What's new

2019 Mac Pro DAW Benchmark Test Results

My 2009 Mac Pro has had an upgrade a few years ago to a 12 core Xeon 2x2.93 motherboard with 48GB Ram, AMD Fire Pro D700 graphics etc. It’s been fine up till now for anything I threw at it both video and audio wise, but I seem to have come up against a wall with my recent Spitfire BBC SO purchase. Like many others I suspect I’m wondering what machine to replace it with - a decent iMac with limited upgrade potential, or the new Mac Pro with all its newbie problems manifesting on this thread? I know we are all in the same boat, so watching this thread interestedly. What are people going for to run a library like BBC SO at the moment?
 
My 2009 Mac Pro has had an upgrade a few years ago to a 12 core Xeon 2x2.93 motherboard with 48GB Ram, AMD Fire Pro D700 graphics etc. It’s been fine up till now for anything I threw at it both video and audio wise, but I seem to have come up against a wall with my recent Spitfire BBC SO purchase. Like many others I suspect I’m wondering what machine to replace it with - a decent iMac with limited upgrade potential, or the new Mac Pro with all its newbie problems manifesting on this thread? I know we are all in the same boat, so watching this thread interestedly. What are people going for to run a library like BBC SO at the moment?

I have been keeping an eye on the new Mac Pro threads here and at Gearslutz since I'm kinda thinking about trading up from my iMac Pro 10 core.

First, I have not seen any real issues reported other than performance with certain circuit-modeled synths like Diva/Repro and questions about the clock frequency control (runs below what is expected based on the stated all-core turbo specs). The Diva thing is real, but IMHO not particularly relevant for how many actually make music (as opposed to running benchmark tests). The patch used earlier in this thread for a test of 50 tracks on a 5.1 MP is a mono patch and based on some really cursory tests on my iMP, I could likely get at least 60-80 instances. However going to one of the poly patches with a track having 5-10 voices, you are talking 6-12 tracks. And these limitations appear to be due to both very poor scaling of the plugin (design/coding) as well as possibly some oddness with how the plugin manages threads and CPU resources on Xeons as opposed to i7/i9 CPU's - although that is far from clear so far.

My take on this is while it would be nice to run 200 voices of Diva in divine mode, I can't imagine ever arranging something that used anywhere near that number of voices with a synth like Diva. It's an interesting test of single core performance, but may not be a real-world limitation for most folks.

The clock speed not being as high as expected may be a bug in firmware/Catalina for the nMP, or it may have to due with the AVX512 instruction set. Intel specs the turbo frequencies to be much lower when AVX512 is being used and we do not know for certain (rumor is Logic uses AVX512) which plugins/DAWs make use of AVX512.

That said, everyone who is using a nMP seems to be pleased with the performance on everything except synthetic benchmarks with certain plugins.

If you style of music making relies almost exclusively on dense arrangements of circuit-modeled analog synths (like Diva), you might be more comfortable with an i9 iMac. If, like many of us here, you typically reply on large number of tracks of sample-based instruments (Kontakt/PLAY/UVI etc), you will likely be over the moon with a nMP (cost aside).

Two things to think about on the iMac approach:

1) It again depends on workflow and compositional style, but the iMacs (and Mac Minis) have very real potential for thermal throttling or loud fans (or both) under prolonged high loads. The iMac Pro and Mac Pro have far more robust thermal designs.

2) The iMacs and Minis have a single TB3 bus with 2 ports. The iMac Pro and Mac Pro have 2 busses with 4 ports (double the bandwidth). Whether this is a problem depends on your storage and peripheral configuration. Since I'm using UAD Satellites and multiple TB/USB3.1 connected SSD's for sample streaming, I need/want the 2 TB busses.

If you are looking for a new rig because of Spitfire stuff, I suspect you would be quite happy with a nMP. Of course whether tha is a cost effective choice is a decision only you can make. There is also no doubt that from a CPU standpoint, an i9 iMac would clobber your 5.1 MP as well assuming you dont hit the TB bandwidth or thermal speedbumps.
 
I have been keeping an eye on the new Mac Pro threads here and at Gearslutz since I'm kinda thinking about trading up from my iMac Pro 10 core.

