Due to the level of interest in the newly-released 2019 Mac Pro, we’ve taken the opportunity to run some performance tests which help determine exactly what we can expect from the new machine, compared with some other choice models.
Whilst Apple have now released a technical document describing the architecture of the new Mac Pro, it tells little about the performance of the machine in a real-world professional audio context - other than repeating their marketing claim that it runs 6.5x the Amp Designer plug-ins compared with the 2013 Mac Pro.
Apple’s performance claims for the 2019 Mac Pro in respect of Logic Pro X performance
Our tests have primarily focussed on finding the limits of performance when running Logic Pro X, and looked to establish the following:
Multi-core CPU performance - as measured by the processing of multiple simultaneous tracks;
Single-core CPU performance - as measured by the processing of a single, complex track and the off-line exporting of multiple simultaneous tracks;
Disk I/O performance - as measured by intensive audio export and playback.
The Headlines
As this is going to be really rather dry, let’s look at the headline outcomes before diving into the methodology.
The new Mac Pro can offer significant performance gains over the previous generation of Mac Pro, ranging from 50% - 600% increases, depending on the area of usage.
We tested the performance against a number of other Macs - details on the machines under test are noted here.
All the performance charts are compared against the Late 2013 ‘Trashcan’ Mac Pro 4-core.
Multi-core CPU Performance
The multi-core CPU usage for a machine is displayed as activity in the left-hand-most bars in Logic’s Performance Meter.
Test running simultaneous Alchemy plug-ins
Test running simultaneous Amp Designer plug-ins
Test running simultaneous Space Designer plug-ins
Further details here.
Summary
The new Mac Pro typically offers between 4-7x the performance of the previous generation Mac Pro;
Unfortunately we’re not able to test exactly the machines Apple quote (Late 2013 Mac Pro 12-Core and Late 2019 Mac Pro 28-Core), but their claimed performance boost of 6.5x between the two machines running Amp Designer plug-ins is roughly borne out by the boost between the machines we have measured (both at comparable point in their respective ranges);
Amp Designer appears to be highly optimised for the new Mac Pro (or vice versa), resulting in the large performance gain in this combination. However this performance gain is not realised with the other plug-ins under test - it would be unwise to expect to gain this much performance increase across the board. However, it is possible that there are improvements in the new Mac Pro which, when adopted by other developers, will yield similarly impressive performance gains;
Other than this, the measured performance is roughly as predicted by the benchmarks.
Single-core CPU Performance
The significant single-core CPU usage for a machine is displayed as activity in the right-hand-most bar in Logic’s Performance Meter.
Test running multiple Amp Designer instances on a single track
Test running multiple Space Designer instances on a single track
Export time for multiple tracks
Further details here.
Summary
The new Mac Pro typically offers around 2x the performance of the previous generation Mac Pro;
Single-core CPU performance has not progressed as quickly as the multi-core performance over the last few years. Indeed - CPUs have become massively multi-core because the single-core performance hasn’t increased hugely;
Consequently, although the capability for running ever-greater numbers of tracks has increased significantly, the possible complexity of a single live track has not become massively greater;
That said, the single-core performance is still greater on any of the current models under test compared with the previous Mac Pro generation;
In this respect, the new Mac Pro offers little advantage over other current models - including the Late 2018 Mac Mini.
Disk I/O Performance
The performance of disks containing audio is not just important for audio track recording and playback, but also for off-line bounce performance and - probably most importantly - software sample streaming.
Disk read speed using Blackmagic Disk Speed Test
Export time for 500 simple tracks
Disk usage for playback of 500 tracks
Further details here.
Summary
The new Mac Pro offers around 1.5-3x the performance of the previous generation of Mac Pro.
The performance of internal disks has increased substantially between the previous Mac Pro generation - which in itself was already impressive- and the current T2-powered Macs, including the new Mac Pro;
However, the performance does vary depending on what tasks are being performed - ranging from 1.5x to 3x the previous Mac Pro performance.
For Your Consideration
Different Models
Looking at these results, it’s useful to be reminded of the spread of performance offered by the different models in the professional Mac range, as predicted by the benchmark scores: whilst the single-core CPU performance is largely unchanged between ‘serious’ current machines, the multi-core CPU performance is significantly higher on the 28-core new Mac Pro compared with even the 12-core new Mac Pro tested here.
Noise Performance
The noise of the different machines is also a consideration here: under full load, the different machines do have different noise performance:
| Remains silent - Apple have designed the fan controller so that the perceived noise remains low at all times |
|
| Remains nearly silent | |
| Low noise, but definitely not silent | |
| Clearly audible noise |
Conclusion
The performance increase of the new Mac Pro compared with previous generations of Mac is extremely impressive in respect of its capability of processing multiple simultaneous tracks. However the single-track performance is only a little better than other currently-available Macs - though other factors, such as the noise performance, may still make the new Mac Pro the optimum choice.
