New Logic Multitrack Benchmark Test

[Ploki]

Try asking u-he support about performance tho...

Their updates are usually timely and support good.

But diva takes a serious bite out of any CPU.

 

I like the legend but i prefer diva...

[TTOZ]

No I don' have a nMP, I am just saying...on my 2010, I can play Diva with plenty of voices, its not a problem. Other people are saying Diva is choking on the nMP, I'm just responding.. I don't know why the nMP would choke on Diva when my 2010 MacPro with 3.46ghz Xeons (no turbo mode though), can play it fine.

You do realise this diva test where people get 6, is on divine mode and playing back 8 voices all at once (enough notes in the midi file to make the voice steal so we know we are using all 8).

 

Let me see if you can get even ONE on this max quality test..

 

do you want me to link you the logic project file with the Diva test?

 

You have to compare apples to apples (no pun intended).. a NMP can do 100+ divas if it's on draft mode and not playing 8 voices all at a time per instance.

[Dewdman42]

hehe, I will do a proper test right now. I'm sure you're right it will not fare well....I'm just shocked the new nMP is struggling too.

 

if you have a test project for diva, yes please link it.

[TTOZ]

Try asking u-he support about performance tho...

Their updates are usually timely and support good.

But diva takes a serious bite out of any CPU.

 

I like the legend but i prefer diva...

Different thing in that Legend is specifically a moog emulation in int hat one regard, it's better than Diva.

 

overall Diva is much more a jack of all trades type analog, with higher polyphony and can emulate multiple machines.. so... yeah of course it's better.

 

I'll be right back with the logic test project, the one that allows my 8 core IMAC pro and the 16 core max pro to only get 6 instances of diva. (15 instances on the MacBook pro 16" 8 core)

 

If you can get more, you'll need to duplicate. If you can't play the 6, just delete till you can.

[TTOZ]

Name your instances! LOL!

 

https://drive.google.com/file/d/15rteHzMTxPFPgyvLqtYjn5mmDMyLCJEB/view?usp=sharing

 

Cheers!

 

EDIT.. this has been edited for only one DIVA CPU test project link. If you need more DIVAS, duplicate the last track and copy the midi clip as much as you need, and your final instance count is only valid with the *audio* track highlighted, so logic is putting all the Divas on the playback buffer. Playback must be stable for MINIMUM 30 seconds looping, at the default Diva settings the project loads at, WITH only the single audio track that is present being selected.

 

Your playback buffer must not be LARGE, as all the the others were tested at MEDIUM.

 

To get an accurate comparison score, all testers must follow the same rules.

Edited January 3, 2020 by TTOZ

[Dewdman42]

I can play 9 of those tracks without dropouts on my 5,1.

 

Consider that a baseline.....

[Dewdman42]

The fact that the i9 can play 20 tracks is all I need to hear. Unless Apple fixes the Turbo mode stuff on the nMP I will be building a hackintosh next.

[Dewdman42]

Note a couple things..the templates I linked to did NOT have Diva in devine mode. They are in "great" mode. Also, I went into each channel and enabled the multicore setting, which made an improvement.

[TTOZ]

Note a couple things..the templates I linked to did NOT have Diva in devine mode. They are in "great" mode. Also, I went into each channel and enabled the multicore setting, which made an improvement.

yes, it's great quality mode with multicore disabled. That's not your result then. I forgot about that, the other test was made with divine mode later, an updated test for more powerful computers. I sent you the exact test that a 16 core mac pro can get 6 on.. and you HAVE to have the audio track highlighted otherwise logic will just add clicks and pops if ANY of the diva tracks are highlighted when you play back, as those tracks go on low buffer mode which prevents logic from stopping with an overload message.

 

Your maximum tracks with the audio track highlighted, logic playing back for at least 30 seconds looping without an overload message, and diva multicore disabled is the result. How can we compare when you are using a different parameter? The highlighted track I am not sure about, only you can confirm you highlight the audio track on playback, as it is when the project loads.

 

Your result is the result with the test loaded AS IS.

 

I do not believe your 5.1 has better performance with Diva than a new mac pro. I do not buy it for a second, I am sorry but I don't.

 

Your computer has a 2700 geekbench 4 single core vs around 5300 in the new one.

