What's new

Cubase/Nuendo 10 Performance Test

Delio Roman

New Member
I'm not shocked because most of us know the performance hit we take using Cubase or Nuendo on macOS but I AM shocked at the difference in performance loss/gain.

So, I conducted a small test today on the performance of Nuendo 10.0.20 on my 2013 Mac pro (trashcan) in comparison to another PC I own that i use for work. It's an X99 system.

Now, this is by no means a formal test, more so a "let's see what happens" kind of test lol. Also, I understand these machines aren't exactly similar in specs, but even if they were closer to spec, the difference I noticed wouldn't be made up I feel, on the macOS side. Here's the specs of each machine:

2013 Mac Pro:
macOS Mojave 10.14.5
6-core 3.5GHz Xeon E5 1650v2
32GB of 1866MHz DDR3 ECC RAM
512GB NVMe Samsung 970 PRO boot drive with adapter (to fit in Apple's proprietary slot)
Vega 64 8GB GPU (via a TB3 Razer Core X eGPU enclosure)
Samples loaded from a Sabrent Rocket 1TB NVMe via a Sonnet Echo TB3 enclosure

X99 build:
Windows 10 Enterprise build 1903
8-core 3.0 GHz 5960X (stock clock, NOT OC'd)
32GB of 3200MHz DDR4 Corsair Dominator Platinum RAM BUT dumbed down to 2133MHz for a slightly better comparison)
512GB NVMe Samsung 970 Pro boot drive in m.2 slot on mobo
NVIDIA Titan Xp 12GB GPU
Samples loaded from a Team Group T-Force Delta 1TB SATA SSD

Interface:
UA Apollo X8 TB3 interface

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

[these were results in real time playing]
[each instance of Kontakt had only ONE library/patch loaded (all different libraries ranging from 8Dio, Spitfire, Orchestral Tools, etc) and loaded as track instruments]
[each instance played a separate midi part]

Here were my findings:

Mac Pro:
82 tracks/instances of Kontakt @ 2048 samples (max buffer)

That was ALL I was able to squeeze out of the MP. AT 2048 samples! Anything more, and it would just click, and no samples played. Nuendo felt extremely sluggish while playing too.

X99 Build:
258 tracks/instances of Kontakt! (same libraries/patches as the mac and yes after the 82 track mark, i added more patches) @ 128 samples!
If i went down to 32 samples, instances went down only slightly to 224.

* I COULD have loaded more, but my RAM was maxed out @ 100% on the X99 build

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

So, this is crazy. that is a HUGE difference in loss of performance on the MP side.

Even IF, I installed macOS on the X99 build, I highly doubt it would be the same as the Windows results.

Even then, the 1650v2 and the 5960X aren't that crazy off in performance anyways. the 5960X is better, but not by an insane amount.

There is definitely some under the hood work Steinberg needs to do on the macOS side, because there are numerous forums reporting the same thing. Even Logic Pro X performs insanely well in comparison on the same machine.

Also note, there is overclocking headroom on the X99 build to warrant a few more instances possibly (CPU & RAM) since the CPU is on water.
------------

If anyone has any input or similar findings, post them here! Or let's talk about this. :)
 

kenose

New Member
That is a rather huge difference! I have the same Mac Pro, and out of curiosity did some similar tests in Cubase vs Logic. I always felt that Logic outperformed Cubase, but the tests had some very weird and surprising results.

I tested with the Spitfire Chamber Strings Cello Performance Legato patch, with a ~16bar midi melody with just modulation data. I duplicated that with the same midi/patch for every instance.

2013 Mac Pro:
macOS Mojave 10.14.5
6-core 3.5GHz Xeon E5 1650v2
32GB of 1866MHz DDR3 ECC RAM
512GB Apple SSD
D500s
Samples coming off a Crucial MX500 in an attached USB3 enclosure.
Focusrite Saffire Pro 40 Firewire interface

Results:
Cubase 10.0.2, Kontakt 5.8.1, ASIO Guard on High


2048 - 78 instances
1024 - 83 instances
512 - 82 instances
256 - 80 instances
128 - 77 instances
64 - 76 instances
32 - 66 instances


Logic 10.4.5, Kontakt 5.8.1, Process Buffer Range: Large

1024 - 61 instances
512 - 73 instances
256 - 77 instances
128 - 79 instances
64 - 79 instances
32 - 79 instances

What the hell is with Logic's performance in this at higher buffers? And why does it seem to stabilize at 128...I'm very curious if this was happening in Sierra before I updated to Mojave and 10.4.5 as well.... I definitely don't remember Logic being so limited at 1024 on Sierra.

