Anyone want to test different rigs/CPU's? EDITED

Steinberg Lounge
1

Hello,

I saved a simple project here:

http://deeaa.pp.fi/TEST/TEST.zip

You can download it and test export to 44.1/16bit wav if you like - it’s built with only basic effects so it SHOULD load in LE/Artists versions as well as 9.5 Pro. Only about 200 megs, so it won’t take but 5 minutes to test it on your rig.

That way we could get some proper real-life test going to see what kind of CPU core count and type is the fastest in basic audio exports in Cubase.

Here’s a list of some results gathered - so far to my surprise Ryzen leads by a good margin:

81.8 sec i7-6700K @3.2Ghz
43 sec - i7 3770K (4096 samples)
35.2 sec i5-7600K 3.80 GHz Buffer 2048
32 sec – i7-3770K @4.5GHz (2048 buffer)
29.72s sec - Dual Xeon (E5645) 2,4 GHz (12c), Mac Pro 5.1
28.7 sec i7-6700K @4.6GHz
27.3 sec – i7-2600 @3.4Ghz
25 sec – Mac Pro Touchbar 2017
23.6 sec 2,7 GHz Intel Core i5 8 Gt 1867 MHz DDR3, Macbook Pro 2015
23 sec – i7-6700K @4.6Ghz
13.19 sec – Ryzen 7-1700 @3.0Ghz
9.98 sec i7-7700K @4.2Ghz (4c)
8.3 sec i7-8700K @5GHz (6c)
8.12 sec i7-6840X @4.2Ghz (6c)

If you do export it, time it with a stopwatch from the moment you hit export to the moment it hits 100% (rather that than finished screen) because it’s easy to be accurate at 100% since you can follow the percentage grow.

My i7-6700K @ 4.6GHz did it in 31.8 seconds at first try.

I also tried it at native 4.0GHz and got 32.5 sec, so it’s interesting already in that 15% increase in base clock only gave 2% increase in exporting speed.
I also ran it without Hyperthreading, and exactly same time was spent. Later I changed to 32 bit audioengine and killed all other processes and the best I got was 23 seconds.

I would guess in a larger/heavier project a higher clock would yield a better increase in overall time used, but still interesting.

I’d LOVE to hear how fast can, say an older 16 core or so Xeon crunch it, and also newer ones like Ryzen or Threadripper builds.

Anyone game?

2

Addition: my machine seems to halt for a moment when I press Export. If I start the watch when I see 0% and it starts exporting, I get a bit over 18 seconds.

3

If I start normally, but with 32bit audio engine instead of 64, I get 23 seconds.

4

25 seconds on my MBP 13" 2017 touchbar.

5

Seems to be quite in line. Surprisingly small differences so far. I have this on another forum as well and first report from there says 27 seconds (i7-2600 @ 4.2ghz), 16GB ddr3 1333Mh. That’s a tough number for a seven-year old PC. Mine is only 3 years old and still no faster in reality. Anyhow, it’s now settled at 23 seconds over multiple tries.

Clearly using a 64bit audio engine makes the time longer. I’m checking if there are other tricks to speed it up some.

6

Does this test do anything that DAWbench so far has not done?

7

Probably not, but we wanted real life results from different rigs easily.

8

If there is a chart with different CPUs at different speeds available somewhere I’d like to see it though. Can’t find one anyhwere. I mean, same project ran with different CPUs and speeds and listed in order. Is there such a thing somewhere, please post if there is.

9

The thing is that with testing one should generally want to isolate different parts of the workstation. So it’s valuable to take one type of workload and create a test project out of that and then test to figure out how one type of parameter varies (such as plugins and CPU capacity).

The other thing is that by sticking to a standard test you’ll end up with more users over time performing the very same test, which in turn gives you far more results to compare. Starting new comparison projects will mean that you’ll have to get more users to run those tests than already existing projects. Unless your test is better or equally good but more used you’ll likely learn more from just using existing tests.

As for DAWbench specifically, here are two charts with testing done by Scan computers in the UK:

7960X-DB-Classic-1.jpg
7960X-DB-Classic-1.jpg1871×967 239 KB

The above test shows the amount of relatively heavy-load multiband compressors you can run before audio starts crackling during playback. So it stresses the CPU mainly, and shows what the CPUs are capable of in terms of just number crunching.
7960X-DB6.jpg
7960X-DB6.jpg1812×939 185 KB

This test shows how many voices can be played back using Virtual Instruments. It’s a different test because the implication is that the lower your latency needs to be the sooner you run out of voices when playing “live” virtual instruments.

