing. If you use C7 without Asio Guard you are effectively using just one core. I think the big bug you are seeing is the new way C7 is structured internally. It is actually much much more efficient. You just need to take advantage of this new feature. C7 uses one thread for live processes, you can see it's cpu usage from the lower bar on the cpu meter. The upper bar marks the total cpu usage. Mostly you should see the lower bar hovering near zero, until you grab a channel with a heavy plugin instrument which then gets loaded to the live thread.

I know what both hosts are supposed to do. Remember - Cubase user since 1995(!), I do have certain experience with PT, Logic, Reaper, FLStudio, yadda yadda. So I'm not totally stupid on that end.

I do have an issue however, if I load the very same project in different Cubase versions (prior to my multi-OS nuking, and post) and then have: 31% ASIO load in Cubase 6.x (which already used Multi Core handling), but a "Guard Load" of at least 40-42%, and way over 100% load if the guard is off in C7.

Same project, same sample buffer size (128 samples), different host revisions.
Now tell me that this is normal.

Oh wait... it should be - since this is a new engine and all. You're actually working for Steinberg by chance? Because this is the very same answer I got in the early weeks of my tech support inquiry.

It is a shame these features aren't properly documented. Learn to use them though and C7 actually performs a lot better than C6.. and on par with Reaper. Actually could be even better. Haven't benchmarked the difference.

So, activate Asio Guard on your plugins and on your audio settings. Try your test again.

You really do like to hear yourself preaching what others do wrong, don't you?

What the hell do you think I've been doing for the last couple of months?! Drowning some pills and halucinate?

Edit: Ah remembered this a bit wrong; Cubase doesn't show you the total cpu usage. It's always kindof just per one core. So if you have 8 cores and you load an instrument that eats up 90% of your processing bar you can still load 7 more.. and the bar keeps at it's place. Just like on C6. Asio Guard OFF you get to do this too. But with Asio Guard ON this is much much more efficient. Sometimes the difference being literally like comparing 1core against many.

Sorry - can't confirm this.

If you run this test with Asio Guard off you are seeing the performance of just one core. CPU generations have mostly upgraded their core amount, while single cores have stayed pretty much the same on their processing power. In C7 you need to use Asio Guard to get the performance of all your cores.

Another test result (which I unfortunately CAN NOT make completely public yet) - with the very same (in)official Steinberg testproject I've got.

128 samples buffer:
Intel i7 920 Bloomfield (3,6GHz), RME HDSPe:
CPU max load: 36% (C7), 28% (C6.5)
C7.06 - Guard on: AVG 11% idle / 40% load // RT: 12,5% idle / 15% load
C7.06 - Guard off: AVG 12,5% idle / 40% load // RT: 22% idle / >100% load
C6.5: 13% idle / 31% load

Intel i7 2600K Sandy Bridge (4,5GHz) - RME HDSPe:
CPU max load: 39% (C7/C6.5)
C7.06 - Guard on: AVG 6% idle / 40% load // RT: 15% idle / 12% load
C7.06 - Guard off: AVG 6% idle / 40% load // RT: 15% idle / >100% load
C6.5: 8% idle / 27% load

Intel i5 4440 Haswell (3,1GHz stock, found and old TXT where I scribbeld that down - not in my results sheet), Digidesign MBox 2:
CPU max load: didn't check that
C7.06 - Guard on: AVG 9% idle / 37% load // RT: 10% idle / 10-12% load
C7.06 - Guard off: AVG 10-12% idle / 37% load // RT: 20% idle / 90-100% load

The higher the sampling buffer, the better the C7 ASIO performance (ideally/stable at about 1024). C6's performance is better overall, for all buffer sizes up until 1024, then it starts to wear thin a bit.


These are three different rigs, completely different configured. Do you still think that the ASIO engine is working "correctly"? Do you still think I'm doing something wrong?

Are my fellows and I lying?


I still do these nonsense tests (like further BIOS edits with EIST, and the CPU Parking tests) just to rule even that(!) out of the equation. I did my work already. Why is it so hard to understand that it is not(!) the hardware that is causing issues. At least on my end.

You are killing it. Intel specs the absolute maximum you should have for i7 920 as 67.9°C.

I aksed this question to several rig builders, OC'ers, fellows. They all were like "you might have found a good CPU batch where you can undervolt - but else your rig is absolutely fine".

Before I OC'd I actually had 2-3°C higher values per core! So I consider this a really good performance.

These... "intel specs" are only there to evade warranty issues. The web is full of comments with "hot" CPU's prior to OCing and post OCing. You might even find my old post in the "Computer Setup" section where I asked about this.