First, I have not seen any real issues reported other than performance with certain circuit-modeled synths like Diva/Repro and questions about the clock frequency control (runs below what is expected based on the stated all-core turbo specs). The Diva thing is real, but IMHO not particularly relevant for how many actually make music (as opposed to running benchmark tests). The patch used earlier in this thread for a test of 50 tracks on a 5.1 MP is a mono patch and based on some really cursory tests on my iMP, I could likely get at least 60-80 instances. However going to one of the poly patches with a track having 5-10 voices, you are talking 6-12 tracks. And these limitations appear to be due to both very poor scaling of the plugin (design/coding) as well as possibly some oddness with how the plugin manages threads and CPU resources on Xeons as opposed to i7/i9 CPU's - although that is far from clear so far.

My take on this is while it would be nice to run 200 voices of Diva in divine mode, I can't imagine ever arranging something that used anywhere near that number of voices with a synth like Diva. It's an interesting test of single core performance, but may not be a real-world limitation for most folks.

The clock speed not being as high as expected may be a bug in firmware/Catalina for the nMP, or it may have to due with the AVX512 instruction set. Intel specs the turbo frequencies to be much lower when AVX512 is being used and we do not know for certain (rumor is Logic uses AVX512) which plugins/DAWs make use of AVX512.

That said, everyone who is using a nMP seems to be pleased with the performance on everything except synthetic benchmarks with certain plugins.

If you style of music making relies almost exclusively on dense arrangements of circuit-modeled analog synths (like Diva), you might be more comfortable with an i9 iMac. If, like many of us here, you typically reply on large number of tracks of sample-based instruments (Kontakt/PLAY/UVI etc), you will likely be over the moon with a nMP (cost aside).

Two things to think about on the iMac approach:

1) It again depends on workflow and compositional style, but the iMacs (and Mac Minis) have very real potential for thermal throttling or loud fans (or both) under prolonged high loads. The iMac Pro and Mac Pro have far more robust thermal designs.

2) The iMacs and Minis have a single TB3 bus with 2 ports. The iMac Pro and Mac Pro have 2 busses with 4 ports (double the bandwidth). Whether this is a problem depends on your storage and peripheral configuration. Since I'm using UAD Satellites and multiple TB/USB3.1 connected SSD's for sample streaming, I need/want the 2 TB busses.

If you are looking for a new rig because of Spitfire stuff, I suspect you would be quite happy with a nMP. Of course whether tha is a cost effective choice is a decision only you can make. There is also no doubt that from a CPU standpoint, an i9 iMac would clobber your 5.1 MP as well assuming you dont hit the TB bandwidth or thermal speedbumps.

Will mention that Apple addresses thermal throttling in the i9 iMacs by actually throttling wattage. Here's a review: . Basically, it somewhat neuters the full performance of the CPU at high load. Still gets pretty hot, but with a fan control app can cool it down. Don't know how loud the fans are on the thing though.

Also, the 2018 mac minis have 2 TB3 busses and 4 TB3 ports I'm pretty sure. The iMacs do only have the 1 bus and 2 ports though.
 
Oops, you're correct. Still only a 6 core i7 though, so not too compelling compared to the 8 core i9 in the iMac.

Agreed. It's pretty wild that the mini benches multicore similar to my hacked 5,1 MP 12-core. Single core, it does even better. Would be pretty wild to see a mini refresh with the i9 that's in the MBP (where the minis are all laptop CPUs).

I'd kill for a Mac tower that's HEDT and not Xeon based. The iMac is actually priced decently for what it is, but about $1k of that is a 5k monitor I don't really care that much about. Imagine a tower with i9 X-series processors? It'll never happen because it'd kill their pricing model, but a man can dream. (I have weighed the idea of a hackintosh, but it just seems too risky atm).
 
Imho the diva tests floating around are highlighting a potential problem with the nMP that may show up later when trying to use increasingly cpu-heavy instrument or fx plugins as they come out. Right now people are trying to blame diva as if it is the culprit but I think the only thing diva is doing wrong is using a lot of clock cycles, which increasingly sophisticated plugins will need to be able to do. Diva just happens to be a bit at the bleeding edge for that.

