Updating Computer Hardware Performance

I am thinking about upgrading my memory, processor, and motherboard from a Z370/I7-8700K 32G ram to Z790//I7-13700K 32G (maybe 64G ram). The cores increase from 6 to 16 (8-8).
My overriding question is: what sort of performance increase might I see (I know that is very subjective). Is Cantabile able to scale over that number of cores and does it really increase performance that much?

I want to be able to run multiple large plugins at once (I do use states, etc) like Diva, IL-Vb3X, XV-5080, etc. I am looking to reduce timeload percentage without having to disable plugs.

Does all of this substantially increase the amount of data being sent to my audio interface which is USB2 based (MOTU 808X). I currently run the buffer at 128 without problems.

I’m guessing some of you have at least toyed with the idea of a higher performance system and I would be interested in your input.

As long as you don’t change the number of audio channels you send to your audio interface, the amount of data flowing between Cantabile and your audio interface won’t change - a stereo sum consumes a fixed amount of bandwidth, independent of how complicated it is to create that stereo sum…

The buffer size you can get away with is actually an issue of the audio performance of your system. Your system needs to have all its audio calculations for your sounds done before the current buffer runs out - the smaller the buffer, the faster your system needs to be. So your overall processing speed (for CPU-intensive plugins) and your hard disk and memory throughput (for sample-based plugins) is crucial for your ability to run small buffers.

Regarding number of cores, your mileage may vary - that really depends how many plugins you can actually distribute across cores - at some point, everything needs to come together to create one audio stream, so single-core performance will always play an important role, independent of the number of CPU cores.

See @brad’s explanation of Cantabile’s multi-core capabilities: https://www.cantabilesoftware.com/guides/understandingMultiCore

Cheers,

Torsten

1 Like

@Torsten, thanks for the link to Brad’s explanation. Sounds like things like memory speed, bus speed, SSDs and clock speed will provide more throughput than simply more cores.

I have a hard time imagining what kind of plugin combo you would run in a live situation that would require 8+ cores and 64GB of RAM. I guess if you pay no attention at all to shared racks and you have a really long set list with lots of layers per song part, but in that case load times would be extreme no matter your CPU/RAM. I have 32GB but that’s really mostly for the dual channel performance as I had a single 16GB chip before - I did notice a difference there.

Since your performance is already solid at 128 buffer size and you don’t mention any problems with loading your plugins, I don’t understand what you’re trying to achieve. What is “timeload percentage” and how do you mean “disabling plugs”?

For an example, my regular “basic” setlist contains
-Kontakt The Grandeur
-Kontakt Scarbee EP88
-Kontakt Vintage Keys wurli
-Kontakt Scarbee Jazz Bass fingered
-Kontakt Scarbee Jazz Bass slapped
-Kontakt Double Bass
-Kontakt Session Brass Pro
-Kontakt Tarilonte Musette Accordeon
-Sonivox String Ensemble (full)
-Reaktor Monark
-Korg Polysix
-Korg M1
-Blue3
-a bunch of fx

To this I will add synths as needed, but with modern SSD aided disk streaming I don’t think I’ve even hit 8GB of RAM, even though there are at least 9 large sample libraries loaded.

My laptop is getting long in the tooth and I could benefit mainly from a bit faster single core performance, so I understand your “itch”. Just don’t overdo it with things that are largely irrelevant to live playing. Having a quick SSD and keeping your system lean and well optimized are at least as (if not more) important. People run Cantabile on very light systems, you just have to know what you’re doing.

These are two basic concepts of Cantabile.
The manual explains this well.

Thanks, I looked time load up. I know what disabling a plug means. But the OP says he runs the buffer at 128 without problems. So I don’t really understand what the issue is. I’m sure the OP doesn’t simply want to reduce time load for laughs. But the system he is proposing is vastly overpowered for generic live playing to the point where it doesn’t really add anything, while simultaneously not addressing some key points like system optimization and load times.

Sorry, i guess I wasn’t clear on my question. My underlying question was if the proposed upgrade would be a lightning bolt from heaven or just a nudge in the overall ability to load multiple instance of CPU hogs like Diva, Keyscape, Omnisphere, etc and would that CPU load affect the load being placed on the audio interface.
I don’t want to spend the money if I get negligible return in performance. One thing that is very clear from comments here as well as in other forums is that single core performance is more important than cores. And like @Torsten was saying you can expect a mixed bag. I guess i was kind of assuming that Windows would assign each large plug to a core so all could run simultanously; doesn’t sound like that is necessarily the case.

@cpaolo are you talking about GlitchFree or something else? I can’t seem to find a “manual”.

Unfortunately, this is not how it works. There is only one audio buffer, so true parallel computing is not possible.

You will certainly see an increase in performance after a tuning of the processor. But don’t expect lightspeed.

Not GlitchFree, Cantabile online guide.

