2019 Mac Pro DAW Benchmark Test Results

Lawson.

Senior Member
Another quick test. Same settings as above, except with 850 instances of Absynth playing 16th notes. (bonus points if you know what score that little ostinato is from)

While there may appear to be a decent amount of CPU headroom left over, Cubase started getting angry when I got over 900 tracks, causing lots of stop and go hitches, even when idling.

Just wanted to chime in and say the ostinato is Sayuri's Theme from Memoirs of a Geisha. :grin: I'm kind of surprised no one answered it yet!
 

simsung

New Member
Here's a Logic Pro X test (Ver. 10.4.8) for you guys. I haven't used Logic for the last 3 years, so it took me a bit to remember how to set up the project and whatnot. lol

Test:
80 instances of Omnisphere ("agape warmth" patch) playing a 5 note chord with the following 5 plugins on each instance:

FabFilter Pro-Q 3
Soundtoys Tremolator
Soundtoys Decapitator
FabFilter Pro-C 2 (4x Oversampling enabled)
2C Audio B2 (Duo Den Exp: Cinematic Long Hall preset)

Buffer: 256

For reference, my 8 core trashcan could barely run a project with just 8 instances of the B2 loaded in as sends.


I tried the same test on a mac mini 2018 i7 6 Core. The only difference is the reverb, since i dont have the B2 i used the Fabfilter R. Result: 24 Tracks max before dropouts
 

JazzDude

Member
55 OMNI instances with Steinberg plugins pingpong Delay, Frequency Equalizer and Roomworks each track. CPU around 70% ASIO guard high latency in 5,8ms out 7,8ms. running smooth without any glitches. Cubase performance meter around 80%. Disk IO 0%.
Hardware
Ryzen1700x 8+8 core 64GB RAM NVME 480GB NVIDIA 1050TI less than 1000 $/€ value
optimized WIN 10. In WIN 7 i could do abt 10% more instances.
ALESIS IO26 FW Interface 48k 256 samples
 

simsung

New Member
55 OMNI instances with Steinberg plugins pingpong Delay, Frequency Equalizer and Roomworks each track. CPU around 70% ASIO guard high latency in 5,8ms out 7,8ms. running smooth without any glitches. Cubase performance meter around 80%. Disk IO 0%.
Hardware
Ryzen1700x 8+8 core 64GB RAM NVME 480GB NVIDIA 1050TI less than 1000 $/€ value
optimized WIN 10. In WIN 7 i could do abt 10% more instances.
ALESIS IO26 FW Interface 48k 256 samples
that is hard to compare. we did the same test as above with omnesphere1 and even that changed result enormously
 

Damarus

Active Member
I like this - I have a 7700k I can try this test on. Fair to say Pro-R is a comparable alternative to breeze cpu useage wise?
 

JazzDude

Member
why would you use Omni1? Version 2.6 has a lot more to offer: its double than Version 2.
So its a harder benchmarktest. I did a lot of bench tests with Omnisphere and Cubase.
You can fill up Omni 2.6 with multitimbral (8 slots x 4 "soundsources/oscillators") and leave out the 3rd party plugins in Cubase to make it compatible for more people to share their experiences. We could share our files (only spectrasonics soundsources and efx).
 

simsung

New Member
why would you use Omni1? Version 2.6 has a lot more to offer: its double than Version 2.
So its a harder benchmarktest. I did a lot of bench tests with Omnisphere and Cubase.
You can fill up Omni 2.6 with multitimbral (8 slots x 4 "soundsources/oscillators") and leave out the 3rd party plugins in Cubase to make it compatible for more people to share their experiences. We could share our files (only spectrasonics soundsources and efx).
that was the only version on my colleagues imac. we just wanted to do a quick test. since the results caine compare i didn’t mention them
 

JazzDude

Member
ill prepare some Omnisphere bench files in Cubase 10, so everybody can use it and deactivate the tracks, for safe loading. It has preloaded 64 Omnispheres 2.6 with the "agape warmth" patch in folders of 10 and 5 midi notes running in loop.
 
