What's new

ASIO Guard / CPU Cores vs Clock Speed

Blakus

Midi Magician
Clock speed is no longer king in Cubase?
I thought I might start a new thread on this topic, as for me it has proven to be a game changer.
(TL:DR - additional cores are better than just purely focusing on clock speed with ASIO Guard)

So, I just upgraded from an i7 3770k, 4 cores @ 3.5GHz, to a 6800k, 6 cores @ 3.4GHz base. I am using the exact same hard drives (except a new m.2 drive for my operating system), and went to 128gb of RAM.

Being the same clock speed, I would have expected that my ASIO/CPU performance in cubase would have been not much different at all, based on the "clock speed is king" theory that seems to be prominent. My experience with Cubase Pro 9 has been the exact opposite. Projects that used to bring my old 3770k to its knees, now open with only 60% cpu/asio usage - and I'm not exaggerating here. Unless there's some other voodoo going on, I am fairly sure that more cores are better than many think (at least in Cubase). I'm running hundreds of plugins, hungry synths etc, 300 kontakt instances, all with no problems now.

My "hunch" is that Cubase and ASIO guard have a lot to do with maximising core efficiency. ASIO guard effectively means that the entire template isn't required to run at intensely low latency, only the tracks that are currently being selected/recorded. So, most of the template is running with buffers that give 98ms latency, but I never feel this, because my actual live latency is 6ms (with 256 buffer). That 98ms provided by ASIO Guard is crucial, as the majority of the template no longer becomes so dependent on the clock speed required to deliver instantaneous, realtime performance. There is now time for cores to kick in with even distribution, much more effectively.

I think that for live performance, or other DAWs that still are required to run all tracks at live latency without features like "ASIO Guard", clock speed is king. But I am not convinced this is the case for Cubase with ASIO guard.

I just ran a test in Cubase with a large project that is running at around 60% ASIO meter - plenty of room for many more plugins/polyphony etc, it runs flawlessly. I just disabled ASIO Guard. Woah. The project will not even play at all and the core distribution is poor!

Anybody else singing ASIO Guard's praises? :D
 
Last edited:
Anybody else singing ASIO Guard's praises? :D
Reaper works this way too. :)

Single core performance vs number of cores is obviously going to be workload dependent. Having 20 slow cores won't do, since, as you say, tracks armed for recording are dependent on RT performance.

I suppose one way to look at this would be you want fast enough single core performance to handle all FX on tracks you will arm for recording at any given moment, plus some margin for headroom, and beyond that, more cores will start helping.

It sounds to me like previously RT performance wasn't your problem, due to ASIO guard, but rather you just didn't have enough total cycles across all cores to support all loaded FX.

For the stuff I do, 4 hyperthreaded cores has been plenty, and I'd rather have better single core performance to allow for lower latency on record-armed tracks, or to improve project load times (since patch loading in Kontakt is CPU bound and single threaded).

But I've seen some of your projects on Youtube, and I can't touch your project complexity. So I'm not surprised to learn about your results. :)
 
Is it possible that part of the performance difference is a result of generational differences between the 3770K and the 6800K? I don't know enough about advances in i7 chips to have a clear understanding of this.
 
Is it possible that part of the performance difference is a result of generational differences between the 3770K and the 6800K? I don't know enough about advances in i7 chips to have a clear understanding of this.
Yeah, this is what I was kinda referring to as other "voodoo" that might be going on hahah! I'd be very interested to know if this would make any kind of difference too.
 
The 3770k and 6800k seem to have relatively similar single core performance, at least https://browser.primatelabs.com/processor-benchmarks (according to Geekbench). The 6800k is marginally faster, but not significantly.

However the fact that the 6800k has a slower clockspeed does underline the fact that single core performance isn't just about clockspeed.
 
Cores can help if you're CPU limited, though I seriously doubt that there's much benefit beyond six. I think an overclocked six-core is the sweet spot for a master DAW these days, especially if you're running only one machine.

Clock speed helps if you're latency limited. In other words, if you're running at, say, 10 ms and want to get down to 3 then simply throwing a bunch of cores into the system likely won't have as much effect as clock speed.

Also, keep in mind that you're not comparing ONLY cores in what you've described - you have two completely different systems, so you're comparing cores along with a lot of other stuff on the motherboard/chipset that affects real-time performance, quite often to a larger extent than processing power. So your conclusion that cores matter more than clock speed is not justifiable based on your data.

Regarding ASIO guard, I've tried it. Doesn't play nicely with VE Pro in my setup so I've never used it. It's a good idea, it's just really hard to implement.

rgames
 
Clock speed helps if you're latency limited. In other words, if you're running at, say, 10 ms and want to get down to 3 then simply throwing a bunch of cores into the system likely won't have as much effect as clock speed.
Yes. That part makes sense to me. Also, in my experience, I've seen drastic difference with audio interface and drivers with regards to latency.