 

because you have so many cores, you are able to get away with it by enabling multi processing.. I don't think your machine could even play back two with multi processing disabled. You gave me the result I didn't ask for LOL!

 

If you are telling me yours can play 9, with the audio track highlighted, and multicore disabled, that means your cheese grater is WAY more powerful than the new 16 core 3.2ghz mac pro for certain VI tasks, and that is 20 grand in Australia for a decent config! If so this needs to be put on VIDEO ASAP so we can send it apple, so they can SORT THIS OUT with an EFI update.. Evidence would be much appreciate to show apple how much the NMP is underperforming with some VI's.

Edited January 3, 2020 by TTOZ

[TTOZ]

For example, someone claimed his mac pro 16 core got 30 of them but he had one of the divas highlighted, which put logic into "no overload but will click and pop" mode.

When run correctly, the result was 6. If you want to put it in multicore mode you can, as long as you first include your "DIVA SETTINGS AS IS WHEN PROJECT LOADS" result.

I have updated the initial post with the link and made the terms and conditions clear, to accurately match the results against all those who have tested with their iMac pros and mac pros at gearslutz. Cheers.

[Dewdman42]

what you sent me is not devine mode. I played what you sent me. what don't you want to use multicore mode?

[Dewdman42]

Dude, you are making this too confrontational. I'm only trying go provide info. If you want to see a baseline from an older Mac Pro, then provide exactly the project you want me to run and exactly the instructions. Also why not use multi-core mode?

 

cheers

[Ploki]

on the test as is (opened and run), i could run 10 tracks on my 2018 i5 13" ... running 10.15.2 10.4.8

i doubt multicore mode would help since 7 threads are running full when i do that.

i'll try on the Mac Mini later when i'm at studio.

 

if my i5 can run 10, a 16-core mac pro should run at least 60... (judging by bencmarks)

 

I don't use DIVA in multicore because it almost always causes crackles on any of my systems. actually both repro and diva do.

[Dewdman42]

Here's a quote from Urs (author of Diva) about multicore mode....

 

It also depends on the CPU. Most effect is seen when 4 cores or more are present and if the CPU has Sandy Bridge or newer. Worst results have been observed on Core 2 Duo machines, where it is indeed better to switch the Multicore option off.

 

On top of that, it isn't good to have more than a few instances of Diva (or any plug-in supporting multicore) enabled with this option, as then the overhead increases. It really is menat to help playing an instance or two live, but it isn't meant to replace track freeze and stuff.

 

Cheers,

 

- Urs

 

Came from this thread: https://www.kvraudio.com/forum/viewtopic.php?t=414398

 

I am running out the door so will try again tomorrow with and without multicore mode. My results initially when I first tried this were that I am getting slightly better core utilization and performance with it on. But this may be a factor of it only being helpful when you have a machine with large core counts...whereas the smaller core higher clock machines should probably leave it off, perhaps...

 

Anyway, with it on, I got 9 tracks out of my crusty old 5,1. and as I recall yesterday I wasn't getting even 7 with it off.

[TTOZ]

Dude, you are making this too confrontational. I'm only trying go provide info. If you want to see a baseline from an older Mac Pro, then provide exactly the project you want me to run and exactly the instructions. Also why not use multi-core mode?

 

cheers

it's called standardised testing. That's why. I am not even going to explain further than that as you simply don't seem to understand nor want to. I said it was divine mode and made an error, I corrected it. Big deal. You obviously don't want to give your results when testing the same way everyone at gearslutz tested, cause you maybe can not even get one, and want to pretend your cheese grater is more powerful than a 2019 mac pro. I CLEARLY updated the instructions in the original post, as well as subsequent posts, but you are fighting with me about my own test.

 

Once again for the LAST TIME:

 

Audio track must be highlighted, no changes made to diva whatsoever. The project must be played as is, with the only changes allowed being to duplicate or delete tracks as necessary till you get a figure you can reliably play for a minute without the overload message.

I don't know how much simpler I can make it than that. Any other result is invalid in THIS test as THAT'S how the mac pro users tested.

 

One guy made a mistake of having a diva track selected, and got 30 on his 16 core mac pro., but there were pops and clicks cause logic's low buffer was activated which doesn't give the core audio overload message. This was therefore an invalid result. When he highlighted the audio track, which means logic will overload before there is EVER any audio fidelity loss, he got 6. That's the valid result IN THIS TEST.

