Benchmark tests/Findings- AMD 7950x v Macbook Pro M1 Max

[Norbury_Brook]

I recently bought a very nice macbook pro M1 Max with 64GB ram and 2TB drive. A great machine. I have an AMD 7950x-64GB ram machine in the studio running windows 11.

So after spendiing a week installing everything on the Macbook pro I sarted to do some ‘testing’.

The first DAW test i did was the @ pete @Scan/Dom Sigales Cubase test they’d come up with.

How To Benchmark Cubase? Tom & Dom Make a Test to Measure Performance. #cubase #PC #Mac

It’s basically a small project with a few audio tracks and a set of Retrologue VSTi and Reverlation Reverbs on tracks that you duplicate untill you hit the red.

My M1 macbook managed about 90 tracks before it overloaded.

The AMD 7950x windows machine did 170 so a massive difference.

Now, if I’d just left it at that, you would think that the 7950x was therefore a lot more powerful machine than the macbook pro.

However … I had just finished a mix last week on the Macbook and used that as the basis of another ‘real world’ test. So i took my mix with busses,sends, automation, VCA’s etc etc and added a load more plugin to get it as far as I could without it going into the red and saved that as my test mix.

I then loaded that same mix on the AMD 7950x.

I couldn’t play it back at all. Totally overloaded … not just a bit… it wouldn’t even start to playback as it was so in the red.

So , despite the AMD 7950 windows machine showing a clear performance win in the Scan Cubase test , in a real world mix the M1 actually performed a lot better.

I thought that might be interesting to some, it certainly show’s the issues with ‘benchmarks’ and how specific workflows can show that despite being less powerful in one benchmark a machine can actually outperform the more powerful one in a certain situations.

I did this test to see which machine was going to be my main studio workhorse and after the scan test I thought: ’ the AMD is far more powerful , so it will stay as the studio DAW’ . It was only when I tried to open the Macbook Pro mix on the studio daw that i realised for the work i do , it’s actually more powerful !

So my new studio set up is based around the Macbook pro Max as the main machine and the 7950x is now running as an Audio gridder and VE Pro slave. The PC has a 10GB/Sec SFP+ card and the Mac has a USB3 2.5GB/sec attached to the Sonnet TB chassis, so i’ve got some good network bandwidth to play with via a 2.5 and 10GB switch.

M

[MattiasNYC]

@Norbury_Brook

i took my mix with busses,sends, automation, VCA’s etc etc and added a load more plugin

In parallel or in series?

[Tj99]

It could be that:

1. The buffer size was lower on the AMD, so the CPU had to work harder (you didn’t mention buffer size at all)
2. The audio interface driver is way less efficient on Windows. Which interface are you using?
3. There were some DPC latency issues caused by GPU/USB/Network drivers, which come more into play, the heavier the project gets
4. If you look at the performance meter, is it Realtime, Asio guard, or Peak which is maxed out on the AMD one?

Just to explain:
The 7950x is indeed quite more powerful than the M1, which you can also see in your DAW test you mentioned before. In that case the load is spread evenly on all cores and the AMD performs a lot better, because it has a way better multicore performance.
In your own project the audio has to go through different groups, sends, etc. and therefore the load will be much more on the 1st single core.
Now the AMD should beat the M1 even in single core usage, even if just a tiny bit.
But if you’ve got some different settings or driver issues on your system, then these could get amplified even more on the single core.

Would be great, if you could share your findings, as this is indeed very weird.

[Norbury_Brook]

Sure
regrading buffers… with AG guard on… it was on medium …the buffer plays little part. However The buffer was actually lower on the Macbook pro. Regarding interface, the macbook was running off the internal audio and I tried the AMD using a Steinberg UR22-c on high performance mode. I can redo the test tomorrow and use the same interface for each , however the performance difference is way outside this IMHO. The AMD wouldn’t even begin to play back the project never mind struggle with some pops and clicks. In the interest of keeping things truly equal I can use the RME RAYDAT , my main interface , and post the results. The card is in a TB sonnet chassis and i can plug it in to eithe machine as the AMD has TB.

There are no DPC issues with the AMD machine as it’s been my main Studio DAW for over a year and has been rock solid in day to day opperation.

Regarding the perfomance meter it’s ALWAYS the ASIO guard meter that hits the red and stops things on either machine.

here’s a screen grab from the Macbook running the project from the on board sound. You can see the performance meter. I loaded up the project with a lot of Acustica Aqua plugins that I like to use a lot. .

M

[Nico5]

Might also be interesting to know how the RAM is spec’d and configured in the AMD box, because there’s quite a bit of variation possible, depending on the exact RAM sticks and the BIOS configuration.

And in a large project with lots of FX in lots of channels, I imagine there could be a lot of data flying back and forth between the CPU and RAM.

If my understanding of hardware architecture is correct, one of the big differentiators between the new Apple Silicon and AMD architectures is how RAM interacts with the CPU. – Apple’s SOC design may have a distinct advantage in moving more data at lower latency between CPU and RAM.

Related:

https://www.reddit.com/r/apple/comments/y8z714/m1_max_versus_the_new_ryzen_7950x_cpu_for_data/

[Norbury_Brook]

RAM is : 64GB (2x32GB) Corsair DDR5 Vengeance Grey, PC5-48000 (6000), Non-ECC Unbuffered, CAS 40, 1.35V, AMD EXPO. It’s running with the correct EXPO speeds etc .