The Cubase results seem in line with your experience on the 2013 MP. The difference is definitely odd between the 5960x considering they were released only a quarter apart.
 

IFM

Senior Member
Logic seems to like Medium buffer more for some reason. That is a huge difference! Clearly if I was going to use Cubase exclusively I'd run it on a PC.

What you should do for a better test is install Windows 10 on a boot camp partition on the Mac Pro, then see if it is a hardware or software issue.
 

kenose

New Member
Yes, I think I’m going to do that when I have a bit of time to install everything on my Bootcamp partition. I’m extremely curious now how Cubase will react.
 

IFM

Senior Member
Yes, I think I’m going to do that when I have a bit of time to install everything on my Bootcamp partition. I’m extremely curious now how Cubase will react.
Were you doing playback on a separate audio track so that LPX could properly distribute the load?
 

kenose

New Member
Were you doing playback on a separate audio track so that LPX could properly distribute the load?
Yep, was always on an empty track for playback. I pushed the track count to the point where all 11 threads were at 100% with clean playback at least twice through. (no overloads) Logic does seem to warm up sometimes, if it failed I gave one more attempt and required it to play through twice before calling it stable. On a few settings it would overload the first try, and then the next attempts played fine.

Cubase is a little different since it doesn’t stop playback in the event of an overload like Logic does, so I listened closely for any degredation of playback. Same system of testing— I gave it one chance to fail, but if it couldn’t playback clean twice in a row after that I moved on.

Will hopefully test Cubase on Win 10 Bootcamp later today.

EDIT: Ugh, looking unlikely as I only have 20gb free in my tiny bootcamp partition and there isn’t a content free version of the C10 installer.
 
Last edited:

kenose

New Member
I spent way too much time on this today, and have other work I must get to, but... here are the results with C10 on Win 10 on Bootcamp.

Code:
2013 Mac Pro:
macOS Mojave 10.14.5
6-core 3.5GHz Xeon E5 1650v2
32GB of 1866MHz DDR3 ECC RAM
512GB Apple SSD
D500s
Samples coming off a Crucial MX500 in an attached USB3 enclosure.
Focusrite Saffire Pro 40 Firewire interface
Same computer, same external SSD running the samples, same Firewire interface, same Cubase project file. I believe the drivers for the interface work a bit different in Windows, with the Focusrite MixControl installed I can choose "ASIO Saffire" as my device in Cubase. The ASIO buffer is then changed in the Saffire MixControl application rather than in Cubase. Same ASIO Guard settings though.

I had to do a reimage of W10 to free up enough disk space, and I also had to screw with the Cubase installer ZIP file. I removed the additional content from the ZIP file with 7-ZIP, so I would have enough space to extract the installer. This apparently works fine, so I was able to install just barebones C10. So this is an absolutely fresh system.

I am running a trial of MacDrive so my samples SSD would show up in Windows 10, this doesn't seem to affect disk performance at all. I benchmarked in CrystalDisk and it hit the same speeds as I get on the Mac side.

RESULTS:
Cubase 10.0.2, Windows 10(Bootcamp), Kontakt 5.8.1
Each instance with (1) Spitfire Chamber Strings Cello Performance Legato patch

2048 - 92 instances
1024 - 100 instances
512 - 99 instances
256 - 95 instances
128 - 94 instances
32 - 89 instances

So this seems to indicate that Mac Cubase is still lagging behind Windows Cubase in performance. However on this machine it's definitely not as an extreme difference as @Delio Roman has seen-- that must just be due to the 5960X dominating the E5 1650v2 somehow. It does show a ~45% multicore improvement here: https://cpu.userbenchmark.com/Compare/Intel-Core-i7-5960X-vs-Intel-Xeon-E5-1650-v2/2580vsm7574

One thing to note is that this last test was performed on a pristine Win 10 install, it is very possible that my Mac side is bloated in some way and needs to be wiped/reimaged. I am still extremely confused how Logic 10.4.5 is showing such awful performance, it should not be getting beaten that badly by C10 on Win/Bootcamp. Something is definitely wrong.

I wish I had done these tests in Sierra so I could compare to Mojave... I have a suspicion there is some weirdness going on with my interface driver in Mojave, but Focusrite claims the Pro 40 is completely compatible.
 
Last edited:
Top Bottom