What's new

2019 Mac Pro DAW Benchmark Test Results

you are still not understanding

i7/i9 starts from a low cpu speed when the computer is idle and ramps up the CPU clock speed on an as-need basis.

Xeon-W does the opposite. It starts at the highest clock speed its going to ever get to and as more cores are utilitized under load it turns the clock speed down.

This has been observed through independent testing. Xeon-W does NOT accelerate to maximum core speed when all cores are being utilized, it does exactly the opposite...

The reason is explained above in my verbose explanation about overclocking which you didn't read carefully.
 
you are still not understanding

i7/i9 starts from a low cpu speed when the computer is idle and ramps up the CPU clock speed on an as-need basis.

Xeon-W does the opposite. It starts at the highest clock speed its going to ever get to and as more cores are utilitized under load it turns the clock speed down.

This has been observed through independent testing. Xeon-W does NOT accelerate to maximum core speed when all cores are being utilized, it does exactly the opposite...

The reason is explained above in my verbose explanation about overclocking which you didn't read carefully.
That's not what my iMac Pro does. It is at low clock speeds when idle and ramps up with load. It's just the base 8 core. I've seen it go down to 1.8ish GHz, when nothing is happening, to 4 GHz under load. For what it's worth...
 
you are still not understanding
i7/i9 starts from a low cpu speed when the computer is idle and ramps up the CPU clock speed on an as-need basis.
Xeon-W does the opposite. It starts at the highest clock speed its going to ever get to and as more cores are utilitized under load it turns the clock speed down.
This has been observed through independent testing. Xeon-W does NOT accelerate to maximum core speed when all cores are being utilized, it does exactly the opposite...
The reason is explained above in my verbose explanation about overclocking which you didn't read carefully.
I read that and you can do something similar or possibly the same in Windows by changing the power settings.
It's not ideal and has some potential downsides I think but I do it sometimes if the usual clock speed changes lead to audible glitches in my DAW which can happen. Then I set the clock speed to the maximum which disables any down clocking; my cooling can handle it.

That's just a configuration option and in itself has no bearing on the maximum rated turbo boost speeds.
This is not exactly esoteric stuff.
I'd still like to know what all core boost speeds people are getting with the 28C chip. Anybody?
 
That's not what my iMac Pro does. It is at low clock speeds when idle and ramps up with load. It's just the base 8 core. I've seen it go down to 1.8ish GHz, when nothing is happening, to 4 GHz under load. For what it's worth...

someone else reported to me their clocks slowing down. let's see a detailed report! I'm not opposed to calling the other person wrong, but I have heard this report from more then one place too. does it not slow down when the core usage comes up? I could also accept an explanation that it idles slow, ramps up with some cpu use, but then backs off as core usage goes up... which is still part of the fundamental problem.
 
Last edited:
I read that and you can do something similar or possibly the same in Windows by changing the power settings.
It's not ideal and has some potential downsides I think but I do it sometimes if the usual clock speed changes lead to audible glitches in my DAW which can happen. Then I set the clock speed to the maximum which disables any down clocking; my cooling can handle it.

Reports are in that the nMP running windows on bootcamp is able to crunch higher scores in benchkmark testing then the same machine booted into OSX. So its certainly possible that some power related settings are allowing the CPU's run hotter.

Perhaps the power user mode that Catlina is going to release soon will be something similar along those lines, we shall see.
 
someone else reported to me their clocks slowing down. let's see a detailed report! I'm not opposed to calling the other person wrong, but I have heard this report from more then one place too. does it not slow down when the core usage comes up? I coold also accept an explanation that it idles slow, ramps up with some cpu use, but then backs off as core usage goes up... which is still part of the fundamental problem.
I'll do some testing and get back to you.
 
I will PM you about that, I have a JUCE plugin I made to help with testing that, its not ready for public release though, but will send it to you.
 
someone else reported to me their clocks slowing down. let's see a detailed report! I'm not opposed to calling the other person wrong, but I have heard this report from more then one place too. does it not slow down when the core usage comes up? I coold also accept an explanation that it idles slow, ramps up with some cpu use, but then backs off as core usage goes up... which is still part of the fundamental problem.
I didn't suggest they are wrong as I can configure my PC to do something similar as I already stated.
It's quite a common thing for DAW users to do as it can reduce the chance of audible glitches due to latency issues as the clock speed jumps around.

Did they say what the maximum speed they noticed?
 
I will PM you about that, I have a JUCE plugin I made to help with testing that, its not ready for public release though, but will send it to you.
I was just going to do the old track duplication + heavy effects load trick, but I could check out your plug.
 
The cooling looks interesting based on this video and this 12C Xeon does appear to keep the clock speed high:


Note: The clock speeds may well be whilst bench-marking and not at idle; my bad.
 
I was just going to do the old track duplication + heavy effects load trick, but I could check out your plug.

People have been having problems with Diva which is notoriously high CPU, but some unknowns about why Diva is a problem and much speculation about different things which nobody really knows. I just thought it would be useful to have a consistent plugin that produces CPU load in a predictable way we can all use to compare. The final version will be free for everyone when its done, and will aim to provide a consistent way to measure different systems and DAW's against each other...over time it may improve as people contribute ideas about how it should be pushing DAW's differently. Right now its just spinning the CPU, but I am adding features to emulate memory use more and other things...
 
