10 U-he Diva pieces, on a Devine option with 5 voice chords, 15 polyphony and 3 voice unison, all cores are used. What is the maximum sounding voices for this plugin and max setup. 256 ms buffer on RME, Win11
What’s funny is that ASIO can use all 100 percent of the processor)
Try this and compare
1 Like
You have to understand that these are very extreme settings for a plugin. But this plugin supports multicore, while others do not generally. And the result can be worse or better
Windows 11 is stable. Please explain why you’d do that.
A reason that isn’t historical FUD, BTW.
Intel is under more pressure because they are used more in portable devices like Surface tablets than AMD.
And that won’t change soon seeing as how Intel is still basically 10nm and has taken the core performance lead back. They actually have more headroom for optimization than AMD. They just have to execute properly.
AMD’s biggest advantage is in multithread, but that is less important on those devices as they don’t tent to run applications that stress CPUs in that way - at least as much as desktops or laptops.
Intel has also improved their Integrated GPUs considerably, which was another area where AMD had historical advantages. But every laptop worth its salt has a dGPU that performs better, these days. iGPU performance isn’t the consideration it once was.
Hmmm…
I’m not sure I entirely agree. I think the biggest problem Intel will have moving forward is energy efficiency. The new CPUs consume a really large amount of power when they’re under full load, compared to AMD. In addition to that AMD has its new smaller node coming out next year. So taking those two things into account I actually think AMD’s biggest advantage is power consumption being low at relatively high performance.
It doesn’t matter if all the working ranges are on the desktop. And that matters if you want a laptop.
Actually, I don’t really agree. 12700 and Windows 11 cope very effectively with the load during normal tasks and when I am texting with you about 35 degrees. Under heavy load up to 80 degrees, but this is usually rare.
a little later, just a little busy)
This is completely for green users who do not understand anything in the settings, I configure myself)
Sorry, not sure what you mean above.
But it’s not about just keeping the temperature down, it’s about keeping the temperature under control while increasing performance.
If you look at extreme overclockers for example they are able to get very high results when processing but it comes at the cost of higher temperatures that need better cooling. It’s the lack of power efficiency that makes the cooing requirement go up.
And as I said that can be a big problem in the future, and my bet is that AMD is much better able to create a more powerful processor at X Watts than Intel. This in turn means that with ‘equal cooling’ AMD could create a more powerful CPU, or conversely a lower power one.
PS: I’m talking about the future now.
We don’t have to unless we want to do DAW-use-specific comparisons. Unfortunately that won’t mean pretty much anything unless we all do the same test on different CPUs, which we won’t. So the load will be different for everyone who runs a test. Best thing I can think of is if everyone ran DAWbench to it’s limit and reported CPU power draw at the point of failure, but nobody here will do that because we don’t have time or the means or knowledge to do so.
In the meantime there’s this article though:
Under the headline “Power Consumption, Efficiency and Cooling:” you can see the power draw for both Alder Lake and AMD’s 5xxx-series CPUs.
Really the question is “If AMD had allowed their Ryzen CPUs to use as much power as Alder Lake, what would performance have been?” Then we take that question plus the charts in mind and we consider what AMD’s new 5nm process node can bring under different power scenarios.
In that context I think AMD is well prepared to battle both Intel and Apple on both the high and low end of the spectrum.
I can play the 10-track Diva test file without issues on my Z390/i9 9900k Hackintosh in a loop, no overload errors, with a 256 sample buffer. Am I supposed to check for crackling? Haven’t switched my monitors on.
Alder Lake with 12700K should be quite a bit faster than my Coffee Lake setup.
2 Likes
not quite so, there are permissible limits, if you go beyond them, then it can burn out, because you violated this and technological norms
Alder Lake performance results:
Talking about the future then I would say the following: It seems that the improvements in DAW performance in Alder Lake mostly comes from using DDR5 memory. Now take into account a possible 15% performance increase as AMD moves to 3D-cache, and then likely another 15% increase as they move to 5nm for the next new CPU architecture, and then add the same improvement in performance as Intel got from moving to DDR 5 for AMD’s AM5 socket and you have - compared to today’s existing Ryzen CPUs:
5xxx - 100%
5xxx 3D cache - 115
6xxx AM5 - 132
6xxx AM5 w DDR5 - 198
In other words possibly twice the performance of today around the fourth quarter next year. Pretty remarkable if it turns out that way, and that would indeed be my prediction. So we can take the numbers for the 5950X in the charts in the link and pretty much double them and that’s where we’ll be in less than a year.
In terms of power efficiency… that’s why that’s a thing.
That test is largely irrelevant (DSP wise, I didn’t look at VI part).
DDR5 only matters with VI, not with DSP.
Also they use “dough in a pan” style test where all cores utilized evenly. In the real world you have busses and subbusses that stress few individual cores much more than rest of the cores so you may run out of CPU power on a 16-core AMD cpu long before you do on 8-core Intel that has a much faster single core performance because depending on the project it simply won’t need that many other cores for channels with low plugin load, but a few plugin heavy busses will totally cripple it.
Doubling 5950x in a year is pure dreaming. 15-20% maybe. VCache may not do a thing for DAWs performance.
The keywords are “may” and “long before”. Yes, that “may” be true. It “may” not.
I’m not sure what test you think would be relevant outside of a DAW specific test that targets a single core, but if you take Anandtech’s single core testing it seems that the results vary. The 12900K is ahead by as much as 22% to as little as 4% in most tests (see below), is behind on two out of ten, and it essentially ties on one.
So it’s likely ahead if we were to isolate a single core.
It’s not entirely clear that “long before” applies, though it certainly could.
Sure, VCache may indeed not make a difference at all… may… again.
And I should have qualified it by saying that I was talking about the VI test. I wouldn’t expect DSP performance (as in the test) to double. VI I could see though.
Still, I think it’d be silly to think there would be zero gains from VCache and zero gains from DDR5 (VI) and then 15-20% only from the process node improvement. Not that it matters, we’ll see soon enough.
Just upgraded to an Alder Lake system after my 6700K / X99 mobo died. I’m running a 12700K / Aorus Z690 Pro Mobo / 32GB DDR5 5200 RAM / Windows 11 / Cubase 11 Pro and so far it’s been running perfectly smoothly.
Not a scientific test, but opened a session that was on the edge of glitching with my old system (Izotope Ozone 9 limiter, Waves Abbey Road Chambers / Plates and Spectrasonics Keyscape -real CPU killers plus assorted compressors and EQs), and it was able to handle everything effortlessly with a few additional Abbey Road reverbs thrown in and a 9ms latency. Performance meters running about 50% and completely glitch-free, running without tweaking any settings. I can easily say it’s the best upgrade I’ve ever done…
2 Likes
Just upgraded from 8700K to 12900K / Asus Z690 Apex / 32 GB 4800 RAM. Dual boot win10 and win11. Except of sonarworks soundId (doesnt work under win11) cubase runs with all vsti smooth on both os. Speedstep off - p- and e-cores on. No problem at all.
I got native instruments komplete, spectrasonics all vstis, real guitar, amplesound. orangetree samples … - works
2 Likes