Test Machines
Our test suite comprised various significant models of professional Mac from recent years.
| Early 2009 Mac Pro |
Late 2013 Mac Pro |
Late 2018 Mac Mini |
Late 2019 MacBook Pro |
Late 2019 Mac Pro |
|
|---|---|---|---|---|---|
| 2.26GHz | 3.7GHz | 3.2GHz | 2.6GHz | 3.3GHz | |
CPU Cores |
8 | 4 | 6 | 6 | 12 |
| Intel Xeon "Nehalem" |
Intel Xeon E5 "Ivy Bridge-EP" |
8th Generation Intel Core i7 "Coffee Lake" |
9th Generation Intel Core i7 "Coffee Lake" |
Intel Xeon W | |
Range |
Mid-Range | Low-End | High-End | Low-End | Mid to Low-End |
| 10.13.6 | 10.15.3 | 10.15.3 | 10.15.3 | 10.15.3 | |
| SATA HDD | PCIe SSD | T2 SSD | T2 SSD | T2 SSD |
‘Cheesegrater’ Mac Pro - Early 2009 2.26GHz 8-Core Intel Xeon “Nehalem”
This was a mid-range model from the 2009 Mac Pro generation, being the low-end of the 8-core models - there were also three lower-end 4-core models;
Officially this Mac runs OS X 10.11.6 as its latest OS - but as the current version of Logic Pro X does not run on this OS, we installed macOS 10.13.6 to run these tests;
We attempted to run macOS 10.15.3 for the tests, to maintain a like-for-like comparison - but although the OS and Logic installed, Logic crashed as soon as we pressed play (the Mac Pro’s default GPU does not support the required software);
Of the test machines, this is the only one to be running a spinning disk drive - all the others were running SSDs.
This was the low-end of the 2013 Mac Pro models, and remained the current Mac Pro until the release of the new machine in late 2019;
At the time of its release, the new Mac Pro was met with some level of indifference with regards its CPU performance compared with its predecessors - though its internal SSD showed very high performance, and the fast I/O ports allowed for a wide range of fast peripherals to be connected;
This range runs OS X 10.9 Mavericks as its earliest supported OS, and will run the current macOS Catalina release.
Mac Mini - Late 2018 3.2GHz 6-Core 8th Generation Intel Core i7 "Coffee Lake"
This is the high-end model from the current Mac Mini range. The Mac Mini had languished for some time, with the prior 2014 release actually being a step down in capability compared with the 2012 models;
The Late 2018 range was released with a big nod to the professional market, with high performance processing options. Most professional users will be best served by this high-end model;
The internal SSD is controlled by the Apple T2 chip, providing high performance;
This range runs macOS 10.14 Mojave as its earliest supported OS, and will run the current macOS Catalina release.
This is the low-end model from the new 16” MacBook Pro range;
Aside from the slightly larger display, the new range features faster processors, increased RAM support (up to 64GB), increased internal SSD options (up to 8TB) and a refreshed keyboard (compared with the previous generation);
The internal SSD is controlled by the Apple T2 chip, providing high performance;
This range runs macOS 10.15 Catalina as its earliest supported OS.
Mac Pro - Late 2019 12-Core 3.3GHz Intel Xeon W
This is a mid-to-low-end model from the new Mac Pro range;
The internal SSD is controlled by the Apple T2 chip, providing high performance;
This range runs macOS 10.15 Catalina as its earliest supported OS.
The test results here are normalised around the performance of the Late 2013 ‘Trashcan’ Mac Pro, to give a representative comparison against current machines. Each machine ran an identical set of Logic settings, based around a 64 sample I/O buffer using the Mac’s built-in audio output.
Our tests were primarily run on the current versions of macOS - 10.15.3 Catalina - and Logic Pro X - 10.4.8.
Multi-core CPU Performance - Further Information
In order to maintain optimum efficiency and provide the highest performance, Logic Pro X uses a large buffer size for any tracks which do not require a ‘live’ output. This gives it some degree of flexibility in terms of allocating resources across multiple processors, and accommodating demands from background applications and processes without interrupting playback.
The CPU usage for such tracks is displayed as activity in the left-hand-most bars in Logic’s Performance Meter.
Testing the multi-core CPU performance was undertaken with 3 separate tests:
Software Instrument Performance - Multiple Alchemy software instrument tracks;
Effect Plug-in Performance #1 - Multiple Amp Designer audio tracks;
Effect Plug-in Performance #2 - Multiple Space Designer audio tracks.
