Converting typical benchmarks to Cantabile use

Cantabile
1

I’ve got my system running well, but I can clearly see that a bit more horsepower wouldn’t be a bad thing. I’ve already had to retire a couple of VSTs that ‘don’t play well with other children.’ One in particular, U-he’s REPRO-5, only runs well if you give it 4 or more cores, or reduce polyphony to 2-3 voices. (Since I only have 4 cores, that’s not going to happen!) Right now, I have an i5-7600T with NVMe SSD and a lot of RAM. Moving to a 10th or 11th Gen system isn’t appealing now because of the cost, but someday that will change.

The problem is, all of the available benchmarks these days are tailored to games and video processing. I know that multithreading helps some apps, but not Cantabile, so I would only upgrade to get more cores. But, newer CPUs seem to have ‘performance’ and ‘economy’ cores (or some similar names), which I assume are not equal to each other. Looking at the benchmarks for an i5-8500T, which has 6 cores to my 4, some benchmarks improve, others actually drop a little – probably due to lower clock frequency per core. I’d hate to spend money and time on a new PC and not get anything tangible for my efforts.

If CPU A is say, 20% faster than CPU B in Cinebench or Passmark, what does that mean for music production?

Thoughts?

Regards,
-BW

2

I’m in the process of upgrading my studio DAW to a new machine and went with the i9 13900KF. I’ve had both Intel and AMD over the years so not really biased towards one brand or another. This particular choice was based on stronger single core performance over the competing AMD by a noticeable margin plus somewhat lower cost, and because single core is still relevant in music software. Plus it has I think 24 total cores and each of the economy cores have more performance than ones in the six core workstation that is being retired. Not sure if those cores are physically different or its just a case that the economy cores are just the ones that get throttled to govern heat. But supposedly you don’t have to assign processes to a particular physical core to get max performance. Fortunately many if not most DAWs can leverage multi core. Plus for keys players, the Unify product by Plugin Guru does a fantastic job of leveraging multiple cores and runs as either a stand-alone or a plugin. Plus they are releasing a version with audio inputs that will make it interesting to us guitar players. All to say, for me I went with max single core performance while also going big on multicore raw power. Fortunately this should cover all scenarios: the single thread hog synth and the more eloquent multi thread synths and hosts being able to spread piles of plugins across cores. All to say the Cinebench and Passmark have some relevance to music production IMO, since usually the results specify both single core and multicore tests. LOL, I suppose I’ll find out this week first hand, since those tests plus recommendations from others are what fueled my choices.

3

@sekim Thanks for the input. A 13th-gen i9 is probably out of the question for me right now, for cost reasons, but your rationale should work well generally when choosing platforms. I agree that single-core performance means something; hence, my comment about the i5-7600T vs i5-8500T: two more cores, but each core runs at 2.1MHz vs. 2.5MHz, so I’m not sure how much overall gain is really there. (The single-core benchmarks actually go down.) I’ll be interested in any follow-on info you can provide.

Regards,
-BW

4

I should have this new box up late this week. Still waiting on the power supply from EVGA that apparently isn’t in the States just yet…

5

Ciao,
usually I go to https://www.cpubenchmark.net/ and compare the “Single Thread” values.
This because unless you are 200% sure that your plugins can manage parallel processing in an efficient way, the single thread values are a safe bet IMHO

Luigi

1 Like
6

Was able to get first light from the new build this morning. Installed OS and ran benchmarks. Several interesting things uncovered, none really unexpected I suppose:

  1. I forgot to set the profile in the UEFI for the high performance memory and as a result got to compare CPU Mark with and without this set properly. Very significant difference.

  2. I also didn’t read the manual and put the memory sticks in slots 1a and 1b. Even with the profile set correctly it still lagged others on PassMark. So once I read the manual and placed them in 1b and 2b like it says, it now tests about 30-40% better than the mean score for these sticks (which is 99% better than the universe).

  3. Also forgot to set the Hi Performance fan/water pump profile in the UEFI. Once I did, one final thing that came out during test is the marginal cpu cooling with this dual fan liquid cooling solution even with the right fan/pump profile.

While running PassMark I also ran Core Temp to monitor temps. One of the CPU Mark tests momentarily took one of the cores to 100 degrees which is the thermal limit and invokes throttling. The other tests were pretty normal temp wise. So a good quality dual fan liquid cooler is the smallest you want for a 253 watt dissipation cpu for DAW imo…and probably worth it to spend for a three fan model regardless.

