DAWbench

[3CPU]

AMD Teases 5nm Ryzen 7000 ‘Raphael’ Zen 4 CPUs, Unveils Ryzen 7 5800X3D with 96MB of L3 Cache

Firing back at Alder Lake

www.tomshardware.com

Thanks Pictus, for sharing that article.

A lot of power layered on one single die, silicon can only reduce some of the heat, I guess there may not be much headroom for this multi-layered beast. Should find out more about this soon after March 2022, once several tests and reviews are completed.

 

[parapentep70]

Right, but what's the *realized* practical impact of that fact?

EDIT: and to be clear, I'm not saying there isn't one. I'm saying I've never seen one.

True: I've never run a project with 300 compressors and nothing else. This is not realistic. But I have run multiple projects with 20...50 tracks each with a Kontakt instance or a synth VST plus several (normally few) effects plus the master FX chain plus bus FX... all without disconnecting the network to check for the latest plug-in deal on the Internet.

My experience says that such projects are comparable to xxx compressors. Dawbench is useful to anticipate that my older computer could not run such projects without freezing track groups, and it also told me how large is my headroom running in the new machine. So I made the decission to upgrade at certian time. I run Dawbench in the new machine and also in the old machine. My results with a good PCI card never listed (EMU1820) were surprisingly close. The CPU and memory headroom as predicted by Dawbench was about 3x better. And indeed my critical project went down to 30% from some +90% (with ocassional drops).

Nobody (but me) is going to run my specific projects, so Dawbench takes a free compressor (and also a typical VSTi, I think it is Kontakt!) as an example of thread loading, then "calibration factor" is of course different for each project, but for me is a ton more than nothing... and it has been surprisingly accurate.

Of course nobody argues that using MY projects in multiple machines / DAWs could be ideal to understand my headroom... but Dawbench is good to estimate the cost / benefit of going to the next machine without actually building it. Or to make an educated decission to go with an 8th generation Intel over a deal on a 1st generation Threadripper (I'd do the opposite for video editing). Or to decide if I'll go Intel or Amd in 2023 or 2024.

Dawbench can be improved for sure, testing low latency audio is more complex than benchmarking CPUs for video edition... but... is there anything better for this kind of audio benchmarking? (I am genuinely interested).

 

[Anthony]

AMD Teases 5nm Ryzen 7000 ‘Raphael’ Zen 4 CPUs, Unveils Ryzen 7 5800X3D with 96MB of L3 Cache

Firing back at Alder Lake

www.tomshardware.com

Glad to see that AMD is not resting on its laurels.

 

[Anthony]

The CPU and memory headroom as predicted by Dawbench was about 3x better. And indeed my critical project went down to 30% from some +90% (with ocassional drops).

So here's a real-world application where Dawbench was clearly useful. And in this case the "calibration factor" that parapentep70 mentions was close to 1 in that Dawbench predicted a 3-fold increase in computing power and this translated into a 3-fold decrease in CPU load.

I plan to do something similar before I build my new system.

Thanks for providing this information!

 

[Anthony]

AMD Teases 5nm Ryzen 7000 ‘Raphael’ Zen 4 CPUs, Unveils Ryzen 7 5800X3D with 96MB of L3 Cache

Firing back at Alder Lake

www.tomshardware.com

Just read the article...

The upcoming 5nm Zen 4 ‘Raphael’ (Ryzen 7000 family) chips could be perfect for DAW use b/c:
1. the new AM5 socket that Zen 4 chips will use supports both PCIe 5.0 and DDR5
2. the 5nm N5 process (which AMD may use for Zen 4 chips) reduces power by 30%
3. all cores (judging from preliminary Halo Infinite results) can be boosted >5.0 GHz

I think I'll wait until the second half of 2022 (as mentioned in the article posted by Pictus) to see if these are available and are as good as I hope. Fingers crossed...