Also, keep in mind that you're not comparing ONLY cores in what you've described - you have two completely different systems, so you're comparing cores along with a lot of other stuff on the motherboard/chipset that affects real-time performance, quite often to a larger extent than processing power. So your conclusion that cores matter more than clock speed is not justifiable based on your data.
The effect that chipset/motherboard have on DAW performance interests me too, as this is hardly ever talked about. Affecting performance even more so than clock speed/cores is maybe a bit of a stretch though? I wonder how much real-world difference there actually is!

A shame that ASIO guard doesn't play well with VE Pro!
 
Last edited:
I'm going to try 'deactivating' VEPRO (ASIO Guard) in the plugin manager - hoping that I at least get benefits from non-VEPRO processes and ASIO Guard. Hopeful but realistic about it. Running 256ms latency now (with 3 slaves/vepro) - 1200+ tracks, plugs, etc - can never seem to get to 128 latency (that's even with MOST of my plugs turned off during writing stage). :(
 
So for someone who uses notation input (I don't own a midi controller) and latency really doesn't matter at all (I mean, 100ms is ok for input. When I play back, 1000ms is even ok) does that mean cores is really queen?
 
So the 1700x is beating the 7700k; I thought there were problems with Ryzen, it scaled with ram-speeds, and the Ryzen boards weren't supporting large amounts of RAM.

But if it's really beating Intel, maybe the 1800x is best.

Is there a big difference with the actual chipset? Should I get an x99 board instead of a z260 or b(I forget what they're called). Even with the same CPU? Are some chipsets simply faster/better?
 
Blakus, when I read your numbers I am about to start crying. I'm on OSX on a 6700k @4.5GHz and can't even manage to load something like 60 Kontakt instances in idle (without any other effects or VEP)... Is Windows 10+Cubase so much more efficient?! If that's the case I have to immediately jump ship!
 
All fine, but looking at his old setup is enough to get me wondering what is going on... He did not need the new setup to get this going...
 
This is the Logic Pro X CPU meter on my 12-core Mac Pro, while playing a sequence. The 24 cores are because it's hyperthreading.

I don't understand the arguments about cores being useless or of limited value, or that anything over six cores is redundant. This is OS X rather than Windows, but it's still running a DAW on an Intel processor (actually two 6-core ones).
 

Attachments

  • Screen Shot 2017-05-02 at 12.11.37 PM.png
    Screen Shot 2017-05-02 at 12.11.37 PM.png
    2.2 KB · Views: 17
All fine, but looking at his old setup is enough to get me wondering what is going on... He did not need the new setup to get this going...

I am no cpu nor windows/mac expert, but Cubase with asio guard on high is rock solid on windows 10 for me. Using a i7 5820k 6 core (at stock clock, 3,30ghz) myself and I can load so many different kontakt, play, engine2 instances, synths (also cpu heavy ones like serum, diva) without having to worry about any cpu maxing out. I am not using vepro anymore btw, this is with instrument tracks.
 
Last edited:
I don't understand the arguments about cores being useless or of limited value, or that anything over six cores is redundant. This is OS X rather than Windows, but it's still running a DAW on an Intel processor (actually two 6-core ones).
Look how much headroom you have per core. Now imagine a rig with 4 fewer cores (8 fewer logical processors) but 30% faster clock speed. Your overall headroom would look pretty similar but you would have a much more capable system for RT performance (or a demanding single-threaded plugin like Zebra).

Also your Kontakt libraries would load faster, because even on cheap flash storage, sample decompression in Kontakt is single-threaded and CPU bound. :)

It's not that more cores are useless, it's just that there is an obvious trade-off between cores and clock speed. Another way to look at it: given equivalent processing power aggregated across all cores, wouldn't you prefer the solution with fewer, faster cores? You get the same overall capacity for multithreaded apps and better performance for single-threaded apps.

Accepting that, then the question becomes how much, if any, multiprocessing headroom would you be willing to give up for faster single threaded performance? That's going to be workload dependent. For me, I'd give up 2 out of 6 cores for an extra 25%+ single core performance all day long. For you, maybe realtime performance doesn't matter and that isn't a sensible trade-off to make.
 
Last edited:
Tack, the most important thing for me is playback - how much can I run on the machine.

It has twelve 3.46GHz cores, plus hyperthreading, and the OS is doing a great job of dividing the load between the cores. The headroom you see is a good thing, because it means I'm not maxed out yet!

Playback is clearly not single-threaded, but if loading is, my entire template loads of SSDs in a little over a minute. Who cares.

What are the chips you buy up to now - 4.2GHz? That's something like 20% faster clock speed and six fewer cores, leaving aside all the other factors that undoubtedly make these old X5690 chips slower than the ones they're selling today.

My argument is with the "more than six cores makes no difference" statement. How can that be, and how is it a tradeoff?
 
Top Bottom