The final results after fixing my stupidity are this machine runs at median or up to 50% better than median than its peers in PassMark. So now that the PC is shaken out, its time for the fun task of turning it into a DAW and installing software, migrating licenses, and all that sort of stuff…
Then we’ll see how DAW performance correlates to these massive PassMark scores. Single thread score was in the very top, so fingers crossed.

7

@luigi I’m learning more and more about the way CPU makers in general, and Intel in particular, handle multiple cores. I think they are playing games to keep the single-core benchmarks impressive.

If we compare the i5-7600T I own to the next logical step, an i7-8700T:

Process: 14nm vs 14nm
TDP: 35W vs 35W
Cores: 4 vs 6
Threads: 4 vs 12 (not apropos in music software)
Base clock: 2.8MHz vs 2.4MHz
Turbo: 3.6MHz vs 4.0MHz

The problem comes because more cores mean more heat. The process didn’t shrink, so 6 cores is ~1.5x the heat. Intel fixes this by slowing the base speed, then bumping the turbo speed, but the trick is deceptive: Turbo only applies when 1-2 cores are running. This means that a loaded Cantabile song would be using 1.5x the cores, but at 85% speed. And a song that uses exactly 4 cores would probably run slower! So, I’m not sure that one can trust the single-core benchmarks anymore – they usually test the fastest core in the CPU, and only that core. The assumption is that 95% of users care about gaming and video-rendering tasks (multicore tests); and the other 4.99% care about desktop/home use (single core). We digital musicians are left in the 0.01% group. :roll_eyes:

8

Yes, you are right, and this is the very issue that affects benchmarks in general… AFAIK all benchmarks (even the most sophisticated ones) are good indications of a specific workload in specific conditions, that are never the same you experience in real life scenarios. I see it everyday when I compare the computing power I get from AWS or Azure machines wrt the one I would expect.
Most published benchmarks were created to mimic the “average user” scenario, and this community is totally outside that average (and that’s good)

So probably the only real answer is: try it with your workload, with your specific Cantabile setup. And of course you can’t do it, because you cannot buy 4 systems and pick the best one… (if you can, give me the 3 remaining ones :grinning: )
My suggestion would be: check with friends or with a friendly local dealer if you can run some tests “for free” and then pick what seems the best one - good luck

9

I monitored heat and load when I benchmarked and what happened appeared to be that cores alternated going into turbo in order to stay within the 253 watt thermal profile, but the ones not in turbo were still fully loaded at non-turbo max freq. So its basically a selective overclocking of cores within thermal headroom of the package. And a few of the cores did appear to go turbo while all of the others were at full non-turbo. That might not be the case with some older cpu’s from Intel with miniscule thermal budget of 35 watts (laptop and eco-grade desktop/server) LOL, the Xeon W3680 I’m retiring is setup in BIOS to send all six cores into OCed turbo simultaneously and it never gets above 55 C with liquid cooling.

10

@sekim That’s an amazing system! And so far from my personal needs that I probably won’t ever own one, even if I win a lottery tomorrow! I play outdoor gigs in the summer, and while my rack was designed with thermals in mind, I don’t trust a CPU that uses more than about 40W in that scenario – I could end up with a PC that keeps locking up during sets. I also don’t want to be lugging around the weight (and complexity) of a water-cooled system to gigs. But your results are still good data to have: it shows where the state of the art is heading. Now, I just need to wait for the ‘state of the technology’ to catch up – which always happens, sooner or later.

It’s unfortunate that as PCs get faster, they also target video needs more. For a Cantabile PC, I’d love to have 8-10 cores running at high speed, coupled to a poor-performing video section. (I run my Cantabile PC headless, connected to a laptop that also runs Cakewalk.) All of the silicon dedicated to graphics is a complete waste for me!

Regards,
-BW

11

To my surprise the i9 13900K comes with onboard video. I ordered the i9 13900KF variant to not get the onboard video. I know a lot of the Intel range has onboard video which is perfectly fine for DAW use. So maybe a micro ATX with a midrange desktop cpu might be worth a look since its not that much bigger than a NUC and can sit performance wise between a good NUC and a decent power desktop. Also FWIW I was able to do this newbuild for right under $2K.

