Thanks for the responses all. I have a UAD Apollo x8p (Thunderbolt 3).

My current PC is an Intel i7 6700k, with an Asus Maximus VIII Mobo and an Asus Thunderbolt Ex 3 add-in card.

While I understand the hesitance about an amd TB4 setup, honestly thunderbolt on PC seems to be wonky however you get it. The thunderbolt on my intel/add-in card system has been very irksome. It's worked pretty well for the past year after a driver update - before then it was truly terrible, with me frequently having to uninstall the card and reinstall and reinstall drivers for windows device manager to recognize it. Thankfully those days ended about a year ago after an Asus driver update, but even now I often have to turn my interface off and back on after the pc has gone to sleep or restarted. It's wonky.

I am definitely going with a built-in thunderbolt mobo. I would like to go AM5 for future upgrades, and I am attracted to lower power usage the 7950X3D and 7900X3D show.

That leaves these two ladies:

https://www.asrock.com/mb/AMD/X670E%20Taichi/index.asp

https://www.asrock.com/mb/AMD/B650E%20Taichi/index.asp

There are asus boards that support a thunderbolt 4 add-in card (though curiously NOT the X670 Creator, but YES on the B650 Creator (wtf asus?)). But given my experiences to date with Asus AIC, I'm wary of that. The only X670 and B650 cards that have built-in thunderbolt 4 support are the Asrock Taichi line.

Yes you'll see plenty of 'USB4' motherboards with the same chip controller as TB4, but I've drilled down on people who've purchased those and the audio interfaces don't work with them or you have to hotrod/go through a bunch of shit to make what is already a wonky thing on PC work. And any given driver update could screw with that setup.

Perhaps one day I'll go RME if UAD continues making more of its plugins available as free-standing vst's on UAD Spark. But for now I'm still a UAD guy, which means I'm a thunderbolt 3 guy, which means I need a motherboard that supports Thunderbolt 4/3.

Is ableton particularly suited to Intel or AMD? I don't use Kontact (how do people deal with that shit interface?). I use the following plugins the most:

Soundtoys (alter boy, echoboy, decapitator, etc.)
Valhalla (verb, shimmer, etc.)
Synth Vsts (serum, model84, arturia V, etc.)
Izotope plugins (Ozone 8, trash 2, etc.)
UAD plugins
FabFilter plugins
Melodyne
Vprom, Addictive Drums, etc.

Thanks for the responses everyone

I’m on a 6 core i7 4970 (over clocked) it’s now 8 years and I’ve been holding out for a new PC hoping to get a big improvement when I do change.

At this point I’m just finishing of a project on my current system.

I’m thinking I should wait for the new intel 7nm 14th gen CPU’s …. does that seem sensible?

I’m thinking the 14th gen chip and 64gb of ddr5 is going to feel like a significant step up from my current workstation?.

I compose in Cubase and I’m probably going to start mixing in Pro Tools.

Okay so with further testing with the same interface : I was able to choke my 7950x3d (with cache preferred setting) with 4 tracks : slate and ash landforms kontakt library (awol preset), bass Ableton instrument, soundpaint library, and an addictive drums track, with 128 buffer setting. Ram is running at 6000.
I know the slate and ash libraries are infamous CPU hogs, but still I was hoping for better performance.
Ordered a Rme aio pro hoping this iron out the driver problem.
DPC latency is very good on the b650e-i motherboard however.

Put querulant mattiasnyc on ignore list. That is my advise about him. It's the last person in the world I would take any advise from especially on thunderbolt. You have been warned

What motherboard are you using @ Jazzmess ?

This is very interesting, thank you for sharing. What are your full specs?

I wonder how many tracks you would get if you set CPPC Preferred Cores to "Frequency".

Do you happen to own NI Pharlight? The default preset, The Alchemist, is known to be a CPU destroyer.

It's the startech adaptor that doesn't work between TB3 and TB4 or USB4 (as I understand). Generally the apple adaptor does work.