 

[jamieboo]

AMD Teases 5nm Ryzen 7000 ‘Raphael’ Zen 4 CPUs, Unveils Ryzen 7 5800X3D with 96MB of L3 Cache

Firing back at Alder Lake

www.tomshardware.com

Very curious about this.
On the brink of going ahead and building a new machine but I'm prepared to wait for these 3d chips if they're worth it. (Although originally I had been hoping the 3d chips would be appearing earlier than spring).
Anyway, the benefits of the extra cache are so far mainly declared in gaming terms, but is this a technology that would benefit us composer/producers?
I do quite big dense Williamsy orchestral stuff (EW HO, largish template, not a vast amount of processing plugins), would the 3d cache tech be a boon for me?

Thanks

 

[Pictus]

Very curious about this.
On the brink of going ahead and building a new machine but I'm prepared to wait for these 3d chips if they're worth it. (Although originally I had been hoping the 3d chips would be appearing earlier than spring).
Anyway, the benefits of the extra cache are so far mainly declared in gaming terms, but is this a technology that would benefit us composer/producers?
I do quite big dense Williamsy orchestral stuff (EW HO, largish template, not a vast amount of processing plugins), would the 3d cache tech be a boon for me?

Thanks

I have no crystal ball or privileged information, but I highly suspect it will benefits...
How much, I do not know...
I also do not know if there will be more models than the Ryzen 7 5800X3D.

 

[jamieboo]

I have no crystal ball or privileged information, but I highly suspect it will benefits...
How much, I do not know...
I also do not know if there will be more models than the Ryzen 7 5800X3D.

Thanks Pictus
Yes, there's of course lots of unknowns. But I don't even know how onboard CPU cache benefits audio work in the first place, so am incapable of even a little reasoned speculation!
Yeah, I'm all over the place in my tentative new build speccing - first I felt pretty certain I was going to base things around the 5900x, then I thought maybe the i7 12700KF, but I was always curious about how the 3d cache chips might perform. But then if I'm prepared to wait until spring for them, then I possibly may as well wait for Zen 4 towards the end of the year.

(In reality, to avoid eternal resolution absence, I'll probably just get a 5900x in the next month or so.)

(Or maybe a 12700kf)

(Dammit)

 

[Anthony]

Yeah, I'm all over the place in my tentative new build speccing ...

You're not alone; I too am going back-and-forth between Intel and AMD. A lot of changes are currently taking place in the CPU/computer industry and I think it will pay in the long run to wait a few months to see what platform is best. I use my systems for nearly a decade before building a new one.

 

[jamieboo]

You're not alone; I too am going back-and-forth between Intel and AMD. A lot of changes are currently taking place in the CPU/computer industry and I think it will pay in the long run to wait a few months to see what platform is best. I use my systems for nearly a decade before building a new one.

A decade - wow! You're probably more adept at future-proofing than me!
I'm on a 6 and a half year old i7 5820, 32GB. Still a good machine, but I'm hitting some limits.
I imagine ANY of these options will seem a night and day difference over my 5820.

 

[easyrider]

Glad to see that AMD is not resting on its laurels.

It’s Intel who have been stagnant.

 

[rgames]

Here's an example of how I would use DAWbench (in the absence of something better).

1. Run DAWbench on my current (laptop) system and record the result
2. Run DAWbench on future potential systems and record those results
3. Plot DAWbench results of future potential systems against system cost
4. Use the above plot to select a system with the highest DAWbench/System_Cost ratio subject to the constraint that the increase in 'computing power' (relative to my laptop) will be sufficient to run current and future projects.

In this example "computing power" would be inferred via DAWbench as explained below.

For simplicity's sake, let's say that my laptop's DAWbench DSP result is 100 tracks and my CPU usage is 75% (which is unacceptably high). I would then make the inference that I'd need a new system with a DAWbench DSP of ~300 tracks because the old/new ratio of 100/300 would allow me to run my new system with ~25% CPU usage (which is 1/3 [100/300] of 75%).

Is this an optimal approach? No.
Do I want to do all this work? No.
Am I certain that I can use DAWbench as described above? No.
Are better (real-world) benchmarks available? (apparently) No.

As a scientist I've realized fairly early in my career that you never have completely optimal tools to do your research and, so, you learn to use what you have until something better comes along. I'm a pragmatist in this regard.

Cheers...

If the project runs at 75% CPU on one computer and 25% CPU on another computer then it runs on both computers. So from the standpoint of writing/producing music there is no difference between those two computers.

