Is there any host sequencer on Windows that can effectively use 32 cores/64 logical cores?

[electro]

If you're doing video rendering (heavily multi-threaded - where ultra low latency isn't an issue), that's where Ryzen shines.

And there is still no AMD Ryzen chipset with Native Thunderbolt3 support? I read there was a new revision to the PCI-Express SPEC in the works (lower latency) due out around 2022. I assume this latency improvement would translate over to Thunderbolt 4.

[jonljacobi]

“All but useless” is a colloquialism meaning almost useless. Windows has never natively supported Thunderbolt. Intel provides drivers. This may change. TB3 cards are at least partially interchangeable. We use a Gigabyte Alpine card in our Asus X99 Deluxe system via the header.

[antic604]

“All but useless” is a colloquialism meaning almost useless.

[Jim Roseberry]

Windows has never natively supported Thunderbolt. Intel provides drivers. This may change. TB3 cards are at least partially interchangeable. We use a Gigabyte Alpine card in our Asus X99 Deluxe system via the header.

FWIW, Windows doesn't "natively" support many things.
Doesn't really matter as long as a quality driver is available.
Thunderbolt is an Intel technology...

• Intel chipset motherboard,
  Intel CPU
  Intel Thunderbolt controller

Rock-solid under Win10


Prior to Win10, there was no support for "PCIe via Thunderbolt".
Any audio interface connected via Thunderbolt-2 (under Win7/Win8) is running Firewire protocol.
IOW, Intel Thunderbolt-2 driver for Win7/Win8 (itself) won't allow PCIe level performance.
The OS itself has to support "PCIe via Thunderbolt".

[tooneba]

Some NVMe manufacture is offering their own drivers even though OS is natively supporting it. So native driver doesn't necessarily mean better.

[jonljacobi]

I was not intimating that there were any issues or that one driver is necessarily better than another.

[Freaqpeak]

There are several AMD boards with Thunderbolt3.
Actually all x570 motherboards from Asrock have Thunderbolt 3. Even at the sub 200 dollar range.

The low end ASRock X570 motherboards don't have a Thunderbolt-3 controller, they have a Thunderbolt-3 header (for their Thunderbolt-3 add-in-controller).

I don't know if you've tried to find their Thunderbolt-3 AIC Card... but it's rarely in stock.
On the Intel side, the Asus Thunderbolt EX3 (AIC) works with ASRock motherboards.

ASRock X570 AQUA has Thunderbolt-3 (as I mentioned earlier).
Limited edition board and costs $1000

ASRock X570 CREATOR has Thunderbolt-3... and it's somewhat reasonable at $500.


I've got a 3970x and will pickup an ASRock X570 motherboard today.
I'll do a build with Thunderbolt-3 and will provide figures.

Ah, I see. Didnt know that.
Creator looks really good. Please tell us how things went for you later on.

[Amelia70]

PS I found some of those DAW bench scores, and even the 3950X is annihilating the 9900K

Actually that was 3900X that Scan Pro Audio tested, so that's 12 cores, not 16. 3950X DAW benchmarks are still not available.

And I don't see it annihilating 9900K in DAWbench VI test, where 3900X pulls 100 voices less than 9900K at 64 samples buffer. At the same buffer, 3900X only has 3 plugin instances more than 9900K. So 3900X is not really annihilating 9900K, in fact

Am I seeing different chart?? I can see 3900X pulls 20 more voices than 9900K and 26 instances of plugins than 9900K at both 64 samples buffer.

exactly and that's just the 12 core.. impressive in my humble opinion, considering how strong the single core of the 9900k is

[Jim Roseberry]

Got the TRX40 motherboard, have to get some work done... then I'll get the build assembled and get some figures posted.

[EvilDragon]

PS I found some of those DAW bench scores, and even the 3950X is annihilating the 9900K

Actually that was 3900X that Scan Pro Audio tested, so that's 12 cores, not 16. 3950X DAW benchmarks are still not available.