[edit] Also, I think some of the midrange cpu are available in low power variants at about 60-75 watt TPD. But they are packaged in the same physical package as the full power variant so you end up with all kinds of thermal headroom using the “normal” cooling solutions either air or liquid.

12

Even 65-70W TDP is rough for outdoor use, IMO, especially since I’m using a ‘micro’ PC. At around ~22C of a typical indoor gig, cooling 65W with a 10cm heatsink fan is reasonable, if occasionally noisy on stage. Bring the ambient to 30C, though – which often happens around here in the summer, even 35C or more on asphalt parking lots – and the fan starts running at 90-100%. I’m concerned about some kind of glitch or lock-up in the middle of a set. So, 35W TDP is my maximum; 28W would be even better.

I just found a good deal on a reconditioned i7-8700T micro, and decided to take the risk that it would be ‘better’ than my current i5-7600T. Since I can build a 2nd Cantabile PC with similar specs – 35W TDP, Focusrite USB audio, NVMe SSD, 32GB RAM, Windows 10 Pro – it will be interesting to see if the 8th Gen PC buys me anything tangible. More cores, more L3 cache, higher Turbo speeds, but in the end it comes down to: 1) will it run more VSTs at the same ASIO latency; 2) will it run a couple of VSTs together that currently only run as solo entities; and 3) will it run at around the same temperature? I’ll keep the group posted on my progress.

[FWIW, the published single-core benchmarks are 5-10% lower on the '8700T, and the multi-core benchmarks are 25-75% higher. But we’ve kind of agreed that those benchmarks are meaningless for music production.]

Regards,
-BW

13

Another i5-7600T CPU here. I enjoy the low current drain (< 22-25W system typical), and low noise, but not always the performance.

My main dilemma with a lower power, lower clock CPU has been latency vs instances of Kontakt racks and other VSTs that will run before CPU clicks and issues start to raise their head. For piano of course the lower the latency the better, but this limits how much else can run.

Recent single-system use of Audiogridder has been encouraging: I can run with low buffers and low latency for the piano direct in Cantabile, but load some other libraries using the Audiogridder VST client within Cantabile. Doing this allows you to select a different buffer setting albeit with higher latency - OK for some libraries and ideal for reverb plugins. Audiogridder allows each client instance to run with different buffer settings (as an option), so essentially you end up with a system that can run VSTs “hot and direct” through to slow but frugal in terms of CPU load if that makes sense. I also read and believe Audiogridder does a good job of using additional cores. For clarity, I’m running both Audiogridder Server and Client on the same PC.

Anyone else with more in-depth experience of this?

14

So the new i9 13900K is burned in, optimized, tested and transformed into a DAW.

Ran some audio tests and the results are interesting to say the least.

A known offender synth in Cantabile still hits a 25% time load while the system cpu is under, get this, 1%. LOL. This with 4ms latency at 44KHz, playing eight note chord.

Then I loaded 16 instances of the offender, played eight note chord on all 16 simultaneously, and time load was unchanged and system cpu rose to 1.7%, LOL. And of course the output audio meter was pegged… The interesting thing is while I was adding them the time load without playing notes kept going up. But then play notes and it rose to the same value as with only one instance.

So not sure what is going on with the Cantabile time load, but apparently whatever it may be for a particular synth, you can add as many as you want and it isn’t going to increase in this scenario probably until 24 since that is the number of cores.

In Reaper, a project that took my old box to its knees and pegged both real time and system cpu (before freezing tracks), the project under the i9 put Reaper’s real time cpu below 10% and system cpu under 3%. So looks like extremely generous amounts of headroom are available now and probably will never need to freeze a track again no matter how complex the project.

Based on the above preliminary findings I’d say the general benchmarks were loosely predictive of what to expect for audio. About the only head scratcher is the Cantabile time load behavior. But clearly its taking full advantage of all the cores.

1 Like
15

Really appreciate the comments in this thread. This is anecdotal but when deploying new business laptops and desktops I noticed that setup time has dropped to about 1/3 what it was taking a few years ago. These are machines where we’re installing SQL Server for industry specific applications plus the usual MS Office, remote connectivity apps and configuring browsers for applications like Salesforce. Installations are CPU intensive as a lot of decompression and configuration occurs.

16

Hi All,

[I’m not sure if this should become a new thread. Opinions welcome!]

