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

[Jim Roseberry]

If/when AMD can get clock-speed up around 5GHz (across all cores), that's when "Ryzen" becomes a whole lot more exciting.

Another thing to keep in mind about going AMD; there's effectively no Thunderbolt-3.
A lot of our professional composer clients (TV and Film) are moving to the Antelope Orion (via Thunderbolt-3).
Lack of Thunderbolt may not be a show-stopper, but it's taking a significant number of (low-latency) options off the table.

[EvilDragon]

TB2 should suffice tho?

[Jim Roseberry]

but at some point if ONE core gets oversaturated (due to a heavy plugin, or something), it doesn't matter how many cores you have, you will get dropouts. This is why you want to have the fastest cores possible. Even if you end up having fewer of them.

+1

And I'll repeat, performance from adding cores doesn't scale 1:1

16 cores running at 4GHz will not out-perform 8 cores running at 5GHz.
I'd also caution that you may not see 4GHz across 16-cores (if talking about AMD).
That makes the comparison even worse.

For AMD to really compete, they need to get clock-speed higher... and especially across all cores.

[Jim Roseberry]

TB2 should suffice tho?

Other than limited edition motherboards ($1000+ USD), I've not seen any AMD motherboard options that include Thunderbolt.

Microsoft only supports "PCIe via Thunderbolt" with Thunderbolt-3 controllers.

[EvilDragon]

TB really just IS PCIe over cable, tho?

[Jim Roseberry]

TB really just IS PCIe over cable, tho?

Yes.
It's essentially external PCIe.

[Jim Roseberry]

BTW, I have zero bias against AMD.
I'd love to see Ryzen (or whatever code name at the time) make serious advances in clock-speed.
We used Athlon CPUs in the past (when floating-point performance was significantly better than the Intel P-III and P-IV).

[EvilDragon]

And I'll repeat, performance from adding cores doesn't scale 1:1

16 cores running at 4GHz will not out-perform 8 cores running at 5GHz.

This is exactly what I'm saying.

[Amelia70]

It's not that simple of a calculation, Amelia.

32 cores will help and be an improvement over your 2013 MP. But they should be at the VERY LEAST 4.0 GHz, all of them. Sustained at all times. Not boosted in/out of.

However once your FX chain becomes too long/hard to process, you'd have wished you had those 5 GHz cores. But there's no 5 GHz 32-core beast out yet.

I am pretty sure the 3970x can do 4 on all cores

[Amelia70]

I am starting to think this must be a reaper limitation. I am not sure if you know how Pro Tools works (apparently logic and cubase can do it also) but if you arm a track to record, it frees up that core entirely for the armed track, and room permitting, moves everything else on a higher internal buffer rate, 1024 samples I believe, to the other cores. I am certain this is why we have never had an issue even playing large kontakt patches with effects.

However, I WILL meet you half way - something like a 2.5ghz is much too slow.. which is why I don't like the new 28 core mac pro (for example). When it is under serious strain, it seems all 28 cores go around 3.1 ghz.. This is absolutely NOT ideal for a DAW workload, we are agreed there, especially since I am moving away from pro tools DSP.
BUT, if it is indeed true that the 3970X can sustain 4ghz on all cores happily (with adequate cooling of course), then, IF i can find a DAW to use all those 32 cores (really, it's 64 logical cores), I would prefer that to an 8 core machine that is 1ghz faster per core. For me and my workflow it's not comparable. I really don't want to argue about it cause it's silly.. I know you are trying to help but you have cautiously avoided telling me how much more powerful a 9900K may be to the Xeon 3ghz 10 core I currently have. In geekbench, the single core scores are almost double. This should relate to Vi performance on a single core, shouldn't it? I mean relatively speaking of course.

