What's new

2019 Mac Pro DAW Benchmark Test Results

Yea well on another forum someone sent me an LPX project to test diva on my 5,1. It was able to play at least one more track then the 16 core nMP according to what he reported. So yea something is definitely amiss, the nMP should have slayed my 5,1.

My 5,1 doesn’t have AVX at all. I don’t agree with the AVX related theories floating around unless we hear it directly from Urs’ mouth. That is pure speculation by the gearslutz crowd trying to guess at it.

but yea there is definitely something amiss.

we need to see more tests with another equally cpu hungry plugin And other standardized situations.

I may have started teh AVX512 thing over there as a result of simply "what's different about Xeon-W vs i9 or Xeon-E/Westmere.

Repro-5/1 and Massive X exhibit the same sort of "combinatorial explosion" in thread loading as Diva. Roland Jupiter-8 too, although it isn't as extreme.

Zebra, Omni, Alchemy, the Roland D-50/1080/5080, Kontakt, Korg M1/Wavestation and TAL-U-NO-LX are very well behaved. I don't own the Korg analog synths to try.
 
Fascinating. Using Cubase, my Intel 4790k hackintosh can do about 32 instances of Omnisphere with the same per-track processing. Only difference is the reverb (I dont have the 2C Audio B2 so used Liquidsonics Reverberate instead).

I decided to do a little more testing with Diva (too much free time). I loaded Logic and Diva on my 2017 MBP (3.1GHz quad core Kaby Lake i7) and compared some scenarios with my 10 core iMP. I used the Deep Space Diva preset in "great" quality mode with multithreading both on and off. I used an 8 voice chord that I just held for several bars. I monitored CPU use with LPX, iStat and Intel Power Gadget. I checked the freq with iStat and IPG. THe laptop was plugged in to maximize performance. All other apps were shut down. Both are running Catalina and LPX 10.4.8

iMac Pro 10 core:

1) With multithreading ON, 8 tracks. CPU was about 85-90% loaded across all 20 cores (iStat/IPG), 3.75-3.85 GHz

2) With multithreading OFF, 6 tracks. CPU was about 35% loaded but the load was spread only on the 10 physical cores in iStat. performance meter in LPX didn't look much different except for fewer threads used.

2017 Macbook Pro:

1) With multithreading ON, 5 tracks. CPU was 60-65% loaded across all 8 cores (iStat/IPG), freq was 3.5 GHz for the first few cycles until the temp shot up, fans went crazy and freq started dropping.

2) With multithreading OFF, 11 tracks !!!???. CPU was still 60-65% loaded, same frequencies.


This all seemed fairly reasonable (although the iMP should have done better considering both the slightly faster clocks and 2.5X cores) until I turned multihreading off on the MBP. The i7 doubling the track count with multithreading turned off is rather amazing since it was the opposite effect of what happened on the Xeon. The multithreading option in Diva behaved on the Xeon as I would expect (in spite of the modest improvement in LPX track count). I have no idea what is going on with the i7.

To be completely fair about all this, With Omni, Alchemy, Zebra, Kontakt or the standard Logic benchmarks using VI's and effects within Logic, the iMP & nMP kills the MBP (as it should). However, this tells me that something bizarre is going on with MacOS+Xcode+current Xeons with whatever coding techniques/libraries are being used in heavyweight analog-modeled, oversampled synths like Diva & Repro.

Yes, it appears the 2019 MP's are underachieving on clock speeds in some cases compared to the iMP's, but the real news here is the plugin behaving completely different on an i7 than a Xeon-W. Clock Speed Matters, but I think the reason current 8 core i9 iMacs and Macbook Pros are killing the nMP's on these type of tests is primarily some compiler/OS anomaly on the different architectures, not just the single core speed difference.
 
One more thing - the latest version on Intel Power Gadget has a speed test menu now. You can select single core, all-cores or a couple other options. And that load test can be run with normal scalar instructions, AVX-256 or AVX-512 (i7 & i9's do not have AVX-512 instructions - the menu item doesn't appear there). On the iMP (socket 2066 Xeon-W Skylake-X part), running the all-core AVX-512 test drops the speed from 3.8+ immediately to 3.4. However, I always see 3.7-3.8 in Logic unless I'm above 90% total CPU load per Activity Monitor. So while my previous SWAG that the Xeon/desktop "thing" might be due to AVX-512 compiler weirdness is not disproven yet, it seems more likely that the lower clocks relative to all-core turbo on the 2019 Mac Pros are probably not due to extensive use of AVX-512.
 