That's what I'm getting at. With dawbench there's nothing being measured that has any relationship to productivity or workflow. It's a benchmarker's playground without a tie to reality.

The response to this fact is the "but one day" argument: if I lower my CPU usage then that's good because one day I'll need it. But we never get to that day. I haven't been able to create a practical project that is CPU limited for about the last 10 years. Sure, some run at 50% and some run at 15%. But they all run, and at latency equal to or better than acoustic instruments. The bottlenecks are elsewhere and don't really matter any more. The only projects I can create that are CPU limited have no relation to anything that people actually do (like adding 400 compressors).

Here's another example: let's say one desk can hold 500 pounds and one can hold 10,000 pounds. Should you buy the one that holds 10,000 pounds because maybe one day you'll need to put 10,000 pounds of gear on it? You could, but who has ever done that? Dawbench measurements are like comparing a desk that can support 500 lb against one that can support 10,000 pounds: the comparison is pointless. (Again, as far as I can tell - maybe I'm wrong - but these discussions come up a few times a year and they always wind up just like this thread: nobody can provide examples of how dawbench results relate to workflow/productivity).

A meaningful comparison is not 75% CPU vs. 25% CPU. A meaningful comparison is a project that won't run vs. one that will. If latency is an issue then min achievable latency for some reference project is also a valid basis of comparison.

rgames

 

[Anthony]

It’s Intel who have been stagnant.

Right, that was implicit in my post: "Glad to see that AMD is not resting on its laurels (despite having pulled so far ahead of Intel who, by contrast, switched from technology-development to corporate stock buybacks which caused their product line to stagnate).

Nonetheless I still worry about buying an AMD CPU b/c the only one I've ever owned (K6) actually failed. Yes it's anecdotal (N=1), but none of the Intel processors I've owned ever failed; most of those systems are still in service living re-purposed lives.

 

[Anthony]

If the project runs at 75% CPU on one computer and 25% CPU on another computer then it runs on both computers. So from the standpoint of writing/producing music there is no difference between those two computers. [emphasis added]

Respectfully that's a bit specious. There are clear, measurable, important differences between 25% and 75%. But before I list them, let me say that I chose those numbers for simplicity's sake. The CPU loads incurred by my latest projects are now routinely close to 100% such that some no longer run without audio drop-outs, clicks/pops, etc.

Running at 25% CPU usage is preferable because:

1. CPU temperatures are much lower. On my system a lean project (25-30% CPU usage) produces CPU temps of 45-50 deg-C compared to heavy ones (85-100% CPU usage) which produce CPU temps of 80-90 deg-C.

2. Distracting, annoying fan noise does not occur. On my system the fan comes on at ~50-60% CPU usage and runs at full >75-80% forcing me to go into the next room to hear what I'm working on. That's a real PITA.

3. I don't have to worry about being frugal and keeping projects lean to avoid the issues described above. I have a lot of cool plugins that I cannot use b/c I lack sufficient CPU resources, and this kills productivity and inspiration as I search for CPU-friendlier (but inferior-sounding) plugins.

Cheers...

 

[rgames]

Respectfully that's a bit specious. There are clear, measurable, important differences between 25% and 75%. But before I list them, let me say that I chose those numbers for simplicity's sake. The CPU loads incurred by my latest projects are now routinely close to 100% such that some no longer run without audio drop-outs, clicks/pops, etc.

Running at 25% CPU usage is preferable because:

1. CPU temperatures are much lower. On my system a lean project (25-30% CPU usage) produces CPU temps of 45-50 deg-C compared to heavy ones (85-100% CPU usage) which produce CPU temps of 80-90 deg-C.

2. Distracting, annoying fan noise does not occur. On my system the fan comes on at ~50-60% CPU usage and runs at full >75-80% forcing me to go into the next room to hear what I'm working on. That's a real PITA.

3. I don't have to worry about being frugal and keeping projects lean to avoid the issues described above. I have a lot of cool plugins that I cannot use b/c I lack sufficient CPU resources, and this kills productivity and inspiration as I search for CPU-friendlier (but inferior-sounding) plugins.

Cheers...