I do have one TB2 hub that doesn't work *well* on USB4 with x670e ProArt. It works, but sometimes you have to disconnect and reconnect to get it to start again. Otherwise, I've only used that one with TB3/4 stuff.

On b550 ProArt (TB4) everything TB2 that I've tried works (hubs, quantum, focusrite red8pre)- with the apple adaptor. I haven't tried much with the startech (I do have one) because I think it officially does not support TB4 or USB4 and the one thing I tried didn't connect.

I pretty much only use the apple adaptors and they work even on either end of a TB2 cable conecting between TB3 device and USB4 or TB4 computer (mac or win). I know this because I have long TB2 optical cables to the machine room...

Right, you're right in terms of how Reaper works and I apologize (I think, with 1566 it seems clearly to work this way but when monitoring other plugins and VIs it confuses me)...

And you're correct that I haven't been able to dig into this enough to figure out something empirical- but on the other hand, as per the other thread, most users are expecting/experiencing more disparity between CPU hit at different buffer sizes because that has generally been our experience.

IE, this is our day to day workflow multiplied by thousands of days and many users...I know that isn't empirical, but it isn't off the cuff either.

In any case, I think I might know why. For whatever reason SGA1566 really doesn't seem to care what buffer size it's using- the CPU load simply doesn't go up much at lower buffers, even with the tracks armed/monitored.

On the other hand, most of the more state-of-the-art plugins that I personally want to use (and be able to track through), do have quite a bit more instance count disparity between buffer sizes.

I just only had some limited testing on this this morning, but I think it might explain why the DAWbench DSP has always felt out-of-sync with many of our real-world experiences.

I'm treading lightly though, because more empirical data is needed....

I'm pretty sure I'm already (happily) on daskeladden's ignore list, but just for general reference to anyone who doesn't hang out in this thread this is what you need to know-

Where every thread on every forum has that one "it's my way or it's CRAP" person, Dask is that person for this thread.

If it's not Intel it's crap, if it's not gigabyte it's crap, if it's not a presonus quantum it's crap, and so on. Basically if he doesn't own and use it, it's crap (duh).

Outside of that aspect, he's really knowledgeable with windows tweaks, BIOS tweaks etc...Just a tad myopic in the hardware choice department.

I don't know how married you are to Ableton, but without a hybrid buffer it's going to be a tough choice for some of these modern CPU hog sample libraries.

The issue is most other daws at this point, would be able to switch all those heavy CPU VI up to a much higher buffer when the tracks aren't armed or monitoring. But, Ableton does not have that option, so everything in the project has to be able to run at the buffer setting your interface is set to.

In your case, I also find that a lot of these newer sample libraries are progressively using even more CPU at buffers of 128 and below. Which makes them kinda designed for a daw with a dual/hybrid buffer, something Ableton does not have.

It will be significant especially at low buffers. Be aware that protools is not an efficient daw with regard to computer resources. In a test I did I ram 6x instances of acustica jade on a 4700g CPU in protools before CPU peaks on all cores. I ran 3x instances in samplitude before a single core peaked, the rest were virtually at idle. That's just a single, simplistic test, but is in line with the general trend of PT being inefficient. If planning to use PT I'd plan on a more powerful CPU than you'd need on some other daws. Pt has strong suits, efficiency isn't one of them.

I'm not aware of any daw that reccomends one brand or the other. Some plugins like melda are optimized for avx-512.

7nm 14th gen isn't goming to desktop, so smaller node will be 15th gen desktop. The rumor is that instead they will do a 2nd Alder Lake refresh (or a Raptor Lake refresh). On the upside LGA1700 might have one more update cycle after all.

All good, trust me when I say it has taken me a long time to get my head around the extended buffering dynamic that AFX can bring to the table, and I still have plenty of unanswered questions which I am still working through.

No argument there, the benchmarks are simply a vehicle by using a DAW to apply an empirical incremental load , and using a metric that can scale with architectural improvements. I have always maintained how that translates to individual session working environments will depend on lots of factors, from session logistics to end user understanding of managing the logistics.