With ASIO guard on there’s no low latency happening in a project

M

[Nico5]

@Norbury_Brook

RAM is : 64GB (2x32GB) Corsair DDR5 Vengeance Grey, PC5-48000 (6000), Non-ECC Unbuffered, CAS 40, 1.35V, AMD EXPO. It’s running with the correct EXPO speeds etc .

That sounds reasonably maxed out for your AMD system’s hardware without overclocking manually.

[Tj99]

@Norbury_Brook

regrading buffers… with AG guard on… it was on medium …the buffer plays little part. However The buffer was actually lower on the Macbook pro

I meant the buffer size you have to choose in the interface settings, not in Cubase directly. That one makes a really big difference. The lower the buffer size, the lower the interface lateny when recording and the higher the CPU usage.

@Norbury_Brook

In the interest of keeping things truly equal I can use the RME RAYDAT , my main interface , and post the results. The card is in a TB sonnet chassis and i can plug it in to eithe machine as the AMD has TB.

That would be great! RME interfaces are top notch.

@Norbury_Brook

Regarding the perfomance meter it’s ALWAYS the ASIO guard meter that hits the red and stops things on either machine.

Which version did you do the test in? As in C13.0.10 there was a bug regarding the asio guard, which should be fixed in C13.0.20.

@Norbury_Brook

I loaded up the project with a lot of Acustica Aqua plugins that I like to use a lot. .

Yeah I also use some of their plugins, they are really outstanding!
But they are also known to behave quite differently performance and stability wise on different systems. Depending on the system some people can use more or less of them, regardless if one CPU or the other seems more powerful…
Perhaps not the best 3rd party plugins for an apples to apples test. Would be interesting to see a similar test using Fabfilter and Waves plugins only, as they are wider used and probably also less quirky.

@Norbury_Brook

Regarding the perfomance meter it’s ALWAYS the ASIO guard meter that hits the red and stops things on either machine.

Thats interesting and seems strange to me, as my experience showed me, that all three meters should be somehow active, in your screenshot ONLY the asio guard is maxed out, while the other two meters are nearly on 0.

[Norbury_Brook]

I was refering to the interface buffers sizes. this is a mix project so ASIO guard on means that the interface buffer size doesn’t really come into play. However on the Mac it was running @128 samples and on the PC I had it set to 512.

Regarding the interface itself, it really doesn’t matter which one i use as it will be the same on both machines In fact in the test i showed the Mac had a dissadvantage over the AMD in that I was using the on board audio NOT a specific device.

I’ll check the version again today if possible and re run with the same device, however as i said the difference wasn’t slight so i doubt it will make much difference.

regarding the meters if you watch the Dom Sigales/Scan cubase benchmark test you’ll see that it’s awlays the ASIO guard meter that hit’s the red first.

bear in mind this a a real world MIX project, no VSTi’s or midi tracks, at’s all audio and plugins, and heavy CPU /memory ones at that.

The Dom Sigales /Scan Cubase test showed the AMD 7950x to be far more powerful using retrolgue and Reverence as the test plugin, but my point is to show that depending on YOUR particular workflow and plugin choice it doesn’t actually give the whoe picture.

There’s no point in me doing a test with plugins i don’t use as that will show soemthing else, like a lot of benchmarks do, you have to work out what works for YOU and your workflow and unfortunately that’s really trial and error and a lot of testing and reading etc.

I’ve been using Cubase since the Atari days and expect for a couple of years on MacOS (snow leaopard) I’ve always found cubase to work beter on windows. I’ve a feeling , and this isn’t going to be popular, Cubase now works better on mac OS with Apple Silicone!!!

M

[Norbury_Brook]

UPDATE:

well after @Tj99 mentioned a bug in ASIO guard in V 13.0.10 I realised the windows version of Cubase on the AMD machine wasn’t updated to the latest 13.0.2 like the macbook so I updated it this morning.

Wow!!!what a difference… the AMD now playsback the project with the same performance as the Macbook

I tested this time using both on the RME and both on the UR22-c and same results the windows version now runs very, very nicely.

I’m surprised that there was no real mention of such a perfomance boost with this update.

anyway , that will certainly please all the windows users out there who’ve read this myself included.

Onwards and upwards.

M

[Dr-Mad]

Thank you for taking the time to do all of this. Very useful information in this post discussions…

[Hippo]

@Norbury_Brook

I’m surprised that there was no real mention of such a perfomance boost with this update.

I`ll mention it, found the same here BUT its only got me back to where it was with CB12.

Hippo

[Norbury_Brook]

I bypassed C12 so have no metric to judge. I don’t know what happened with my system and C12 but i couldn’t get it to run properly. I built a new machine and did a fresh install when c13 came out and it worked better, however I did experience crashes that were down to windwos WDM midi. I then bought the MBP max and out of curiosity started to push it to see how it performed and found it to be way more than i expected, and I had zero issues with the GUI /midi etc so I decided to give it a go as my main machine. I always have to option to switch back to the AMD as my system now runs entirely from a sonet TB dock so one cable I can switch between windows and the MBP

Having the 7950x as a Ve pro /Audio gridder slave means I have an incredibly powerful system now that i doubt I’d ever reach the limits of.

M

[Tj99]

@Norbury_Brook

Wow!!!what a difference… the AMD now playsback the project with the same performance as the Macbook

Oh wow!
Thanks for testing again, as I wondered myself how such a big difference could be possible

[D.Mel]

thank you really for this i am now searching to upgrade and I was trying to find out the best way to do it with a PC cause Mac on the same specs is… Triple the price maybe more actually… So thank you really.