OK that's fine. Then that's your basis of comparison: noise, or heat. So do noise/temp measurements, not # of compressors. Measure what you care about. You're still assuming a relationship, not demonstrating it.

Computer A runs at 40 db and 70 C and Computer B runs at 35 dB and 65 C or whatever. If that's what you care about then go measure that.

I don't care about those numbers (and I strongly suspect most people don't) but maybe it matters for someone.

I've never heard anyone say "I could write a lot faster if only my CPU temps were lower."

And I'm still waiting to see my first "burnt up" CPU after many decades of tweaking computers in all kinds of ridiculous ways. It doesn't happen in consumer products unless it gets struck by lightning or engulfed in flames or manually juiced with insane voltages.

So I'm afraid I'm still going to have to vote "irrelevant" on those metrics.

But hey, have fun

Cheers,

rgames

 

[thevisi0nary]

Respectfully that's a bit specious. There are clear, measurable, important differences between 25% and 75%. But before I list them, let me say that I chose those numbers for simplicity's sake. The CPU loads incurred by my latest projects are now routinely close to 100% such that some no longer run without audio drop-outs, clicks/pops, etc.

Running at 25% CPU usage is preferable because:

1. CPU temperatures are much lower. On my system a lean project (25-30% CPU usage) produces CPU temps of 45-50 deg-C compared to heavy ones (85-100% CPU usage) which produce CPU temps of 80-90 deg-C.

2. Distracting, annoying fan noise does not occur. On my system the fan comes on at ~50-60% CPU usage and runs at full >75-80% forcing me to go into the next room to hear what I'm working on. That's a real PITA.

3. I don't have to worry about being frugal and keeping projects lean to avoid the issues described above. I have a lot of cool plugins that I cannot use b/c I lack sufficient CPU resources, and this kills productivity and inspiration as I search for CPU-friendlier (but inferior-sounding) plugins.

Cheers...

You do you, but that’s like buying a gallon container to hold a pint of water instead of just buying a pint container.

Also you probably have inadequate cooling. I’ve regularly pushed my aging 4790k with a noctua fan to 70%-100% and hear nearly nothing.

 

[Anthony]

OK that's fine. Then that's your basis of comparison: noise, or heat. So do noise/temp measurements, not # of compressors. Measure what you care about. You're still assuming a relationship, not demonstrating it.

Computer A runs at 40 db and 70 C and Computer B runs at 35 dB and 65 C or whatever. If that's what you care about then go measure that.

I don't care about those numbers (and I strongly suspect most people don't) but maybe it matters for someone.

I've never heard anyone say "I could write a lot faster if only my CPU temps were lower."

And I'm still waiting to see my first "burnt up" CPU after many decades of tweaking computers in all kinds of ridiculous ways. It doesn't happen in consumer products unless it gets struck by lightning or engulfed in flames or manually juiced with insane voltages.

So I'm afraid I'm still going to have to vote "irrelevant" on those metrics.

But hey, have fun

Cheers,

rgames

You cleverly avoided responding to "3. I don't have to worry about being frugal and keeping projects lean...". That is an explicitly music-related issue.

Indeed, a system with more CPU power will enable me to write the kind of music I want to write rather than be limited to only what's possible given my CPU's limitations. And DAWbench is useful in that it'll enable me to assess CPU resources of future machines.

I should thank you for this exchange; prior to starting it I really never thought much about DAWbench, but everything you've said has, ironically, convinced me just how valuable a tool it really is!

Cheers...

 

[jamieboo]

So, broadly speaking, as someone who composes orchestral stuff (EW HO, largish template, dense 'Williamsy' orchestration, not a vast amount of processing plugins), do these DAWbench results suggest that Intel would suit me more than AMD?
I'm considering either the 5900x or the 12700kf. My early instinct had been inclining towards the 5900x but maybe the 12700kf would be better for the kind of stuff I do.
I know there are imminent new developments but, as things stand, what do you think?

 

[Pictus]

The Intel 12700kf is only better with DDR5.

 

[jamieboo]

The Intel 12700kf is only better with DDR5.

Ah, silly me - I missed that rather crucial detail! Thanks Pictus.
So it seems, of the two I'm considering, my initial instinct was right: for the kind of stuff I do the 5900x seems the winner. Right?