 

If you want to turn on multicore, feel free to do so and report the difference. But the multicore result is NOT valid to compare to the results recorded so far since the test started.

 

If you don't want to test like everyone else thus far to get an actual comparative result, stop wasting my time. This is the last time I post about it to you.

[Dewdman42]

Man dude, calm down. You are just being so confrontational, I don't have any desire to entertain a test for you now. if you want to be nicer about it I will try.

[Dewdman42]

FWIW, I reloaded the test you provided, changed the quality to divine mode on all diva instances, made sure the audio track was selected....9 tracks, same as before. No message box, no audible pops or clicks. Beautiful divine mode Diva.

 

With Multicore mode turned off, I get 7 tracks.

 

Its somewhat debatable about whether multi-core mode should be turned off while comparing a high-core-count machine against a low-core-count with higher clock speed. Its no duh that the higher clock speed will have the advantage when many cores aren't being used optimally. Listen I am a huge fan of i9 also, I wish Apple would build an 8 core i9 nMP, I'd buy one.

 

I suspect the nMP would perform perhaps more admirably with multi-core enabled in Diva, but still the results being reported here are pretty disappointing. It probably is something related to power and turbo mode. I have read elsewhere that high-core-count machines often do not overclock very well, for example, in the PC world. Sometimes not all the cores speed up, etc. Who knows what is happening with the nMP turbo mode as it ramps up the clock speed...is it doing it for all cores? Or is it perhaps just avoiding turbo mode altogether due to power supply concerns or something? We don't really know right now I guess its speculation, but honestly, I got 7 tracks with the test TTOZ is saying that a 16 core nMP could only do 6. That's with divine mode, multi-core turned off, audio track selected.

 

cheers

[Dewdman42]

I want to say also, it does not surprise me that my 5,1 can do 9 tracks of this test compared to 20 in the i9 as reported. This is completely predictable as the benchmark scores between the two machines show about the same difference in scores.

 

What is a big surprise to me is that the nMP is scoring so poorly, however. And in particular, I think the 16 core nMP ought to be able to slay the i9 for multi-core performance and mix down...so... something is seriously wrong with that setup, either the way the testers have it configured, or Apple has design issues in this new mac that are not very conducive to audio production since we depend usually, on each channel taking up a thread, on a core...we need higher clock speeds...and it would appear the nMP not only has a lower clock speed then the i9, but its not ramping it up to its full capacity apparently, making it even worse.

 

I suspect if you were mixing down hundreds of tracks, then all the 16 cores would be utilized more, and possibly the nMP would outperform the i9 in that scenario, but who needs to mix hundreds of tracks really? Mostly we don't. That's why the i9 would have been a much better CPU for an audio production nMP.

 

Perhaps.

 

its also not that common to have 10 or 20 instances of Diva on 20 tracks in divine mode! In a real world 100-150 track mix...could be the nMP does just fine. In mean if you had 100 tracks with stuff typically less cpu intensive then Diva, then the load would spread to all 16 cores and perhaps it would hum along in some way as good or better than an 8 core i9. The i9 will excel when you have particularly CPU-hungry plugins, in live mode or with only a few tracks. But... honestly the fact that the nMP could only 6 tracks of that Diva, should be very concerning to everyone thinking about getting that mac for audio.

[Grilled Cheese PhD]

But... honestly the fact that the nMP could only 6 tracks of that Diva should be very concerning to everyone thinking about getting that mac for audio.

Perhaps a more accurate statement would be “should be very concerning to everyone thinking about getting that mac for DIVA.”

 

Most benchmarks so far indicate that the nMP handles audio very well. Diva seems to cause problems for some unknown reason. We could speculate about this all day, or just accept that Diva performance is not really representative of the way nMPs perform, and move on.

[Dewdman42]

do you have an example of a benchmark that is using some heavy DSP per channel, with something other than Diva? Using a lighter instrument such as Omnisphere does not count. You need a test with say Omnisphere + a reverb plugin + another reverb plugin, etc...some way to see what happens when you have some channels with heavy CPU use. Diva is well known to consume a lot of CPU. I don't particularly like the theory that there is some mystical thing Diva is doing that the nMP cpu instruction set can't handle....that doesn't resonate with me at all. However, I can definitely subscribe to the idea that too much CPU crunching on each channel (each channel needing one thread), could expose the nMP's slower clock speed, especially if its not ramping up to turbo mode well as some have been reporting on Gearslutz and other places.