Yet, the 9900K is only around 10% faster in single core than many 16 core CPU scores I have seen.. therefore how can the 16 core not be the superior overall processor for large track counts? I see scores for the new mac pro 16 core at 5600 single core and something like 6200 for the 9900K (geekbench 4). This tells me that it's not THAT far behind and in fact, the AMD seems to be getting very strong single core scores. Is there any sort of test or database showing how many plugins one can run on a variety of modern cpus? That could be a great help.

I am asking for a computer that can do everything on the one machine, natively. I just don't see how an 8 core 9900K can do it. I am sorry, I have tried every avenue in my mind, and although not a computer genius, I understand enough to know that I will overload it if i try to do everything I previously used one 10 core 3ghz xeon processor and 36 HDX dsp's for. Do also note, before one of the cards died irrecoverably, we had THREE of them..There were hundreds of plugins on those cards, including HEAT. Now, the tape will have to be done 100% natively. That in itself is going to add some serious overhead. Yes I am stubborn about tape plugins, but only cause I like the really good ones.. Contrary to what you might think, I do not overload on the analog. I don't start adding extra saturators, channel emulators, transformer modelling EQ's, etc. I just like the process of one nice tape plugin on each audio track, and that is the way I work. I tend to use clean linear phase EQ's for 90% of my EQ tasks or just standard digital EQ's, like fab filter in it's default mode. If you are saying Satin can honestly do what the IK tape can, at much less CPU, then of COURSE i will try it. In fact I am going to go and download the demo right after this posting.

[EvilDragon]

I am pretty sure the 3970x can do 4 on all cores

Actually it probably can't. Those boost speeds it can do are only on a few cores, not all of them.

https://www.tomshardware.com/reviews/am ... x-review/2

Also, those boosts are very short-term, not sustained like Intel's are. And you NEED sustained speeds above 4 GHz. Well, at that point you might call it a full-blown overclock, rather than a temporary boost.

[EvilDragon]

I am starting to think this must be a reaper limitation. I am not sure if you know how Pro Tools works (apparently logic and cubase can do it also) but if you arm a track to record, it frees up that core entirely for the armed track, and room permitting, moves everything else on a higher internal buffer rate, 1024 samples I believe, to the other cores. I am certain this is why we have never had an issue even playing large kontakt patches with effects.

It's not a Reaper limitation. Reaper deals with CPU completely differently. It has something called "anticipative processing" for unarmed tracks, where it processes things in advance in order to maximize available CPU. It works much better than whatever PT is doing for multithreaded workflows

[EvilDragon]

In geekbench, the single core scores are almost double. This should relate to Vi performance on a single core, shouldn't it? I mean relatively speaking of course.

It is also the main measure for a DAW - the main real-time audio thread is always run on a single core, it cannot be multithreaded. It is your master track. If this gets bogged down, no matter how much free CPU you have left, you will get dropouts. THIS is why single core speed is still the most important!

[EvilDragon]

Is there any sort of test or database showing how many plugins one can run on a variety of modern cpus? That could be a great help.

ScanProAudio does these, but it's not up to date currently. There is a DAWbench VI test (google that) that tests Kontakt usage, and then there's a regular DAWbench test that tests pure processing power by counting how many instances of SGA1366 plugin a CPU can run until it chokes.

[antic604]

I just like the process of one nice tape plugin on each audio track, and that is the way I work. I tend to use clean linear phase EQ's for 90% of my EQ tasks or just standard digital EQ's, like fab filter in it's default mode. If you are saying Satin can honestly do what the IK tape can, at much less CPU, then of COURSE i will try it. In fact I am going to go and download the demo right after this posting.

I think the suggestion was to use plugins - if possible - on groups / busses / folders, rather than on each track individually, unless the settings are very, VERY different for each track.

And BTW, if those are reliable (only 7 samples for 3970x) then it looks really promising:

https://www.cpubenchmark.net/compare/In ... 3334vs3623

It's really a pity there's no website and/or YT channel dedicated to testing DAW performance in a systematic, organised manner. It's always only Adobe Premiere / Final Cut Pro and sometimes a gaming benchmark that's useless for DAW workloads...