Again, no argument as per my previous response.

Its not only 1566, the same dynamic applies to the RXC and even any of the plugins used in more recent BETA sessions I developed with the help of Mikko to include Protools/Cubase/StudioOne. I also had a suite of 7 plugins for the testing around 10 years ago, but with most being full versions , it limited the testing to those that actually had the 3rd party plugins. All the test plugins displayed the same dynamic.

What they have in common is a smaller to moderate individual load so that we can apply the processing load in small increments to the break point.

Of course, plugins with larger individual processing requirement would not scale in the same manner incrementally. We can assign a far higher load in 1566 to emulate that if need be.

I also supply blank ( DSP ) session templates in the downloads, so anyone can configure custom versions of the sessions using their own preferred load plugins.

With any and all the benchmarking options available, whether they are synthetics or based in respective DAW's , YMMV !

this is not confirmed only speculations. But of course 15th-gen Arrow Lake can be the "true desktop replacement" for Raptor Lake. If I remember correctly this has happened before with Intel (Ice Lake and Tiger Lake which were only for laptops).
No big deal i9-13900K should be good enough until 2024. What we need now is some better air CPU coolers

Unless using the same exact system Scan uses, different people will get different results using the same dawbench method.

This is why dawbench can't be used as anything but anecdotal. Not only does it not translate to real world usage, it's results are not comparable to each other because the test itself is system dependent.

This causes issues like the unknown reasoning behind the exceptional performance of 12th gen Intel with ddr5. Or me not being able to compare result reliably with other users.

The methodology is just too abstract and unscientific to be used for anything beyond a marketing write up, or fun little test.

True. Widely circulated already, though. I do hope that it's true and they'll keep the 6 P-cores architecture to mobile only.

Ok, presenting hopefully the first of more tests (done by more people) of DAWBenchDSP- ARMED. The 2nd amendmend edition (funny, not-funny).

This test is done by using the standard DAWBenchDSP but deleting all the sine waves, and arming the actual tracks as you go across the various buffer sizes. The audio is played and the number of tracks
is reported at the last point that get's through the loop glitch free.

I'm reporting the number of tracks here because the FabFilter results are Pro-C2 and Pro-R both on default except Pro-R changed to ~3% wet.

It's also necessary to have actual audio inputting and listening to it (for glitches) and I suggest that future tests should be invalid if the audio input is anything other than Smooth Jazz Scat Soloing (also funny, not funny).

Anyway, here is what I get.

2 interesting observations. I'm seeing quite a bit more disparity between 64 buffers and 512 with the tracks actually armed. And 2) The FF combo (hardly the most state-of-the-art) has greater disparity between 32 and 512 than the 1566 does.

It would be great to see other people try this.

I'll add to this that on my personal testing comparing 2 12th gen chips and one 13th gen, I've also done testing with several "modern" plugins that I use myself and compared them to the ones used in the DAWBench tests. I haven't found notable differences in scaling behaviour, but the sample set is of course very limited and it has been just my own informal testing for my own curiosity so this is definitely just anecdotal. FWIW, plugins I've used include Oeksound Soothe, Fabfilter Pro-L2 (with zero to massive oversampling) and Liquidsonics Cinematic Rooms Pro (Anecdotal comparisons against the DAWBench DSP and Bus plugins on Reaper, as well as the plugins used for the Pro Tools Bus beta testing).

Would be interesting if some plugins were found that behaved clearly different.

Just a follow up here to highlight what is different, even with SGA1566, between DAWbench Armed and Dawbench not-armed-

On the (scan?) posted test that I found of 5950x, they have

64 | 381
128 | 405
256 | 426
512 | 432

On mine (which has PBO kicking and 2x ram to fclk), and with tracks armed I got

64 | 256
128 | 304
256 | 456
512 | 520

I tried to be as judicous as possible with the glitches. IE, I didn't start opening new animated web windows while one was playing and then not do that for the other. But with armed tracks at low buffers you do kinda feel like if you wait long enough, there will be a glitch. At least until you give it even more overhead.