Generally speaking plugins do not have any influence or control over core usage, that is all managed by the host and OS X. In rare exceptions some plugins offer some ability to create more threads on other cores, such as kontakt and diva; as an OPTION. Sometimes those features can be helpful and sometimes not. Most people turn off kontakt’s multicore option for example.

I ran the recent diva benchmark on my 5,1 with and without multicore option enabled. I got 7 tracks without and 9 tracks with it on. So that helped me in that case.

But the fundamental problem with all macs today is that they are based on a speed stepping cpu architecture that is designed to keep the cpu clock speed as low as it can get away with to control the heat. That cpu tech was primarily introduced for laptops originally since they don’t have space for a big heat sink, etc. Apple calls it “turbo”. But bear in mind the purpose of turbo is not to optimize cpu speed for performance, it is to optimize down the cpu speed for heat control. All the mbp’s, minis and iMacs need that heat control and the vast majority of Apple users are not playing the hottest games or running daw plugins and they don’t need an Olympic sprinter for a cpu. They need something that doesn’t fry the cpu and looks pretty nice on their ikea office desk.

With the non towers this heat control is of very high priority. With the nMP it is probably not as necessary but hard to say for sure right now until and unless Apple changes something so that the nMP will run the cpu at higher clock rates. Right now it appears to be constrained to lower clock rates, as are the other macs too, but still the i9 clock rate, even constrained, is higher then the Xeon constrained clock rate.

In daw work we are really in need of higher clock rates. Any constraining to those cpu speeds is going to seriously effect things when cpu heavy plugins are used.

In the pc world, people over clock their computer to the highest clock speed possible and use water cooling and other things to do it. The nMP does not have water cooling. It has better cooling then the other Apple products but still pales in comparison to what is possible commonly on a pc. Therefore it will probably not be possible to crank up the nMP to always turbo mode for sustained periods like is possible on a pc today. I think the constrains Apple is placing on the Xeon clock speeds are not by accident.

In some applications, work can be spread to many threads across cores and then a 16 core monster would be more then fine even at constrained clock speeds. But for daw work that is a serious concern because of the way threads are typically allocated per track or channel. That means each core needs the clock speed to prevent drop outs.

I look forward to seeing more tests with more cpu hungry plugins besides diva so we can really see what the nMP can do with heavy dsp on tracks. For me it’s a concern about the platform.

Other people that use few fx in every track and lightweight instrument dsp may never have a problem with it. But then a 5,1 would also probably work fine for that user too.

I have also read that people trying to run xeons full speed with many cores are unable to get all the cores up to the rated max clock speed. They say smaller core counts are a lot preferable for overclocking. I’m sure that is related to heat and/power requirements. But none of that fares well for the prospect of expecting a 16 core nMP to ramp up to theoretical max clock speeds via “turbo”
 
Last edited:
The clock speed not being as high as expected may be a bug in firmware/Catalina for the nMP, or it may have to due with the AVX512 instruction set. Intel specs the turbo frequencies to be much lower when AVX512 is being used and we do not know for certain (rumor is Logic uses AVX512) which plugins/DAWs make use of AVX512.

Pretty sure this is correct. I have a hackintosh, overclocked it for a couple days and decided it wasn't worth it, (the gains didn't justify the high temps and higher bills)... Anyway long story short...

I had an AVX offset when I overclocked the machine and noticed in Logic that the all core turbo dropped by exactly the AVX offset amount I set when OC'd, whereas Logic would hit full speed when not OC'd...
 
Sorry if i put out a different perspective on the diva thing but couldn’t you work in a way like you would with hardware synths? Meaning committing faster to audio and move on?
 
Overclocking is not for everyone, but bear in mind, that overclocking is an advanced technical topic and it really takes quite a lot of time and effort to dial in just the right motherboard settings for any given configuration to get acceptable heat and some amount of improved clock speed. Its not a black or white thing. You can dial it up bit by bit until its too much and then dial it back and it will be fine.

Generally, to achieve a reasonable amount of overclocking you will want to make sure you have a bigger power supply, a better-then-standard heat sink with proper thermal paste in place, etc. This will allow you to push the settings before it starts getting too hot or wonky. The CPU will do wonky things once it either is pushing the limit of what the power supply can provide or the excess use of power causes all the heat to exceed the improved heat sink and fans you put in. In extreme overclocking, you would need to get liquid cooling. Most people don't need to overclock to the extreme level, you can overclock a little bit. The i9 is capable of going to 5ghz, but a lot of people would be in a very good position with it just running at something higher then 4ghz all the time, for example.