And I don't see it annihilating 9900K in DAWbench VI test, where 3900X pulls 100 voices less than 9900K at 64 samples buffer. At the same buffer, 3900X only has 3 plugin instances more than 9900K. So 3900X is not really annihilating 9900K, in fact

Am I seeing different chart?? I can see 3900X pulls 20 more voices than 9900K and 26 instances of plugins than 9900K at both 64 samples buffer.

exactly and that's just the 12 core.. impressive in my humble opinion, considering how strong the single core of the 9900k is

Yeah that was a bad choice of colors in that chart. I was totally looking at wrong things, apparently.


Regardless of that, 3900X needed 4 cores more to compete, 20 voices more is neither "annihilating" nor impressive, 26 instances is let's say somewhat decent, but again it needed 50% more cores to get those results. Which simply shows that CPU clock and having all cores running at higher speed is definitely beneficial in the long run. Not only that, but also the design of those cores plays a huge role. AMD and Intel have quite different designs.

[Amelia70]

PS I found some of those DAW bench scores, and even the 3950X is annihilating the 9900K

Actually that was 3900X that Scan Pro Audio tested, so that's 12 cores, not 16. 3950X DAW benchmarks are still not available.

And I don't see it annihilating 9900K in DAWbench VI test, where 3900X pulls 100 voices less than 9900K at 64 samples buffer. At the same buffer, 3900X only has 3 plugin instances more than 9900K. So 3900X is not really annihilating 9900K, in fact

Am I seeing different chart?? I can see 3900X pulls 20 more voices than 9900K and 26 instances of plugins than 9900K at both 64 samples buffer.

exactly and that's just the 12 core.. impressive in my humble opinion, considering how strong the single core of the 9900k is

Yeah that was a bad choice of colors in that chart. I was totally looking at wrong things, apparently.


Regardless of that, 3900X needed 4 cores more to compete, 20 voices more is neither "annihilating" nor impressive, 26 instances is let's say somewhat decent, but again it needed 50% more cores to get those results. Which simply shows that CPU clock and having all cores running at higher speed is definitely beneficial in the long run. Not only that, but also the design of those cores plays a huge role. AMD and Intel have quite different designs.

The point is that if i can find a DAW (besides Reaper) that CAN use 24 or 32 cores, it WILL be annihilation.. I mean it won't even be comparable.. You are telling me the 16 core at 4ghz will not beat the 8 core at 5ghz and I am willing to put money on that, that it will win both in VI polyphony and overall effects count by a large margin. The only thing the 8 core will have over it is that it will be able to do say 15% or maybe even 20% more per core, because it can be set to work at 5ghz permanently.

I really don't know what effects chains people are using here, but I have never maxed out single core workflow using native effects even on my pathetic (by comparison) and slow 10 core. Even with the IK plugin, I can use 2 instances per core on this computer, meaning I can get 20 of them in Pro Tools on the 10 core Ivy Bridge. How does a higher core speed help in this case? If people keep saying speed matters but they don't explain why, I will not understand it, when what I see with my own eyes is pro tools using all cores evenly. This means to me that if I have 32 cores, I presume I would be able to get 4 or 5 of the IK tape on each core of such a powerful new processor, therefore 32 cores would be able to get me 150 instances. It's still not enough to put one on every track in larger songs, but it means there will be a lot less bouncing and freezing, and 64 track smaller projects will be a breeze with plenty of room for other stuff. Wouldn't a good way to test it be to see how many instances of the IK tape plugin a 9900K at 5GHZ can get, over say a 64 track project, vs a thread ripper? Isn't that the real world answer?
Oh, I tried the other tape plugin recommended to me, it's not even close. There is nothing on the market that sounds like the IK. It's in a complete league of it's own in my opinion. I was even able to improve an already excellent mix by replacing the tape plugin I was using (softube) with the IK and barely tweaking it. yes it was a long process with freezing and bouncing and it used more than 5x the power, but I had to know if my ears were deceiving me, and they weren't. All 5 of us that did a listening test blind picked it as the best mix.