My opinion at the moment is that Xeon is throttling down the clock speed as more cores are utilized more. This is probably due to the fact that more power is needed when more cores are hitting higher levels of utilization. Also more heat produced. In any case, super cpu intensive plugins would start to suffer if and when the clock speed goes down.

Having more cores is only helpful when you can spread the work load out to many threads, but a heavy plugin is heavy on one thread and needs the clock speed to keep up.

So basically when you have a few tracks going, fine...clock speed is fine, but as you add tracks, which adds threads, and multi-core utilization starts to increase...then the clock speed starts to drop and eventually a heavy cpu plugin like Diva will start to crap out.. whereas other plugins that are not as intensive on the CPU can still make it through and allow more and more threads to be created, utilizing more and more cores.

The i7 and i9 appear to not be throttling down the clock speed, rather the opposite...they start from a low idle clock speed and throttle it up as load increases.
 
My opinion at the moment is that Xeon is throttling down the clock speed as more cores are utilized more. This is probably due to the fact that more power is needed when more cores are hitting higher levels of utilization. Also more heat produced. In any case, super cpu intensive plugins would start to suffer if and when the clock speed goes down. Having more cores is only helpful when you can spread the work load out to many threads, but a heavy plugin is heavy on one thread and needs the clock speed.

So basically when you have a few tracks going, fine...clock speed is fine, but as you add tracks, which adds threads, and multi-core utilization starts to increase...then the clock speed starts to drop and eventually a heavy cpu plugin like Diva will start to crap out.. whereas other plugins that are not as intensive on the CPU can still make it through and allow more and more threads to be created, utilizing more and more cores.

The i7 and i9 appear to not be throttling down the clock speed, rather the opposite...they start from a low idle clock speed and throttle it up as load increases.

The iMP and MP appears to behave as if Apple disabled the C-state/SpeedStep stuff in the EFI and the clock speed is managed by the OS to presumably hit some desired power target. It does not appear to be driven by actual temperature. The question is why Apple seems to be so conservative given the thermal capacity of the 2019 MP.

The laptops behave like all the Intel speed-fiddling is turned on. The clock is low under light load to keep temps down and preserve battery and only ramps up as the load increases and more horsepower is required. It also seems to be actually using the temp as well, unlike the Xeon machines.

The real news though is that plugins like Diva behave radically different on Skylake/Cadylake Xeons than on all of the i7/i9 desktop parts, regardless of clock speed differences.
 
The iMP and MP appears to behave as if Apple disabled the C-state/SpeedStep stuff in the EFI and the clock speed is managed by the OS to presumably hit some desired power target. It does not appear to be driven by actual temperature. The question is why Apple seems to be so conservative given the thermal capacity of the 2019 MP.

(shrug). OSX or in the motherboard...no idea.. matters not to me..that is internal Apple decision making. but it may be necessary because more cores being used closer and closer to 100% means more power draw and it may not have enough power to drive all 16 cores full tilt, so to speak without lowering the clock speed as you approach that.

Temps are a result of more power being used. The nMP has better temperature control then most laptops to be sure so it should be able to handle higher temps, but power is also a factor. if the machine starts to use more and more power, not only will the temps go up, but at some point the power-starved cores will start to produce errors.

The laptops behave like all the Intel speed-fiddling is turned on. The clock is low under light load to keep temps down and preserve battery and only ramps up as the load increases and more horsepower is required. It also seems to be actually using the temp as well, unlike the Xeon machines.

I agree. But its not necessarily just laptops. This is probably also a factor of the i7/i9 architecture with typically less number of cores.
 
Interesting...


Apple's Macs could soon have the option for a "Pro Mode" that boosts performance according to data found in the macOS Catalina 10.15.3 beta.

Released in December, the beta contains code discovered by 9to5Mac that references an optional Pro Mode that appears to override fan speed limits and energy saving restrictions to improve performance. Code in the beta says "Apps may run faster, but battery life may decrease and fan noise may increase" when Pro Mode is turned on.

As with Do Not Disturb, Pro Mode is a temporary option that is set to turn off the next day after being enabled, likely to make sure Macs are running in optimal conditions.