OP
BenHicks

BenHicks

Professional Lurker
I like this - I have a 7700k I can try this test on. Fair to say Pro-R is a comparable alternative to breeze cpu useage wise?
I don’t own the Pro-R, so I can’t really say. I also don’t own 2CAudio’s Breeze 2, but according to their website, the B2 and Aether are their heaviest as far as CPU usage goes. I purposefully used the B2 and the Duo Den Long Cinematic Hall preset in that test because it was the heaviest I could find on my system. The Breeze 2 seems to be 2CAudio’s response to the criticisms of their heavier verb offerings.

I’ll see if I can’t run that test again using VSS3, Valhalla Room or Seventh Heaven, which aren’t quite so heavy. Interesting to see all of these results, though. I know this thread is focused on the Mac Pro, but without context and comparisons to other CPU’s running similar tests, it can be hard to see how things stack up. I do think that there are a bunch of different factors that need to be accounted for, so results are going to vary, wildly. Still, it’s interesting.
 
OP
BenHicks

BenHicks

Professional Lurker
Just wanted to chime in and say the ostinato is Sayuri's Theme from Memoirs of a Geisha. :grin: I'm kind of surprised no one answered it yet!
There’s an extra 16th note at the end of the phrase in my version, mainly because I wanted constant 16ths playing, but otherwise, yeah. Lol
 

Living Fossil

Senior Member
I do think that there are a bunch of different factors that need to be accounted for, so results are going to vary, wildly. Still, it’s interesting.
One interesting thing would be to see how an extremely heavy chain holds up.
E.g. Two instances of Ozone 9 in a row - with limiter set to algo IV and true peak - and some additional heavy weights.
That would have some real world relevance, since quite often there is one heavy chain that taxes a single CPU thread.

p.s. thanks a lot for posting your results...
 

Dewdman42

Senior Member
exactly. Also...what happens if you have 50 or 100 tracks mixing down and you try to record one more track, that is also feeding a fairly heavy signal chain.
 

Dewdman42

Senior Member
see this result: https://www.logicprohelp.com/forum/viewtopic.php?f=1&t=138612&start=120#p757529

The new 16 core Mac pro is crippled with 6 instances of U-HE Diva playing 8 voices each.. too much for single core load so it can only load 6 and leave all the other cores free.. it's the single core performance that is at fault here.. My iMac pro gets 6 also, yet my MacBook pro gets 15.. yes an 8 core MacBook pro gets 15 x 8 voice divas all playing back without logic overloads, vs 6 in a mac pro configured to about 14 grand.. There's a large topic at gearslutz explaining it all (the last 5 or so pages are the ones to read about this from memory), but basically OS X is not allowing the Xeons to keep a steady all core turbo speed and drops below Intel's actual rated speeds, whereas Windows is letting them perform at their maximum. For example in my iMac pro, i get 12 divas in Pro tools/bootcamp vs 6 in Pro tools/Mojave. It's a power limit issue in OSX and we all hope it gets worked out. The reason the MacBook pro gets one per available thread in Logic is because there is enough headroom left over after each one, to put the next one on.. With the mac pro's, the reason they can only play 5 or 6 is because even though those first few instances can be loaded on one core each, every new instance adds overhead even to previous cores, just slightly, but there is no headroom left in the single core performance so it flakes out much quicker. The 9900K gets like 20 or something.

Yet, with omnisphere, there is a demo of the new mac pro playing 80 of them and all 16 cores/32 logical cores getting hammered to the max. So really, it's only U-HE synths that seem to not be able to be fully taken advantage of on the Xeons.. I have other heavy cpu synths and I can use them all.. just not Diva.
All the Arturia stuff is fine, Dune 3, the legend which sounds *amazeballs* used about 1/5th the cpu of DIva's moog filter mode. Try the legend, you will love it.. Only 8 voices but he used an AVX trick to get it to perform like other synths with one voice!
 
OP
BenHicks

BenHicks

Professional Lurker
I've seen that as well and I'm glad you've highlighted this. The i9 (or even the i7 in some cases) will absolutely crush the Mac Pro in single core performance, so synths like Diva will perform *much* better on those CPU's. That's to be expected, granted you've done your research, but not everyone will be aware of this. If you primarily use Diva or some other CPU hungry softsynths, then I definitely wouldn't recommend the new Mac Pro.