So then let me propose a second option; assuming the current processor handles the load, I would have 2 options:

  1. increase buffer size to 256 or
  2. get a faster audio interface. As mentioned, I have a MOTU 828x which is USB2 based. It has a Thunderbolt port, but support told me that in the end the processing runs at USB2 speed (so why have the TB port?). Anyway, in theory are there interfaces out there <$1000 that would perform better?

Not really.

Latency is made up of two parts.
The one due to the OS drivers and the interface type. I’s usually the least relevant, speaking of a few ms.

The other part is related to the ASIO audio buffer. This is the part that matters most, up to tens of ms on a slow PC.

So, try setting the buffer to 256, don’t rush to replace your sound card with a very expensive one.

Hope this help.

I will experiment in that direction. Thanks @cpaolo

Steve,

Here is my experience (not with latest Intel processor).

I replaced my old i7-4810MQ with an i7-10700H (not a K-series as it’s a laptop processor).
Not much difference, as you can see in our dedicated PC configurations comparison table (see the thread linked below).

Please note that a minimum level of tuning is mandatory. At the very least:

  • Set processor and all the cores to max performances.
  • Disable core-parking “feature”. There are utilities to do this easily.
  • The other GlitchFree hints.

USB 2 “speed”, i.e. bandwidth is not a bottleneck in your setup - unless you are sending an insane amount of stereo channels to your audio interface. USB 2 has a bandwidth of 480 Mbps, while a stereo signal at 48 kHz requires just 2.3 Mbps (ignore all the protocol overhead etc.). So if you send one or two stereo pairs to your interface, your USB 2 connection is mostly bored…

The more expensive interfaces typically have other differentiating features: better AD/DA converters, lower noise, integrated DSP capabilities etc, but communication speed is not the issue.

One thing that can contribute to better performance of an audio interface is the quality of the driver (the piece of software on your PC that handles the task of moving audio data to the interface) in combination with the specific hardware - there are some vendors that have a reputation of very efficient drivers, e.g. RME. That’s where investing in a “better” interface may pay off in lower latencies.

But the key of real-time audio processing is still the “computing and throughput capability” of your PC, i.e. its ability to make a lot of calculations and have data flowing through storage --> memory --> CPU --> memory --> output as quickly as possible.

The key challenge in real-time audio is managing the flow of data to your interface. The smaller the “chunks” of data are (buffer size), the more hectic the process of filling that buffer and handing it over to the audio interface becomes, and it puts more strain on the throughput of your system (not only CPU, but also the communication between the processes in your system - some processes need to wait for the output of others, which causes a delay). So this process of moving data to your interface becomes a bottleneck. Increasing the buffer size lets your system concentrate more on actually calculating the data (running plugins) and moving the data between plugins vs. just shoveling it out to the interface. The price you pay is more latency.

Essentially, no single solution will bring you a quantum leap in performance - a well-balanced, modern system with good drivers, using some of the typical real-time-audio tweaks can perform really well.

See the update that @cpaolo posted recently: Feedback computer config . My little “Live Cube” (ASRock Deskmini, first in the list) using a middle-of-the-road Core i5 processor with 6 cores compares pretty favorably - and it’s not a “high-performance monster” with 10 or more cores.

Cheers,

Torsten

2 Likes

Thank you all for you input. Last night I did some experimenting with adding more plugs and then checking the profiler to see who the big CPU hogs were. Surprise IK B3X 18%, next Roland XV-5080 13%. Everything else, including 4 instances of Diva, 1 Hive, 2 Omnisphere, 2 Keyscape, Fabfilter, Blackhole and others were 3% or below. This was with no notes being played; timeload varied between about 30-35%. Sample rate 48K, buffer 128. ( could drop sampling to 44K)
Tried different combinations of plugs playing a random amount of notes. Timeload varied from about 60-80%; it took about 5-6 plugs to start getting dropouts. CPU never got higher than about 20%. That’s good enough for me.

The thing that I was missing was the meaning of timeload. I knew it was not a measure of CPU directly, but was nevertheless not seeing it as separate from CPU. Somebody either said or I read that it is a percentage of the available audio cycle. When I realized that it all started to click. If needed I can bump the buffer to 192 or 256 and still not be too affected by latency.

Thanks @Torsten for explaining how the audio interface bandwidth is affected. Since I have both the computer and audio interface in one rack, if I do replace the AI at some point, I would like it to be a rack mounted version. i will have to look at RME; that seems to be a favorite here.

Thanks to all who contributed to my increased understanding of the process.

I have a hard time imagining what kind of plugin combo you would run in a live situation that would require 8+ cores and 64GB of RAM. I guess if you pay no attention at all to shared racks and you have a really long set list with lots of layers per song part, but in that case load times would be extreme no matter your CPU/RAM. I have 32GB but that’s really mostly for the dual channel performance as I had a single 16GB chip before - I did notice a difference there.