In order to ensure we were measuring only the multi-core CPU performance, we took a number of precautions:
We ensured that an empty track was selected during the playback tests, so no ‘real’ track was using the ‘live’ CPU core in addition to the ‘non-live’ cores;
We tested audio tracks with multiple instances of each effect plug-in, so as to reduce the influence of the disk playback performance on the result;
We tested two effects plug-ins, as Apple were claiming very high performance improvements with the Amp Designer plug-in, so we wanted to see if this was a special case.
We’ve also included the Geekbench score for each machine, to give some indication as to how the real-world DAW performance compares against the synthetic benchmarks.
Multi-core CPU - Software Instrument Performance
This test used a number of software instrument tracks, each with a default instance of Alchemy loaded, playing back a single repeated note.
Notes:
The real-world performance roughly tracks the theoretical performance, sometimes exceeding it;
There are substantial performance gains to be had from the Mac Mini, MacBook Pro and new Mac Pro, compared with the ‘Trashcan’ 2013 Mac Pro or earlier;
The mid-range ‘Cheesegrater’ Early 2009 Mac Pro was roughly comparable with the low-end ‘Trashcan’ 2013 Mac Pro.
Multi-core CPU - Effect Plug-in Performance #1
This test used a number of audio tracks playing back a repeated region of pink noise, with each track holding 8 default instances of Amp Designer on the track.
Notes:
The real-world performance roughly tracks the theoretical performance - until the new Mac Pro, which massively exceeds it.
Multi-core CPU - Effect Plug-in Performance #2
This test used a number of audio tracks playing back a repeated region of pink noise, with each track holding 8 default instances of Space Designer on the track.
Notes:
The real-world performance much more closely tracks the theoretical performance for the 2019 Mac Pro, compared with the performance for Amp Designer;
Otherwise the performance roughly correlates with the first chart.
Single-core CPU Performance - Further Information
In order to run ‘live’ inputs - that is, any audio track which is in input-monitoring mode or any software instrument track which is primed to respond to incoming MIDI - Logic Pro X uses a small buffer size and processes the track and some master plug-ins on a single CPU core.
The CPU usage for such tracks is displayed as activity in the right-hand-most bar in Logic’s Performance Meter, and is the cause of most processing overloads: a single complex instrument and effects plug-in chain can overload a system more readily than hundreds of ‘non-live’ tracks.
Testing the single-core CPU performance was undertaken with 3 separate tests:
Plug-in Performance #1;
Plug-in Performance #2;
Off-line Track Export Performance
In order to ensure we were measuring only the single-core CPU performance, we took a number of precautions:
As with the multi-core CPU performance tests, we tested two effects plug-ins, as Apple were claiming very high performance improvements with the Amp Designer plug-in;
We ensured the single track was in live mode, to ensure all its plug-ins were being processed on a single CPU core.
We’ve also included the Geekbench score for each machine, to give some indication as to how the real-world DAW performance compares against the synthetic benchmarks.
Single-core CPU - Plug-in Performance #1
This test used a single software instrument track running an instance of Alchemy, playing back a single-note repeated region. Multiple instances of Amp Designer were inserted on the track, plus on Bus 1 (fed from the track), plus on the Stereo Output.
Notes:
The ‘Cheesegrater’ 2009 Mac Pro would not perform this test correctly, and allowed a full complement of plug-ins to be inserted without any error message - despite the audio output going unnaturally silent and the CPU usage never peaking beyond 87%. Consequently the figure used here is the number of plug-ins the system could handle before the CPU usage stopped rising;
Like the multi-core performance, the real-world single-core performance substantially exceeds the theoretical performance again;
Whilst still a significant gain over older machines, single-core performance has not increased as far as the multi-core performance.
Single-core CPU - Plug-in Performance #2
This test used a single software instrument track running an instance of Alchemy, playing back a single-note repeated region. Multiple instances of Space Designer were inserted on the track, plus on Bus 1 (fed from the track), plus on the Stereo Output.
Notes:
Interestingly, the capabilities here very closely match those of the previous, Amp Designer-based, test. This contrasts significantly with the multi-core CPU tests;
Again, the performance significantly exceeds the theoretical performance gain for the new Mac Pro;
It’s possible the new Mac Pro would have been capable of a few more plug-ins - but the limit of 15 plug-in slots of each of the instrument, aux and output channels was reached.
Single-core CPU - Off-line Track Export Performance
This test used 30 software instrument tracks, each with a default Alchemy plug-in, 2 x Amp Designer plug-ins and 2 x Space Designer plug-ins. All tracks were exported offline, a process which is run on a single CPU core - the time taken to do so was measured.
Notes:
Whilst the new Macs still demonstrate a performance gain over the older models, the gain is not as significant as with the other single-core CPU tests.