There are numerous settings that you have to play around with in any given motherboard and its not an exact science as everything effects everything else. Memory timings, multipliers and several different things. You have to do the research on the motherboard you're using to find out what other people are doing and then you have to play around experimenting until you can find the sweet spot where the cpu clock speed is higher then before without causing the CPU to start to do wonky things from not enough power or too much heat. The only way you find out if its doing wonky things is basically because when you run a sustained cpu stress test for quite a long period, it should be able to run without corrupting the data (the tool will keep track of that) and it should not blackscreen or shutdown. If it can sustain that for a long time and you're comfortable with the fan noise and temperature readings while doing that...then you've found the sweet spot...and your sweet spot might not be as high as some game player that has a tricked out water cooled monster machine, but it will still definitely be better then the stock base clock speed. You also have to be careful NOT to try to run your main operating system while testing all this out because if the CPU does wonky things it can corrupt your data and corrupt your OS installation, forcing you to reinstall from scratch. You should be doing all stress testing way before attempting to run your carefully configured OS with stable and tested overclock settings. Like I said, this is not for the faint of heart.

I had an overclocked hackintosh for years and it was not liquid cooled, just a better-then-stock heat sink, bigger power supply then originally had, and an extra fan. It still runs like a charm now overclocked as a windows machine. The factory clock speed on that is 2.4ghz and it has been running at around 3.2hgz for years. I can't even imagine letting it run at 2.4ghz that would be so dog slow in comparison and why? It was WELL worth the effort for me.

But I have spent, literally hours and hours, dialing in the settings, monitoring the temperature until I was comfortable that its ok, etc.

I would venture a guess that most of the people on this forum do not want to mess around with any of that.

But I do feel the base rate of most CPU's is well below what they are capable of. There is no reason to have an i9 slugging along below 4ghz, for example. Apparently the Turbo technology is supposed to allow Apple computers to run very slow and quiet when not in use, but then ramp up under load. The reports around are that nMP is not ramping up very well under load... That's all.. The i9 has a higher base clock speed to begin with so it already has an upper hand that way, but apparently the i9 macs are ramping up much more liberally for some reason. And so the effective clock speed difference between an i9 mac and an nMP is bigger than reported due due to that it would seem.

How does any of this overclocking talk relate to the nMP?

As I said before, people that do overclocking on PC's, report that machines with high core counts are unable to do high overclocking settings, and often feel that not all of the cores are getting overclocked properly. The CPU starts to do wonky things a lot sooner, constraining their ability to overclock very high. There are theories about it, but I suspect that unless you're hearing it from an engineer at Intel its just theory. But I suspect its related to power and heat...with more cores...more heat starts to happen a lot quicker...so there is only so high they can overclock.

Even though you aren't overclocking an Apple computer, its not even possible.. But that is EXACTLY what turbo mode is supposed to be doing automatically for you. It's supposed to detect load and then dynamically ramp up the clock speed, as if it were being overclocked dynamically under load.

Can you see how for various reasons related to power supply, heat, high core counts, etc...the nMP may in fact be kind of limited in its ability to ramp up the clock speed on say 16 cores? Thus it would be stuck down in the lowly base clock speeds..which are mediocre to begin with compared to say the i9 macs. This big difference in effective resulting clock speeds between these two machines is, in my opinion, why someone recently reported that a 16 core nMP could only run 6 tracks of Diva while an iMac with i9 could run 20 tracks of the same project.
 
Last edited:
Sorry if i put out a different perspective on the diva thing but couldn’t you work in a way like you would with hardware synths? Meaning committing faster to audio and move on?

Unfortunately no it really doesn't work that way. Smarter algorithms can be faster then dumber algorithms in that they require less DSP cycles to get something done, so a lighter weight plugin does what you are imagining, it uses the fewest clock cycles possible and gets out of the way. If a plugin is using more cpu cycles, then its take more time to get it done, it's not like the cpu runs faster for one plugin vs another. A cpu-hog plugin requires more cycles and they require more time. If they take so much time that they can't get the buffer filled in time, then that's when you get audio drop outs.
 