Are there plugin synths on the market today that a 4GHZ processor can not cope with? Is this why single core speed is so important? I never had an issue with kontakt which is the sample engine all our libraries use, and it has it's own multithreading. Really, I am at a total loss as to why anyone thinks I would get better performance with a large 200 track project with an 8 core cpu vs a 32. I just flat out don't believe it if I'm to be completely honest.

[Jim Roseberry]

Are there plugin synths on the market today that a 4GHZ processor can not cope with? Is this why single core speed is so important? I never had an issue with kontakt which is the sample engine all our libraries use, and it has it's own multithreading. Really, I am at a total loss as to why anyone thinks I would get better performance with a large 200 track project with an 8 core cpu vs a 32. I just flat out don't believe it if I'm to be completely honest.

FWIW, I've tried to explain why high clock-speed is important.

- Not every process in a DAW can be multi-threaded (spread across multiple cores).
A developer from FLS explained this in detail (months back).
- The lower the latency at which you're working, the more important single core performance.
ie: Monitoring in realtime thru software EFX/processing at a 32-sample ASIO buffer size is not something that lends itself to heavy multi-threading. You can verify this by looking at the load of each core.
- I'm not sure if this has changed recently, but UVI's Falcon wasn't able to use multiple cores.

If it comes down to choosing clock-speed or cores, clock-speed is the single most important factor.
Xeon CPUs often have *significantly* slower clock-speed (but significantly more cores).
If you've ever used one for DAW purposes, they're an expensive performance hit compared to a standard CPU (fewer cores but significantly higher clock-speed).
A CPU with 16 cores running at 3GHz will not outperform a CPU with 8-cores running at 5GHz.
At the risk of repeating, performance increase from adding cores doesn't scale 1:1.
IOW, Quadrupling the number of cores does not quadruple performance.

If you're running at higher latency settings... and using a DAW that has "Hybrid Buffering" (where tracks that are merely playing back are using a *large* playback buffer), that's going to make more efficient use of multiple cores. If you're working at the smallest ASIO buffer size (say 32-samples), that's not going to make efficient use of multiple cores.
You can see this for yourself with Cubase.
If your audio interface allows, set it to a 32-sample ASIO buffer size.
Load a project into Cubase... and note the Audio Performance meter.
Now, switch off the Activate Multi-Processing option... and note the Audio Performance meter.

I'll have some specific figures for the 3970x in the next couple of days.
If it can run all 32 cores at 4.5GHz, I'll be amazed... and the performance will be fantastic.
If it can't run anywhere close to all cores at 4.5GHz, it'll be underwhelming.
Details to follow...

[EvilDragon]

Jim said it all. You don't have to believe it.

And yes, there are virtual instruments that can eat a lot of CPU for breakfast. Falcon (it does NOT multithread, hence single core speed is paramount!), Omnisphere, Repro-5, Reaktor Blocks once you start adding stuff, Softube Modular, and then on FX side Acustica Audio, for example. But there are many more examples.

[EvilDragon]

You are telling me the 16 core at 4ghz will not beat the 8 core at 5ghz and I am willing to put money on that, that it will win both in VI polyphony and overall effects count by a large margin.

I am not telling you, the charts linked told you. 20 voices more and 26 instances more between 12-core 3900X and 8-core 9900K is NOT winning by a "large margin". 4 more cores on a 3950X won't double those differences, things don't scale 1:1 when adding more cores, as Jim already said a number of times!

Now imagine how would it be if Intel had a 16-core 5 GHz monster. A 32-core Threadripper couldn't touch it, or it would be slightly above in certain multithreading workflows, but the difference in clock speed would prevail for the most part.

[tooneba]

As intel halved the price of corex so 9940x costs around $400 more than 3900x CPU&MB. It looks tempting.