Some might be lucky to have an Nehalem, Westmere, Sandy, Ivy, Haswell that is actually cooler running than expected. This really depends on the build. So you were lucky, and I wasn't so lucky. But I am still(!) within real world specs. And I don't reach 100% load constantly - if ever. Currently I'm happy to even REACH 50% CPU load before the ASIO system tells me "sorry - you used up all your spare".

Actually the i5 Haswell I recently build, runs with merely 84W compared to my 130W CPU (hexa cores do run on 130-150W still, modern Sandy/Ivy/Haswell do run between 60-85W nowadays) - and even with an oversized, aircooled heatsink and low noise adapter I still reached above 50°C with a PRIME stress test.

Now tell me that this is wrong yet again. Actually - no, I don't want to hear it.

I'm running a hexacore and its 47°C after 6 hours of max load on all cores (max load on the pcie GPU too.. so even extra heat compared to DAW work). You should also take note of your Mobo temperature as that is where all your data is moving. They have cooling needs too. Mine is at 37°C after 6 hours of full load.

My mobo does't even touch 40-45° on full load, I have 2x 12cm case fans, an ANTEC case, excellent air management (air sucked in from front and top, blown out in the back), Noctua heatsink with 2x12cm fans. Heck my HDDs run way hotter than my whole rig alone. And they all have at least 5cm space in between (non SSD, mind you).

So you definitely don't have an adequate cooling.. or you are OC-cooking it too much..

YMMV. And I worked at an IT division where I had to handle industry type PC's every day. 60-70 degrees room temperature at extreme high humidity was normal torture for these devices. My i7 is bathing in luxury compared to that (room temperature reached 28°C this summer, floor temperature is usually 1-2 degrees lower - average room temperature: 22degrees).



Now could you please stop barking up my tree?

You critisized me earlier that I constantly "shred" every Cubase thread. You on the other hand permanently chime in to tell me that I'm wrong and stupid. This is slowly starting to annoy the sh*t out of me.


READ more closely(!!!) from now on. I won't repeat myself anymore.





SIDENOTE:
It's funny that "Kate's" rig problems turned into "my" (Compyfox) rig problems now. As long as other users can benefit from that - fine with me.

[mkdr]

Now could you please stop barking up my tree?

Just helping out noobs around here.
How else could they learn?

But sure I won't help you anymore..
You aren't that keen on getting help from the pro's anyway (and i dont mean just myself..)

[mkdr]

Oh, just one more thing..

I am trying to determine if C7 is really such a problem for you. When cubase 7 is using 20% extra, is asio guard ON and individually ticked for those fx?

FX is automatically ticked for FX, and automatically un-ticked for VSTi. Cubase seems to be that intelligent at least.

Then again - no real difference if guard is on or off.

Cool that Cubase has Asio Guard on for fx on default, didn't remember that. But it doesn't do anything on channels which have live inputs; that is vsti instruments that dont have Asio Guard on or live audio from your i/o. You can't process an effect beforehand if you can't do the same to the source.

[Compyfox]

Just helping out noobs around here.
How else could they learn?

So you do consider me as newbie, even if I'm clearly not!

But sure I won't help you anymore..
You aren't that keen on getting help from the pro's anyway (and i dont mean just myself..)

Right...

Cool that Cubase has Asio Guard on for fx on default, didn't remember that. But it doesn't do anything on channels which have live inputs; that is vsti instruments that dont have Asio Guard on or live audio from your i/o. You can't process an effect beforehand if you can't do the same to the source.

Again, I do know that (no newbie, remember?). And that is also not the issue.

I am talking pure ITB editing here. No recording.




Be back in a bit... further testing for you people.

[Compyfox]

Sorry, double post - but I want to separate this.


I went ahead, and retested (by request) with the following things altered:
- CPU Parking turned off (thanks CableChannel)
- EIST/Speed Stepping and C1E (CPU watt usage management) turned off (on request by TheoM and Mr "I only want to help noobs")


Additional findings (to the ones I already had):
- the CPU Parking Tool revealed that 4 cores (the "virtual"/HT cores) were in park mode, even though the energy profile was at 100% (set to performance) - the Windows internal task manager still showed usage of all cores (real and virtual ones) however prior to the switch
- Cubsase 6 and 7 already adressed all 4+4 cores! (multi core mode active)
- the test results only differ slightly, by that I mean 1-2% maximum (!!!), so even these "modifications" didn't offer a significant boost in usability
- with C1E and EIST/SpeedStep off, my rig actually slowed down!



I still can't (and won't) provide my full test specs. I'm also still waiting for further results from fellow testers(!). But I'll post the first three pages of information regardless. Just to show (and further prove/emphasize/reinforce) what's happening.