Disk I/O Performance - Further Information
The performance of disks containing audio is not just important for audio track recording and playback, but also for off-line bounce performance and - probably most importantly - software sample streaming. A disk with limited I/O performance will limit the number of simultaneous record and playback audio tracks, will increase the time required to bounce tracks, and will limit the number of voices which can be simultaneously played in software samplers.
Whilst there are a number of disk benchmark applications about, many simply measure the raw throughput of a disk’s performance; whilst this can be a good approximation for real-world performance for video applications (which require a high data rate for relatively few files at any one time), audio applications require modest data rates - but typically for hundreds of files simultaneously: each audio track and each microphone position of every note being played in a software sampler will require access to a different file (or part thereof) at the same time.
In the days of spinning hard drives, this could markedly reduce the apparent performance of disks in that the time required to ‘seek’ to a new file would typically be limited by the rotational speed of the disk: the drive would position the disk head to the right position across the spinning disk’s platters, and then it would have to wait for the required physical location on the disk to move into position under the head. Multiply this process by hundreds of different files at once, and the seek time becomes an extremely significant factor in a disk’s performance.
With modern SSDs, there are no physical moving parts - so accessing different areas of the disk could (theoretically) be no different from accessing sequential areas of the disk. In practice, there are factors that complicate matters - the SSD’s controller active spreads data across its flash memory chips in order to prolong its life, and will fetch data off the chips pre-emptively and store it in a short-term cache in order to deliver future requests more rapidly. Consequently it’s difficult to predict how the real-world performance for an application will compare with the theoretical benchmarks.
Testing the disk performance was undertaken with 3 separate tests:
Benchmark Performance;
Off-line Track Export Performance;
Audio Track Playback Performance.
Disk I/O - Benchmark Performance
This test used the Blackmagic Disk Speed Test application to perform streaming data tests on the drives. The figures shown here are for Read performance only:
The figures varied up to 20% between subsequent runs of the test, so they should only be treated as rough guidance;
The Write performance was so close to the Read performance it was within the margin of error.
Notes:
Whilst the Trashcan Mac Pro was a disk I/O miracle when it was released, it has now been utterly trounced by Apple’s newer models. This is presumably due to the custom T2 security chip that controls the SSDs, as well as performing other underlying functions for the system;
All the newer models feature comparable performance - including the new Mac Pro.
Disk I/O - Off-line Track Export Performance
This test used 500 tracks, each with a pink noise Signal Generator plug-in on it. All tracks were exported offline (which is processed on a single CPU core) for a 100 bar duration, and the time taken to do so was measured. Whilst there is an element of single-core CPU performance variation in this test, the disk performance is the dominant limitation.
The benchmark comparison here is taken from the above Blackmagic Disk Speed Test.
Notes:
The disparity between the Blackmagic Disk Speed Test performance and the tests here show how the disk speed varies according to the precise nature of the application being undertaken;
Again, although the modern Macs perform better than the older models, then only offer a ~50% improvement over the Trashcan Mac Pro.
Disk I/O - Audio Track Playback Performance
This test used 500 tracks, each with a pink noise audio file on it. The tracks were played back together, and the performance of the system was noted from Logic’s Disk I/O meter.
The benchmark comparison here is again taken from the above Blackmagic Disk Speed Test.
Notes:
The spinning hard disk drive in the ‘Cheesegrater’ 2009 Mac Pro couldn’t actually play back 500 tracks at all - the maximum it would allow was 128 tracks. Consequently the figure used here is the extrapolated amount of disk usage it would have taken to deliver 500 tracks;
The performance here otherwise more closely mirrors the performance measured by the Blackmagic Disk Speed Test, with modern machines delivering 500 tracks without breaking a sweat.
Bonus Track
We also had opportunity to run a couple of these tests with an iMac and iMac Pro for comparison:
| Retina 5K 2019 iMac |
Late 2017 iMac Pro |
|
|---|---|---|
| 3.7GHz | 3.2GHz | |
CPU Cores |
6 | 8 |
| 9th Generation Intel Core i5 |
Intel Xeon W | |
Range |
Mid-Range | Low-End |
| 10.15.3 | 10.15.3 | |
| T2 SSD | T2 SSD |
Multi-core CPU - Software Instrument Performance
This test used a number of software instrument tracks, each with a default instance of Alchemy loaded, playing back a single repeated note.
Single-core CPU - Plug-in Performance #1
This test used a single software instrument track running an instance of Alchemy, playing back a single-note repeated region. Multiple instances of Amp Designer were inserted on the track, plus on Bus 1 (fed from the track), plus on the Stereo Output.
Single-core CPU - Plug-in Performance #2
This test used a single software instrument track running an instance of Alchemy, playing back a single-note repeated region. Multiple instances of Space Designer were inserted on the track, plus on Bus 1 (fed from the track), plus on the Stereo Output.