The code found in the beta seems to suggest Pro Mode is coming to the 16-inch MacBook Pro, but MacRumors has also received an anonymous tip suggesting the Mac Pro will also get a Pro Mode option that will override maximum fan speed software limits for improved thermal performance.

More information on Pro Mode should be available when macOS Catalina 10.15.3 sees a release, provided this is the update where it will be introduced.
 
I put a Highpoint 7101A RAID card in today with 3x 2TB SSDs (saving for a 4th!). The read/write speeds are great, but bear in mind this is mainly for video where large sequential files are more common. It might not work as well for sample libraries where 4k speeds are more important.

Screen Shot 2020-01-14 at 2.03.34 PM.png
 
Temps are a result of more power being used. The nMP has better temperature control then most laptops to be sure so it should be able to handle higher temps, but power is also a factor. if the machine starts to use more and more power, not only will the temps go up, but at some point the power-starved cores will start to produce errors.
That’s not going to happen as it would imply a serious design flaw that we would know about by now as there are billions of Intel CPUs that use on chip performance management.
The CPU monitors itself very frequently and adjusts clock speed and voltages primarily, based on measurements of thermals, power usage, demand etc.
The chips are carefully calibrated to work within certain parameters so what you suggest is very fanciful.
 
Exactly.

I wasn't meaning to imply that MacPros are going to produce errors, I was speaking hypothetically about what WOULD happen if CPU clock speed management was not in place to throttle down the CPU clock speed under load.

That is why, by design the macpros have to slow down the clock speed as core utilization goes up. It is that way by design to prevent errors as well as overheating. It is a physical constraint of the hardware that 16 cores can’t run full tilt.

unfortunately that it’s also one reason why this architecture is not great for digital audio.

If you are questioning my assertion that CPUs can produce errors when starved of power, then do a little research about overclocking. That is exactly what happens. The system is calibrated to run within limits where that won’t happen. When you push the system it will happen. It will either overheat from all the power use or crash from not enough power. If the macpro did not throttle the cpu clock speed down while increasing core utilization it could cause one or the other to happen.

And this throttling down has already been measured and observed by Xeon-w users
 
Last edited:
Exactly.

I wasn't meaning to imply that MacPros are going to produce errors, I was speaking hypothetically about what WOULD happen if CPU clock speed management was not in place to throttle down the CPU clock speed under load.

That is why, by design the macpros have to slow down the clock speed as core utilization goes up. It is that way by design to prevent errors as well as overheating. It is a physical constraint of the hardware that 16 cores can’t run full tilt.

unfortunately that it’s also one reason why this architecture is not great for digital audio.

If you are questioning my assertion that CPUs can produce errors when starved of power, then do a little research about overclocking. That is exactly what happens. The system is calibrated to run within limits where that won’t happen. When you push the system it will happen. It will either overheat from all the power use or crash from not enough power. If the macpro did not throttle the cpu clock speed down while increasing core utilization it could cause one or the other to happen.

And this throttling down has already been measured and observed by Xeon-w users

So, could something be done to fix this on Apple's part? Or are us Mac Pro owners screwed? o_O
 
Well they just announced a so called power user mode that they may be bringing into catalina, which allegedly will run the fans at high speed for 24 hours before it automatically resets back to the previous behavior. No idea how that will effect the new MacPros, we shall see.

I think this is fundamentally a hardware constraint, its not simply something that can be fixed. In the overclocking community many people complain that machines with large core counts do not overclock well. What that means is that the clock speed on 12 or 16 or 28 core machines simply cannot run as fast as machines with less cores. And I believe the primary reason is related to power draw. But power draw and heat are interlinked with each other. Its the power that creates the heat. When you overlock you have to find the happy balance between turning up the voltage to give your CPU, memory and motherboard enough power to run at a higher clock speed, but that in turn creates more heat. At some point you are either not giving it enough power for the desired clock speed, and the system crashes, or if you ramp up the voltage...presuming your power supply can handle it...and then you start to get a lot more heat.

So overclockers play around with that to try to go beyond the safe limitations imposed by the the default settings from Intel mainly. Intel knows with reasonable certainty that if the cpu operates within certain limits, it will not crash and will not run too hot...so systems are shipped and spec'd with fairly conservative values. Overclockers try to push that limit by improving cooling in their system in order to push more power into it, and raise the clock speed without problems. They have to test their system until they find a setting that doesn't crash and doesn't run the fans too much or whatever...and then they can enjoy a faster-then-stock computer.