In other words I would say that the “regular” or “classic” test shows you more about a mixing scenario where you can have high playback buffers (high latency) because it doesn’t matter, and the second a scenario more for those playing / recording in realtime using VSTi with lower buffers.



It’s from here…

10

Yes, those are indeed interesting, but in my view rather academic and not real life results for anything usable.

Any daw PC can run enough instances of fx these days required I think, and what we’re interested in is actual performance increase over different platforms and CPUs over a spectrum of choice.

I have never come across a situation where my DAW would stutter under load, be it 100 tracks and as many or more VSt effects etc, I’m not interested in that. BUT I am very interested in what kind of gain might I expect upgrading my machine to, say the latest Intel hexacore and whether that would be justifiable expense.

And so far while we only have half a dozen tests run, it would appear that there is very little effect what CPU and system you run within some boundaries.

If a 7 year old i7 can get equal or even better results than a 3 year old clearly faster i7 or a new PC, that is interesting to me. That’s something that the CPU crunching charts don’t show.

The only place where I want more speed is exports. I do up to a dozen exports a day, and it makes a huge difference whether that takes me 3 minutes a pop or 2 minutes a pop.

But so far it seems in that respect it doesn’t matter at all whether you have an old 2nd gen i7 and a HDD and slow memory or an 8th gen at about same clock speed and an m2 drive and plenty of memory three times the speed - every CPU chart claims there is a big difference. Also the clock speed is interesting. Getting 15% more clock speed gives only 2% increase to actual output speed.

There’s a lot of data out there but we feel it is skewed and shows huge differences in percent - but who needs 500 compressor instances? It’s all about export speed.

We want to know whether it makes any sense to upgrade even from a decade old quad-core to a new CPU in real life or not. Or whether it is faster to export on an old Xeon architecture but with 32 cores or a new hexacore at five times the cost.

That I can’t see in any charts, they all just talk about calculation power and not entire systems and architectures as wholes and in real life scenarios that would actually matter to the user.

11

If any PC can run enough then why do you need to test anything? “Enough” literally means that you have no need for any more beyond that value. So, you already have what you need. No need to test anything.

But if you are going to test something then “actual performance increase” is exactly what the test shows you. You call it “academic” but that’s missing the point. The test will isolate the performance of a CPU relative to another CPU. That’s what you say you want. You say you want to see increase and that’s what you can see in those charts.

Same as above. If you have no problems today then you have zero ways of justifying your expense. You already have enough.

And same as above (again): The charts show you “what kind of gain” you might expect.

Yes they do. You have to understand that what that company does is sell computers set up for professional audio / video work. They deal primarily with customers who are looking to buy a new computer with newer components. This means that their tests will show the past generation or two of CPUs, not CPUs that are 7 years old.

But that doesn’t mean they haven’t done tests on those older CPUs, it just means that you have to go find those test results. A good place to start is Gearslutz.com where you can find threads on a bunch of older CPUs.

CPU speed then. Look at the charts. And btw; those charts and charts just like it for video show exactly what you know for professionals who aren’t looking to save 12 minutes in a day, but more like hours worth of export/rendering. So the resource and testing methods are pretty legit.

You make it sound like “it’s all about export speed” for everyone. It isn’t. Really the only group of people I’ve encountered until today that care about export/rendering speed are video professionals, and that’s just because it’s not a 3 minute render to get even a TV show out, it’s far, far longer than that, and under really tight deadlines. But for audio all I’ve seen people complain about so far is running out of computational power and/or glitches.

At any rate, if you want to look into what your bottleneck is, since you’re saying it isn’t the CPU, the RAM or storage, I would suggest looking into:

  • Buffer settings during export (as far as I know larger buffers = faster export)
  • Plugin choices (I know Equilibrium for example slowed renders on PT by a huge amount)
  • Plugin settings (i.e.; could some have settings that slow down the export for no good reason?)
12

Ok, I now tested your project with my brand new computer setup!

Ryzen 7 1700 at stock speed (8 cores @ 3GHz). Profile in Windows is “Balanced”.
16GB DDR4 @ 3200MHz
Source/Project drive is SSD over SATA, and is also the target drive

Export time: 13.19 seconds.