OK, so I bought another Dell micro-PC, this time an 8th-gen version with an i7. I wasn’t sure if the i7 with multithreading was better than an i5 without, but the (retired) engineer in me was curious. And the refurb that I found was cheaper than some i5 offerings. So, I set out to test a) G7 Kaby Lake vs. G8 Coffee Lake CPUs; and b) whether multithreading helps a Cantabile PC. 2.8GHz veteran vs 2.1GHz youth, so to speak. Except, it turned out to be way different than I expected …

Both PCs have 32GB of RAM, and use 12GB for a RAM drive to load samples and VSTs. Both are running ‘headless’ using Remote Desktop for video/kb/mouse, and rtpMIDI to pass MIDI data back and forth. They aren’t identically set up, but close, particularly when driven with prerecorded MIDI from a DAW.

I tested a set of 4 songs that I ended up having to simplify a little in my current setup, because my i5 would glitch now and then. Would the original versions work on the new machine? Nope. In fact, they were worse – a lot worse in some cases! I started looking for configuration mistakes, or some unneeded app that was hogging the CPU. That’s when it dawned on me: my ‘i5-7600T’ is really an i5-7600: 65W (vs 35W) TDP, 3.5GHz (vs. 2.8GHz) clock. There was a note in the ad saying that, “We may substitute items of greater value in some circumstances.” So, I paid for an OptiPlex 5050 with an i5-7600T, and got an OptiPlex 5070 with an i5-7600. :roll_eyes:

So, how does a slower, cooler 8th-gen T-class i7 processor made for laptops stack up against a midrange 7th-gen i5? Not so well, I’m afraid. The i7 has twice the L3 cache and 11% faster RAM, but it still runs at 2.1GHz, and has problems running songs that have evolved into ‘typical’ for my rig. The Cantabile Time Load was particularly disappointing, running near-100% in the best case, and well over 300% in some songs. Surprisingly, the reported CPU load was low – 20-30%, on average – but the audible results were glitchy. I wasn’t able to get things to settle down without resorting to 20-25 mSec of latency in the ASIO drivers. Since both CPUs are locked, I can’t play with either clock to try to level the field.

So, for a system filled with VA synths like mine, it seems that CPU (FPU?) speed is the dominant factor. I tried using the new system for sampled guitar, but even there complicated patches were showing weakness. The setup worked – as long as I didn’t change anything mid-song.

I’ve ordered an i5-8600 with 6 cores and a 3.1GHz clock to upgrade the i7; we’ll see how that fares. At that point, the test will be 6 cores at 3.1GHz vs 4 cores at 3.5GHz. The 8th-gen CPU should win, unless it runs hot enough to start throttling.

Regards,
-BW

17

Really interesting write-up - thanks for sharing. It sounds like your i5-7600 was a good choice back in its day (even if you ordered a lower-powered T variant).

Good luck with the i5-8600 - it will be very interesting to learn how it performs for you…

18

@TonyG My '7600 reliably runs:

  • 1 guitar (+ AmpliTube) and 2-4 lower-CPU synths, plus a media player;
  • 2 guitars + 2-3 synths + media player;
  • 3 guitars + 1-2 synths.

This covers 90-95% of my must-have needs. But there are always like-to-haves, right? What it doesn’t always do well is tricks like playing the same part on two similar synths, panned left and right. And Cherry Audio’s GX-80 does some magical things, but not always alongside other VSTs. I’m hoping that the '8600 improves the situation a good bit. My backup plan is to use both PCs, one for synths, the other for guitar + media + usb hub. It’s still smaller, lighter, and less costly than the six keyboards gigged with in the '80s! :wink:

Regards,
-BW

19

Thanks for the breakdown on your '7600 capabilities. It sounds more powerful than my low-power T version.

It will be a fascinating update to see what your '8600 delivers. As you say, even 2x boxes is considerably smaller and more powerful than the reality of hardware only (which I still love, but we get to keep what we like these days and have a mix of the best).

20

In the FWIW department, I researched the latest in mini-ITX form factor and it is possible to do a 240mm liquid cooler and i9 in a 13" x 9" x 6" case these days. TBH if I were going to build one as a performance box I’d look at the i5-13600K since TDP is a lot lower than the i9 and i7 series and it still packs a very good single core punch. It has six performance cores and eight economy cores. Plenty of thermal headroom with a good quality 240mm cooler and the 181 watt max / 125W normal TDP. Plus it can be clock limited and undervolted in the UEFI setup if even more thermal headroom is needed. Or just perma-park some of the cores.

1 Like