These tests were all post OCing, all with optimized OS (actually, tests with driver version 3.38 have an even more optimized rig!), the "CPU parking" and "EIST/C1E" tests are marked separately.


TEST RESULTS (Status 06-12-2013):
Test Results - Cubase 6.5/Cubase 7, Studio Compyfox and "guest rig"



I can now, once and for all, rule out(!) that this is an issue with my rig. Period!

I did everything I could. Involved more parties than I initially planned, they all devoted their sparetime to help me troubleshoot and get this issue resolved. Especially since Steinberg didn't want to care (I am still awaiting counter results!).

All in all - I alone(!) invested 5 weeks of worktime in a row, spread over the course of the last 4-5 months! This is the longest time I've ever invested for in-depth testing anything (beta testing included!). And even after that long period of time, I still don't see a significant difference.


I spend enough time on this.
I still await counter argumentation and counter results to prove my findings wrong.


I hope this finally shuts up those people that:
a) still do not believe whatever I say (and declare me as crazy and "being on a vendetta")
b) still think I'm an idiot in front of a PC
c) doesn't know what the heck is going on

You know who you are.



To emphasize it once more:
Steinberg is aware and informed of all this - and they will be updated on the status of further findings (as soon as I have more results!). There is no public mention about the issues yet, they are working on it, but do not expect an imediate fix. Or even a fix within the C7.x cycle.



[/end of line]

[mkdr]

Ah, now i see what the problem is!


I did the unthinkable and installed Cubase 6.5.4 back to my system. Just to make tests for you Compy. You are seeing a performance drop of 20%-ish.. and i can confirm it! It does really happen!

But..

The clue you gave me was the fact that you were doing just audio editing, and that Asio Guard is on. The difference between C6 and C7 is that C6 multiprocessing uses up all cores, where in C7(+asio guard) uses all cores minus one. That remaining one is used for the live thread and to run the channels you are editing/playing etc. So if on an 8-core machine you have full cpu utilization on all cores in C6, if you load the same project in C7 you are maxing it because one core sits idle.. and of course the load needs to go on some other core.. and the meter shows only the worst core performance of all the cores.

So in C7 when you use samplers or synths etc you are getting the benefits of running them at higher buffer sizes(that's what Asio Guard is for) when you're not editing them. For example Kontakt can have tremendous difference in performance if run at 128 buffer or at 1024 buffer. This is where the real gain in performance comes for C7. If you run only effects that don't benefit from bigger buffer sizes you are getting a slightly lower performance compared to C6. Especially as audio tracks are always automatically run before hand(in Asio Guard mode) even if they are selected.

Just look up what your Task Manager says your core cpu usages are when running your tests on C6 and on C7. On C7 there's that one core that sits idle as there's no live streams.


So the only bug there is that if Asio Guard is off in C7 it's still reserving that one core for the live stream, which now does nothing, even though it could just run it like normally on C6 and use all the cores for processing.

Problem solved. One more noob saved.
And C7 still is much more efficient.. most of the time

[Compyfox]

For once, I wanted to stay in bed, don't think of Steinberg or even KVR - no, I was stupid enough to check KVR still - only to see my most favourite smarty-pants post!

Ah, now i see what the problem is!

No, you still "assume" what my problem is.

I did the unthinkable and installed Cubase 6.5.4 back to my system. Just to make tests for you Compy. You are seeing a performance drop of 20%-ish.. and i can confirm it! It does really happen!

You don't say!
Wow, that really catches me off guard.

The clue you gave me was the fact that you were doing just audio editing, and that Asio Guard is on. The difference between C6 and C7 is that C6 multiprocessing uses up all cores, where in C7(+asio guard) uses all cores minus one. That remaining one is used for the live thread and to run the channels you are editing/playing etc.

...

Just look up what your Task Manager says your core cpu usages are when running your tests on C6 and on C7. On C7 there's that one core that sits idle as there's no live streams.

Sorry to burst you bubble of glorius deduction right in the middle, where you're still in full pursuit.


You see... I am not totally stupid, something you still assume I am. Else you would have stopped calling me "noob" by know, which is highly insulting btw.

There is this one nifty program called "Task Manager" (which I constantly used while testing) and I also used this program called "Park Control" yesterday, so really see what my cores are doing.

The one thing where your deduction fails, is that it adresses ALL CORES (real and virtual ones) in both Cubase 6.0/6.5 and Cubase 7. I have no core idling, I neither is one "not adressed and reserved for live streams.


Your theory is highly flawed.