13

Fun.

i7-3770k overclocked to 4.5ghz
192 sample buffer: 57 seconds
2048 sample buffer 36 seconds

sata SSD, project drive and destination the same.

Now, I feel like I need a new system. :laughing:

14

I hear you, but it really seems to me everybody I know - people like me, who churn out maybe a few CD’s or a dozen a year or a bunch of videos as a hobby - we all are interested in export times. And we all make music and videos cooperatively with each person in even different countries.

I usually have like an hour to work on a song at a time, say after work and before getting the kids to their hobbies. MOST of that time always goes into opening projects, making very minor changes, exporting a bunch of songs in the works, and getting them online so I can listen to and make notes at work and driving etc. and the other participants can do the same.

We only have maybe 50-track and 100-VST plugin effects; my ongoing CD release is about just that, plus two drum machines (SSD and GrooveAgent) and a few synths, lots of Amplitube and GTR3 plugin tracks…that takes about 3 minutes to export. It’s really no problem to run them smooth, at worst we just need to turn buffers sky high at some point. But the video of it takes about an hour to export. but that’s a different story. I already know I’d need lots of CUDA cores ie a better display card, but for now my main concern is audio.

This whole project started from us noticing that there seems to be next to no difference in export speeds between our machines despite huge differences and what those CPU charts etc. say so we wanted to embark on a test to find out whether there are some types of machines that would be significantly faster.

So far tests have revealed mostly this:

  • Hyperthreading on/off doesn’t affect anything
  • Clock speed affects export speeds at hugely reduced gain rate (ie.15% more clock = 2% more speed)
  • Hard drive and memory speed don’t matter at all in this (an old HD spinner machine with 1600MHz RAM beat an M.2 NVME machine with 3200MHz RAM
  • Soundcard buffer settings and ASIO guard etc. settings have no effect at all in some systems, in some they do - depends on the card?
  • using 64bit engine over 32bit engine adds roughly 15-20% time ie.slows down exports some

And now it appears Ryzen builds are very quick at exports, but we’ll see as we get more results in.

15

Thank you! I’ve been wanting to hear how Ryzens work. I wish we can get more Ryzen users to try it, or even a Threadripper one. That is one sweet render time already!

16

Edited the OP to show results and I can keep updating it for easy access.

17

There has to be something else hindering it now. That CPU should be at least equal to my 6700K and way better than the 2600K so you definitely should expect to get clearly under 30 seconds on it.

Does it ‘think’ before it starts the process (bar moves)?

On my machine when I press export it does nothing for about 5-6 seconds and then the progress bar appears, and if I clock it from that rather than pressing export, it only takes about 18 seconds.

So clearly if I can figure out what does it do in that time (something not related to CPU as CPU usage spikes only after it has already exported some 5% of it)…I could shave off the time quite nicely if I could somehow eliminate that time gap. Although I doubt it makes a big difference in bigger projects.

18

In fact there is a 10 second delay before starting. I think it has to do with some routine involved in accessing drives, I’ve seen it elsewhere, but haven’t troubleshooted it…

19

Seems so indeed. But for some users it doesn’t do that. Maybe it depends on what kind of drives etc. there are in the machine.

20

Accessing drives likely isn’t the issue, and neither is drive type - UNLESS you’re using a spinning drive that’s actually not spinning when you hit ‘export’. Then you’d have to wait for the drive to start rotating etc. But since you “deeaa” said your HDD gave better results than your SSD on this test it doesn’t seem like that’d be the case.

What I would maybe look at is what else is going on on that computer. My computer is as I mentioned a new build, so I’m running the OS off of a brand new 512GB SSD (SATA, not NVME), with a brand new install of Win 10 x64. The only software I’ve installed so far is Nuendo, Spotify, Voicemeeter, one game and a few supporting items like the Lynx drivers etc. So this is a very “slender” build right now, and brand new. This was also done after patching for the Meltdown virus by updating Win 10 with the latest security patches.

So, if you’re looking at a 5-10 second wait before export even begins then apart from a HD spinning up the only thing I can think of is if the computer is doing something behind the scenes that takes time because of running processes and background services. Those things certainly add up over time (i.e. months/years) as people install and change applications.

My thinking then is for you to boot the computer, run the test export, quit Nuendo, open task manager and start shutting down processes and services that you think aren’t needed, open Nuendo again, and export again. See if there’s a difference at all. If there is, I suppose you could be looking at processes or services that compete for attention…