It all comes down to workflow and the type of music you write. I don't own Diva or use *too* many synths like that in my music. I work primarily with large track counts utilizing Kontakt/Sine/Play instruments, so for my purposes, the 16 core has proven to be well worth it. Older projects that used to crush my 8 core trash can barely scratch 30% of my total CPU on the new Mac Pro. I've even found that plugins like the B2 are handled much better. For example, I could barely have 8 instances of the B2 reverb used as sends on my old mac. Now, I can throw 80 of them directly onto individual tracks if I wanted to, as was the case in my last video (and that's on top of all the other processing in the chain).

I created this thread in order to help others get a sense of what the new Mac Pro can/can't handle in order to help others decide if it's right for them. It's a big investment, so people should know what they are getting themselves into. If you use Diva or other single core reliant VI's/plugins, I definitely wouldn't buy one. I'd go with an i9/i7 machine. If your projects have large track counts of primarily Kontakt/Sine/Play instruments, I think you'll be quite happy.

I'm hoping that other members who decided to pick up one of these machines will share their experiences, particularly if they went with the 12 core or even the 24/28 core.
 
OP
BenHicks

BenHicks

Professional Lurker
I ran through that Omnisphere test again, but replaced the B2 with Valhalla Room instead. I've decided to skip creating a video, since that's too much of a pain for me to deal with, especially with my sorry internet speeds.

Here's a refresh on the project settings inside Logic X for that test:

__________________
DAW: Logic Pro X (Ver. 10.4.8)
Sample Rate: 44.1kHz (Turns out my logic project settings default to 44.1 for some reason. Had I known that was the case, I would've changed it to 48, since I never work in 44.1)
Buffer: 256

FX Chain:
FabFilter Pro-Q 3
Soundtoys Tremolator
Soundtoys Decapitator
FabFilter Pro-C 2 (4x Oversampling enabled)
Valhalla Room (Default settings at 20% mix)

RESULT: 103 tracks
__________________


For Cubase 10 users, I replicated everything, including all of the exact plugin settings. The only thing I changed was the sample rate/bit rate. For Cubase, I set it to 48/24. Here are the project settings:

__________________
DAW: Cubase 10 (Ver. 10.0.5)
Sample Rate: 48kHz
Bit Rate: 24
Buffer: 256
ASIO Buffer: High

FX Chain:
FabFilter Pro-Q 3
Soundtoys Tremolator
Soundtoys Decapitator
FabFilter Pro-C 2 (4x Oversampling enabled)
2CAudio B2 (Duo Den - Cinematic Long Hall - [17.3 mix])

RESULT: 71 tracks
__________________

And again with Valhalla Room instead of the B2:

__________________
DAW: Cubase 10 (Ver. 10.0.5)
Sample Rate: 48kHz
Bit Rate: 24
Buffer: 256
ASIO Buffer: High

FX Chain:
FabFilter Pro-Q 3
Soundtoys Tremolator
Soundtoys Decapitator
FabFilter Pro-C 2 (4x Oversampling enabled)
Valhalla Room (Default settings at 20% mix)

RESULT: 94 tracks
__________________


I wanted to test with Seventh Heaven, but, annoyingly, whenever I try to copy that plugin over to the next track, it doesn't keep the same preset/settings. There's no way I'm gonna waste my time tweaking dozens of individual settings. Not even for science. lol
 

Nick Batzdorf

Moderator
Moderator
This is the 2020 Mac Pro I got for Christmas. It has 28 cores, 1.5TB of RAM, two of the most expensive video cards they have, and an 8TB SSD. (Of course I also bought the $6000 display and the $1000 stand.)

It's Logic running 50 Omnispheres with a 5-note chord looped, set to a 128-sample buffer. Each track has five stereo Space Designers.

1577763459364.png
 

Dewdman42

Senior Member
By the way, here's my 5,1 MacPro playing 40 tracks of Diva, with the buffer set to 64 no less, but no tracks are record-enabled.

Screen Shot 2019-12-30 at 10.13.07 PM.jpg