 

Diva is known to be a very CPU intensive plugin, all in one plugin...so another equally cpu-intensive plugin would be interesting to compare...but also a chain of plugins on each channel that add up to maybe a lot of cpu per thread.

[Dewdman42]

All this makes me realize that some clever JUCE programmer needs to make a benchmarking plugin that is dedicated to no other purpose but creating an adjustable, but standardized cpu load in a consistent way.

[Sascha Franck]

All this makes me realize that some clever JUCE programmer needs to make a benchmarking plugin that is dedicated to no other purpose but creating an adjustable, but standardized cpu load in a consistent way.

 

That'd be quite cool. X-platform, x-format please. Maybe with AVX-support being switcheable.

[croixdj]

well this is it:

 

NewLogicBenchmarkTest.zip

 

I really think I have created the perfect new benchmark test.. and props to Evan, didn't realise how long it would take to get the balance right.

 

On my 2015 macbook pro 2.8hghz quad, i get 37 tracks with turbo disabled or 38 with turbo enabled.. Their really is something to this turbo boost switch software.. difference is.. CPU is at 74 degrees instead of 95!

Turbo boost enabled will initially get 45 or even 50 tracks for about 8 or so bars but then overheats and drops the cpu frequency.. the macbook pro just can't maintain it. And then, "audio overload" message.

 

And regardless, as has been shown with first i9 result below, the test is very intense as is.

 

What is this Turbo boost you speak of? Is that a MacBook Pro thing? I am running a MacPro mid 2012 QuadCore 3.2 with 48GB Ram in triple channel.

I can run at best 29 channels of this test - 28 if I want to be able to do other things in the BG while it runs...

[skijumptoes]

Turbo boost is Intels way of speeding the CPU up until thermals/power is an issue, at which point it slows down ("Throttles back") again. It's how a lot of the modern i5/i7/i9's get great performance - but most of the time it's not 100% sustainable in use, at least not at full boost.

 

You Mac Pro doesn't rely on Turbo Boost so you get much more constant CPU power throughout . This is why benchmarks can be a misleading vs running a machine under load for hours.

 

Not sure what software TTOZ is refering to though, i think with the older macbook pros you can get to the CPU settings as they weren't locked. You can't on the newer models however.

[Ninecows]

Anyone care to try out this benchmark on the new iMac’s (if you got them that is )

 

Thanks in advance

[PaoloT]

I'm a bit late at the game, but it seems that the old Mac Pro is still of fashion…

 

Mac Pro 2013 (6,1), 12-core, 64GB RAM, 2TB SSD. Logic Pro X 10.4.8, 128 samples I/O buffer, Medium processing buffer:

 

115 tracks before overload. Sustained, no degradation.

 

Paolo

[cowudders]

yeah it's so strange, this old grater doesn't want to get old at all - on my 12core 3,46 i also get 115 tracks in the newlogicbenchmark and 23 in the divatest. procbuffer medium, 128 samples buffer - RME raydat on Mojave.

 

Did the tests, because i thought about the new 2020 imacs, but according to some youtube video the i7 8 core gets 123 tracks and the i9 10 core 131 tracks.

[Dewdman42]

Long live the cheesgrater!

[JoshJ]

I ran this on my 2012 MP - got up to 92 tracks before an error. 128 i/o buffer Medium Process Buffer

 

How on earth is that possible?!? My 2018 Mac mini can only run 73 tracks on 1024 buffer!

Edited September 26, 2020 by joshj

[JoshJ]

I just got 73 tracks running before any System Overload messages with the following setup;

 

Kit:

-MacMini 2018 3.2 GHz Intel Core i7 (6 core), 32GB DDR4 RAM

-Running Logic project on internal Startup Disk

 

Environment:

-I/O Buffer: 1024

-12 Threads

-Process Buffer Range: Large

-Wifi Off

-No other apps running

 

-Logic CPU registering at 876% in Activity Monitory

-Core temps hitting between 80-100 degrees

 

 

Can anyone else confirm similar performance on a similar MacMini or whether the above results look reasonable for my system?