Overclocking is not available on macs. I'm only mentioning that for some perspective about why Turbo mode even exists at all on a mac. They are attempting to provide a system that will attempt to find a safe happy medium between power and heat and no chance the cpu will crash due to errors. Turbo mode provides a little more leeway by dynamically changing the clock speed according to the load, hopefully pushing that safe zone a little higher in general.

The i7/i9 have less cores..and are often used in Laptops where thermal problems come up more often. So they start with slow cpu speed when idle and only increase the clock speed under load (along with fans and power pushing into it.

The MacPro is using a different CPU architecture, Xeon, which has many more cores...and now apparently the behavior is to run it at ideal clock frequency when idle, but to rather slow down the clocks as more cores are utilized in parallel more, so that it will not be starved of power or overheat. Its kind of the opposite turbo behavior compared to the i7/i9's, same general overall goal, to try to optimize what they system can do without blowing up.

Perhaps this new power user mode that will be coming out for Catalina will be able to minimize how much the CPU's throttle down under load, or perhaps the base frequency will be raised in some way, as overclockers do manually on PC's, perhaps not. It will undoubtedly be much more conservative then what overclockers typically do because it has to run absolutely without crashing...and without the temp raising too high...even if the fans are blaring the whole time.

We shall see, but I don't expect to see that much of an improvement honestly because a lot of overclockers complain that big core counts are simply difficult to run at full clock speed...all the cores consume a lot of power and generate a lot of heat.
 
Last edited:
I had a bad stick of ram which didn't let the mac pro boot up and made the fans run at full speed. Loud as hell but holy shit can they move a ton of air. Hoping pro mode takes maximum advantage of that
 
Exactly.
I wasn't meaning to imply that MacPros are going to produce errors, I was speaking hypothetically about what WOULD happen if CPU clock speed management was not in place to throttle down the CPU clock speed under load.

That is why, by design the macpros have to slow down the clock speed as core utilization goes up. It is that way by design to prevent errors as well as overheating. It is a physical constraint of the hardware that 16 cores can’t run full tilt.

unfortunately that it’s also one reason why this architecture is not great for digital audio.

If you are questioning my assertion that CPUs can produce errors when starved of power, then do a little research about overclocking. That is exactly what happens. The system is calibrated to run within limits where that won’t happen. When you push the system it will happen. It will either overheat from all the power use or crash from not enough power. If the macpro did not throttle the cpu clock speed down while increasing core utilization it could cause one or the other to happen.

And this throttling down has already been measured and observed by Xeon-w users
Ah, you were talking about over-clocking without clarifying that which is why I was scratching my head.:eek:
You seem to be over thinking it, what with over-clocking and the theoretical disabling of the CPU’s own management system!
Either the system doesn’t have the cooling muscle to allow the CPU to hit its various maximum Turbo speeds, or it does but is not allowed to utilise it fully due to noise levels being prioritised.
 
I was trying to respond to the comments from others about why its not ONLY about temperature. Try to keep up.
 
I was trying to respond to the comments from others about why its not ONLY about temperature. Try to keep up.
The Mac Pro is not overclocked, so unless you think the system is unable to supply the correct voltage/amps to the CPU, highly unlikely, the only reason for it to not hit the maximum turbo speeds is down to inadequate cooling or Apple not allowing it for other reasons.
What's that got to do with overclocking? :eek:

The max all core turbo on the 28 core will be around 3.8GHz so has anyone confirmed this?
 
Please go back and read my posts more carefully before responding carelessly. I never said they are overclocked. Read more carefully. Learn something!

The Xeon-W appears to turn the clock speed DOWN as core utilization is increased. the discussion about overclocking is only to shed some light on why, which you appear to still not understand, so go read it all again.

cheers
 
The Xeon-W appears to turn the clock speed DOWN as core utilization is increased.
Pretty much ALL Intel CPUs these days have different maximum turbo boost speeds based on the number of core loaded; those with more cores will see a bigger gap between the turbo boost speeds
THIS IS NORMAL BEHAVIOUR; nothing new to see here.

Do you actually know the rated all core speed for the 28C chip?
It has a base of 2.5 and max turbo is maybe 3.8.
Has it been reported that it can't hit 3.8 with an all core load?
That would be an actual fact for a change.
 
Top Bottom