Sorry if i put out a different perspective on the diva thing but couldn’t you work in a way like you would with hardware synths? Meaning committing faster to audio and move on?


this is exactly why i think bands like prodigy, chemical brothers and crystal method where so good back in the day. they needed to commit and it was a tough choice and went with it. now they have endless undos and options. sometimes limits are great.

but in this context though.. for media scoring the option to just be able to undo and change the synth later on if there is picture edit its extrmely important. if you composer to picture and there is an edit which involves changing the melody or harmony then getting back to the same sound is a bit cumbersome and getting another sound will get the director asking why that cool synth was changed and add it back in.
im sure there are other reasons, but thats the one it comes to mind.
logic has freeze and disable the track so its a good workaround. but the best is to not have any technology limitations when creating.

what was this thread about again? lol
 
Interesting. That article also mentioned the following card which holds 6 M.2 SSD modules. https://amfeltec.com/pci-express-gen-3-carrier-board-for-6-m2-or-ngsff-nf1-pcie-ssd-modules/
That looks great! $800 for the card though. You could potentially have 3 of those in the spare 16x slots, each loaded with 6x 4TB drives for 72TB of ultra fast storage! Total - $15k. :shocked:

My budget won't quite stretch that far though, so for now I'm just running a pair of 2TB M.2 drives in a RAID0 using super cheap PCIe adapters. That gives 4TB with a write speed of 5GB/s and a read speed of 4.3GB/s. :)

Unfortunately I can confirm that the reports on the Asus 4x4 ASUS Hyper M.2 X16 PCIe 3.0 X4 are correct. The new Mac Pro does not support bifurcation and so the card only makes a single drive available.
 
That looks great! $800 for the card though. You could potentially have 3 of those in the spare 16x slots, each loaded with 6x 4TB drives for 72TB of ultra fast storage! Total - $15k. :shocked:

My budget won't quite stretch that far though, so for now I'm just running a pair of 2TB M.2 drives in a RAID0 using super cheap PCIe adapters. That gives 4TB with a write speed of 5GB/s and a read speed of 4.3GB/s. :)

Unfortunately I can confirm that the reports on the Asus 4x4 ASUS Hyper M.2 X16 PCIe 3.0 X4 are correct. The new Mac Pro does not support bifurcation and so the card only makes a single drive available.

so you have 2 pcie cards - each for one m.2 and you can still set both to raid0?
 
My 2009 Mac Pro has had an upgrade a few years ago to a 12 core Xeon 2x2.93 motherboard with 48GB Ram, AMD Fire Pro D700 graphics etc. It’s been fine up till now for anything I threw at it both video and audio wise, but I seem to have come up against a wall with my recent Spitfire BBC SO purchase. Like many others I suspect I’m wondering what machine to replace it with - a decent iMac with limited upgrade potential, or the new Mac Pro with all its newbie problems manifesting on this thread? I know we are all in the same boat, so watching this thread interestedly. What are people going for to run a library like BBC SO at the moment?

I just delivered a few projects using BBCSO, all from my MB Pro. I have nearly every instrument loaded, for a total of 10.5GB (I only have 16GB). Granted, the orchestrations were not massive, but if load only one articulation per instance the SF Player is very efficient. I also have it loaded on my slave, but now I don't think I'll ever need it.
 
I think the MacPros with more than 8 cores are more interesting than the entry-level machine.

The iMac with an i9 processor costs half as much and delivers more processing power. If you can get by with max 128 GB RAM the i9 is sufficient.

It's different if you use older libraries with high resource requirements (VSL, HOD) for which VEP slaves are often used today.
A well developed MacPro with more than 256 GB Ram and 12 - 16 cores is then already interesting.

Mathematically I don't see any reason for a MacPro at the moment, I'd rather add a MacMini or a MacBook Pro as slave to my i9, which is about the same but costs much less (instead of a 12 or 16 core mp).

If you don't do video editing and can do without the internal expandability, you have enough alternatives. At least as long as you don't need more than 2 installations of the VI licenses.
 
Top Bottom