Furthermore, I'm not doing pure audio engineering alone. That Steinberg benchmark test however is not(!) focused on testing the MIDI behavior - blame Steinberg for that (it's their project I've got after all). While songwriting, I have even more crazy behavior. Which brings me to the following:

Since you're already at it, why not also solve (from your deduction) the "lazy initialisation" bug with VSTi what SODDI (and I think also SJ_Digriz) was mentioning several times in here by now.

They aren't so called "noobs" either.

So the only bug there is that if Asio Guard is off in C7 it's still reserving that one core for the live stream, which now does nothing, even though it could just run it like normally on C6 and use all the cores for processing.

Again, all cores are adressed, and Cubase 7 runs (with Guard on) about 20-25% worse in terms of performance (especially in low latency) since day 1 - this is not how it should be! If the guard is off, I have even up to 50% worse performance - depending on the used buffer size. VSTi are already automatically turned off (most of the time, Cubsae does that itself).

So there is something more "wrong" than your "one core isn't adressed" assumption. Nice try, better luck next time.

Problem solved. One more noob saved.

I feel highly offended by that now. It's not a nuisance anymore.

You called me out for quite a while now at all other Steinberg threads I posted about this. I just contacted the mods of this board - I'm sick of constantly being accused of lying and being called "idiot" or "newbie".

And C7 still is much more efficient.. most of the time

No it is not (far from it actually) - you didn't even read the PDF it seems.

You just constantly bathe in your own sublimity rather than read properly, and only believe what you think is "right" in your opinion.

[mkdr]

The one thing where your deduction fails, is that it adresses ALL CORES (real and virtual ones) in both Cubase 6.0/6.5 and Cubase 7. I have no core idling, I neither is one "not adressed and reserved for live streams.


Your theory is highly flawed.

Well.. that is the way Asio Guard works. Think about it for a while. It can't do it any other way. Try it if you don't believe me.


(just to be more precise.. core idling does not mean core parked or no cpu usage at all. it just means it's not doing much at all. so not zero, but clearly not doing full processing either. you can clearly see this from task managers graphs)

[Compyfox]

I give up!
I seriously give up! It's no use debating with you.


The test results I posted clearly show(!!!) that C7 performs 20-50% worse compared to earlier Cubase versions. You constantly throw the same nonsense at my head that Steinberg did in the first two to three weeks where I first filed this issue.


Enough... Believe what you want.
You are right, we other users are wrong. So be it, end of story.

[ObsoleteAcc99]

[DELETED]

[mkdr]

I give up!
I seriously give up! It's no use debating with you.


The test results I posted clearly show(!!!) that C7 performs 20-50% worse compared to earlier Cubase versions. You constantly throw the same nonsense at my head that Steinberg did in the first two to three weeks where I first filed this issue.


Enough... Believe what you want.
You are right, we other users are wrong. So be it, end of story.

So even the maker of the software says it works like the way i say.

And you don't believe it.. or even bother to test if it's true






Btw. the "lazy initialization" bug is caused by the Asio Guard system. It can't play back a sound that starts before about half a beat of the beginning of the spot you hit play at. Asio Guards whole purpose is to work by processing things beforehand and i guess determining when to start this is tricky.

This could be fixed by Steinberg quite easily by delaying the start by half a beat.. or by filling a buffer whenever you move a cursor somewhere. For some reason they haven't done this yet.

I have no issue with this though. I can just press the numpad 1 and enter again.
And it doesn't happen on channels that are selected.

[mkdr]

Oh, another thing that came to mind just now. Bridging some plugins with Cubases own bridge can cause Asio overloads when instantiating them. Bevare of this.

Also, Automap is quite buggy with some plugins. Especially if they are 32bit and bridged in a 64bit Daw. This caused some instabilities for me in the past.

Automap also doesn't work with Cubases own plugins. Better turn it off for them and use Quick Controls instead (which you can map to the "user" midi channels in Automap).


These are the only instabilities or real problems I've ever had with C7.

[bluedad]

I don't know who is right or wrong.
But there seems to be a lack of communication between mkdr and compyfox.
Siince it's compy's thread to begin with, I'm asking mkdr to refrain from here on.

[mkdr]

I don't know who is right or wrong.
But there seems to be a lack of communication between mkdr and compyfox.
Siince it's compy's thread to begin with, I'm asking mkdr to refrain from here on.

Yep. No prob.
I did however solve his problem.
He will see it when he cools down a bit

[CableChannel]

@mkdr: with all respect, you did not solve the problem, you just found an explanation. That does not bring back the performance loss that he has (presumably) because of one core sitting idle.

Also with the lazy initialization you are wrong. That is really an initialization problem that only happens on the very first playback of a VSTi. Subsequent playbacks work alright.