@sanderxpander I thought so, too, but I am seeing a huge difference in headroom since I moved to a 4-core i5 w/32GB for guitar and bass parts, and a 6-core i5 w/32GB for my synths. Granted, I’m playing as a one-man-band, so there are a lot of tracks running at once. But I used to be running at 90-95% in many songs, even after thinning parts and trading off to simpler VSTs on complex parts. It also allowed me to easily use two Focusrite audio interfaces via ASIO, without having to play tricks, so I have 10 hardware audio outs – that might actually be the bigger gain.

What I can’t speak to is the use of modern performance vs. economy cores. Each one of my cores has its own FPU, and none of them can run multi-threading. I don’t know if 12 mixed-bag cores in a 12th/13th gen Intel trumps 10 ‘real’ cores in my setup. I do know that running as much of the VST and media file load from a RAMdisk improves the overall loading of new parts!

Regards,
-BW

A six core i5 with 32GB for a synth setup is something different than a 16 core i7 with 64GB which the OP was originally proposing and I was commenting on.

I think you probably have more RAM than you need but it doesn’t hurt. Like you said, modern SSD disk streaming makes a big difference.

Understood. But to be clear: I have two i5 micro PCs running Cantabile simultaneously in my setup, and each one has 32GB of DRAM; it’s not a case of a ‘studio system’ and a ‘live system.’ I use shared racks just about everywhere, but the one PC still has to load 29 of them, and the other 35. So, my system isn’t all that different from the proposal – albeit from different CPU generations (my 4-core is a G7, and the 6-core is a G8).

And although I haven’t timed the difference recently, I compared loading the VSTs and .wav files I use into SSD memory vs. RAMdrive memory on power-up before I added the second PC. The startup load time for 68 VSTs from a WD Blue 570 NVMe SSD (G3) was about 90-100 sec.; the load time for the same VSTs from RAMdrive was about 50-55 sec. – a significant difference, and probably more than I could expect by moving to a G11 or G12 CPU. Loading sample-based VSTs is one of the biggest areas of gain. So, while SSD memory (especially NVMe) is a big improvement, running programs from RAM is so much faster.

I used to design laptop PCs for a couple of top tier manufacturers, and we did an experiment with running the entire OS + apps + data from RAM at one of them. It was completely impractical in 1998, but the results were stunning! That Pentium processor with 1GB (??) of dynamic RAM – far beyond practical at the time – would probably keep up with many G5-G7 laptops for tasks like word processing and browsing.

Regards,
-BW

That’s interesting about the RAM disk. It seems a rather niche use case though. Doesn’t it require constant power for it to work?

I understand your point and I don’t think we really disagree too much. It’s nice to have a somewhat up to date system and t9 overspec a little bit. But if you check your RAM usage, without RAM disk, let me know how much your entire setup uses. I’d be pretty surprised if you came close to 16GB. I added 16GB to my 7th gen 4 core i7 last year and noticed a difference, although considering my RAM usage, I think it was more the dual channel effect as I had a single chip before.

But, a 4 core gen 7 i5, even with 32GB RAM, is different from a 13th gen 8(+) core i7 with 64GB.

To be honest, I don’t really think you need to run two systems unless you’re playing keys and guitar/bass at the same time, which seems unlikely. From a performance angle at least, I can absolutely see where it makes sense from an organizational viewpoint.

Torsten and others advice is spot on.If you do decide to look for a rack mounted interface, FWIW I’ve been gigging with a PreSonus Quantum 2626 for a couple years now, it’s rack mounted and TB3, 8 in/8 out And I think it goes for about $600 now. It’s been quite solid, and I did get a significant latency improvement, IIRC I’m running at 64 and about 5ms latency, And that’s on heavy songs with about five backing tracks and some heavy plugins… It’s a good driver and they update it regularly. I believe they also have a USB version now. But as others said, you don’t need to break the bank for live work… For several years before that I used a cheap Scarlett 2i4 and a Vaio i7 laptop/ 8GB RAM from 2009… until the 14+ ms latency started to drive me crazy😁. The Quantum is paired with a i7-9700K Rack mounted PC, 32GB Ram, 2 SSD. It barely breaks a sweat… A 60 song setlist with very heavy songs and backing tracks runs solid every time for the past 2+ years, I don’t remember the time load or CPU, maybe 15%, but I think my largest spike on CPU is around 45% only during startup. I don’t even bother to look anymore. Best of luck!
Tom

But that’s the point. I am playing guitar and keys at the same time – or at least, me and Cakewalk are. If I’m soloing on guitar, there are probably 1-2 synths being played on MIDI, plus maybe a rhythm guitar part. Similar situation if I’m on piano, organ, or synth. My motif is to sound like a full band, and that requires at least 4 instruments running at any time.

And I have run the numbers, and loading VSTs and their associated sample sets from a RAM drive always runs faster than from SSD in my system. And yes, it prevents the system from going into standby, but I prevent most of my setup from heavy power management, to prevent glitches on stage.

Regards,
-BW