In any case 1.13x vs 2.03x

Anyway, @ TAFKAT I'd love it if you were willing to mess with this as well.

There is a high probability that I did make some sort of mistake, but I would be surprised if this trend not only didn't hold up, but also was exaggerated for more modern/heavy plugins.

Were you testing different buffer sizes with tracks armed/monitoring though?

That's the main hypothesis that I'm presenting-

Modern, heavier, state-of-the-art plugins tend to stress the CPU incrementally more when monitoring through them at low buffers.

Armed and recording silence ?

Lets move this to the DAWbench thread Ryan, instead of high jacking this thread.

Armed and monitoring, I didn't actually hit record (maybe I should have?).

And sure, lets do it.

My question would be how much of the difference to attribute to the armed/not armed as opposed to the different audio drivers, or ram, pbo, daw buffering, ect?

Oh!

So do this means I may as well get a 13th gen workstation now?

2025 is a long way off!

I’m a little bit concerned as I heard Cubase has issues with drop out and windows 11 and intel’s current CPU’s?

Yep. Pick Intel. Zero problems:

LOSING MY MIND! Focusrite Scarlett 2i4 and Windows 10

Lynx Aurora(n) Thunberbolt with Windows10

Thunderbolt 4 - problem solving thread

Thunderbolt 3 Audio interface connectivity issue PC

Thunderbolt 4 - problem solving thread

Problems with Realtime Performance of UAD Apollo over Thunderbolt 3 on Windows 10

UAD Apollo x4 not recognized - ProArt Z690 motherboard

Thunderbolt 3 Audio interface connectivity issue PC


PS: Kudos on using "querulant". Too bad you couldn't get "advice" right.

I have never seen that problem being reported by a user. I have low faith in their un-recommendation of 13th gen Intel.

For me it works, from th ex4 asus card through apple adapter 3 to 2 , then to thundb2 UAD satelite, than to Apollo x16 (th3 device) again th2 to th3 with apple adapter and at the end 0.5m th3 to th3 cable to Aplollo x8p.
No problem at all, Asus rog strix E with th ex4 card.
If I would sell my old satelitte and change it for new with th3 interface it would be much more simple but it works. Even at low latencies 64 or 128. I say that cause Im using 2X 3m long cables in thunderbolt rig cause my PC is out of control room so I need it, but suprisingly works well.
Only that people know what is possible with thunderbolt... I had last year cheaper MB Asus Tuf gaming Z690 with TH EX4 Card and can also recomend it.
I think it is cheapest combo THb Card+Mb, about 350eur.

Prices for the new ASUS Xeon boards have been listed at some sites

https://wccftech.com/asus-w790-mothe...e-899-for-ace/

£899 and £1279 ($1513usd)

Ok, thank you.

That’s good to know.

Any more conversation about Xeon vs i9 for DAW use, cost aside? Or maybe, new highest i9 tier vs lower/mid tier new Xeons?

Xeon build definitely wouldn't be cheap when you factor in the cost of just a processor and a supporting mobo could cost $3000-$4000.

But for myself, I would consider financing such a build if the benefits were clear.

If you didn't guess, I use Cubase which is a very big program compared to other DAWs.

Some computing tasks being dealt with in my workflow.

- Using Cubases MediaBay - massive complex file manager/browser for sound, asset, preset browsing. Dealing with scanning/updating/search 100,000s of sounds.

- I create very big templates with high track counts, pre-loaded inserts, sound assets.

- Complex sound design with a lot of track layering and intensive FX processing

- Both MIDI and Audio production/composition, sometimes individually sometimes simultaneously.

- Large audio batch processing with thousands of files

- heavy FX stacking both serial on single tracks, and parallel on separate tracks... as well multiple serial track FX stacking

- spectral editing at high resolutions/bit rate

- I work 88.2/96khz 32bit float almost exclusively

I just think going Xeon seems right in my head, but it's hard to say... and I doubt anyone is going to test this stuff for a while in a DAW scenario.