Over on one of the 2019 MP threads at GS, Urs from U-he just posted he took delivery of a 16 core MP last week and has been chasing the Diva performance issue noted on the recent Xeon-based Macs. He found the problem and has a RC build on KVR - final release (1.4.4) should be in the next week or 2 according to Urs.

Someone tried the RC - reported something like 25 instances versus 5 with the current release version!
 
Just reran the original Diva test from GS - here is my post over there:

iMac Pro 10 core, Apollo X6

Running the original "Diva vs Bootcamp" project TNM uploaded when this all started. Using the 1.4.4 RC from KVR.

Original test with Diva 1.4.3 - 6 tracks

With RC 1.4.4 - 30 tracks!! with multicore=OFF and quality=great
thumb.gif
thumb.gif
thumb.gif


Urs fixed the hell out of this!!

BTW, my clock varies a bit between 3.76-3.86 as this runs. Sitting at about 4.12 before I hit play

Going to turn multicore on just to see what happens now

UPDATE: Catalina (10.15.2) and Logic 10.4.8
 
@BenHicks great tests, I've had mine for a few months and it's been incredibly solid with Logic, Pro Tools and Ableton 10. This week I decided to upgrade to Cubase Pro 10.5 to build a larger, more efficient orchestral template and I'm running into some sort of weird glitch with tempo ramps and Kontakt instruments specifically.

Have you run any tests with tempo changes using Kontakt patches (both orchestral and synths) with and without tempo-synced articulations? If I set the buffer to around 1024 or 2048 I still get some really bizarre CPU spikes and audio dropouts. Kinda scary after the beast of a machine I've witnessed firsthand. I didn't think I could crack it.

Have you experimented with adjusting the Kontakt core settings manually? Or do you keep them off and let the Cubase split the load across the cores?

SPECS:
MacPro 2019
3.2 16 Core
96 GB ram
AMD Radeon Pro 580x (baseline)

Sample Libraries on external Samsung EVO SSD via Thunderbolt 3.0

RME Fireface UFX II as the audio interface.
 
@BenHicks great tests, I've had mine for a few months and it's been incredibly solid with Logic, Pro Tools and Ableton 10. This week I decided to upgrade to Cubase Pro 10.5 to build a larger, more efficient orchestral template and I'm running into some sort of weird glitch with tempo ramps and Kontakt instruments specifically.

Have you run any tests with tempo changes using Kontakt patches (both orchestral and synths) with and without tempo-synced articulations? If I set the buffer to around 1024 or 2048 I still get some really bizarre CPU spikes and audio dropouts. Kinda scary after the beast of a machine I've witnessed firsthand. I didn't think I could crack it.

Have you experimented with adjusting the Kontakt core settings manually? Or do you keep them off and let the Cubase split the load across the cores?

SPECS:
MacPro 2019
3.2 16 Core
96 GB ram
AMD Radeon Pro 580x (baseline)

Sample Libraries on external Samsung EVO SSD via Thunderbolt 3.0

RME Fireface UFX II as the audio interface.

I have pretty much the same setup as you and have also been disappointed by the performance on Cubase. Would you say that Logic is more efficient or does there seem to be something wrong with Cubase specifically?
 
I have pretty much the same setup as you and have also been disappointed by the performance on Cubase. Would you say that Logic is more efficient or does there seem to be something wrong with Cubase specifically?

Honestly, I'm not sure because I have seen Cubase templates with worse computer specs running more. efficiently. For example Daniel James (of hybrid two). He's running an old trashcan mac with fewer processors, lower processor speed, and a little less ram with Cubase 10.5, but I think he's on macOS Mojave? If that is so, and Catalina is the culprit when it comes to communicating correctly between an OS and DAW's, we'll just have to wait it out I guess. Perhaps Cubase 11?

Logic X is really good, but I just can't get around the setup and feel of that DAW. Cubase seems like the logical choice for organization, MIDI editing, programming, large templates, etc.

It's really a bummer that the new Mac Pro 2019's can't be downgraded to Mojave, I have a feeling that would solve a ton of issues. But alas, this is truly one of THE most frustrating aspects of this industry. Upgrades, downgrades, compatibility issues and BAM the issue is fixed with perfect synchrony; but only for a couple of months before some hidden auto-update happens overnight and throws the whole system into chaos. SMH.
 
Cubase 10.5 feels dog-slow on my Mac Pro 7,1 with Catalina, especially compared to my Windows 10 PC. Logic is snappy and quick.

As the Diva update shows, there's a lot to be said for optimizing software for new hardware and operating systems. I suspect there are performance gains to be had with Cubase on Catalina if Steinberg puts in the man hours.
 
Turn off sync in Kontakt. Don’t use ramp in tempo changes.
YES! I've written on this many times... it is super important to take this into account. Tempo ramps while using tempo sync will bring powerful systems to their knees. I've learnt to write VERY intricate steps into the tempo track... because I use instruments who need tempo sync in Kontakt. Its frustrating beyond belief.
 
I’m thinking about picking up some OWC Accelsior 4M2 PCIE SSD’s to put in my Mac Pro to run my sample libraries. Has anyone seen or heard of these yet? Anyone know how in the world 6,200mb read/write speeds are possible?
 
Top Bottom