Actually, it is possible to process serial plugins in parallel. Cakewalk does this with their plugin load balancing, there is a nice explanation in their manual (also describing the limitations and trade-offs).
It’s “possible” but is there a benefit? Even if you put plugin A on one thread, and Plugin B on another thread, the information from plugin A still needs to be processed and then sent to Plugin B even if it’s on another thread.
So it’s not realllllly parallel is it? I think the benefit perhaps depends on the intensity and amount plugins on one channel and in the project to prevent singular threads from getting bogged down. It’s not really parallel though, it’s not possible, otherwise for an example, Plugin B (compressor), wouldn’t respond to changes in the EQ curve of Plugin A.
That’s my understanding, and also, because CPUs are so powerful now and thread counts are so high, that sort of distribution isn’t really necessary. Correct me if I’m wrong.
Read the linked document, it is pretty well explained there, also the limitations and scenarios where it wouldn’t work or bring no benefit.
There is a clear benefit, depending on what the plugins are doing and if they are dependent on each other. To make it simple, there are two possible cases.
One is, there are plugins where the output of one plugin is required as input to the other plugin. These plugins can be started on separate CPUs and as soon as the first plugin starts output of processed data, the second plugin can take them and do its work. This is something called pipeline processing, which is kind of the simplest way of using parallel processing.
Then there is real parallel processing in case you have a plugin that is loaded in two or more flows. In this case this plugin can be started multiple times, on multiple CPUs and run fully parallel, processing independent data from independent flows. If the output data of theses plugins need to be processed by another plugin in the chain, the data need to be combined (gather) and then sent into that plugin.
There are many ways of combining data back to a single stream, for example something that is called Interprocess Communication.
In reality this is much more complicated, but it is the basics of parallel processing and standard technology these days.
The linked document above is explaining exactly this kind of processing and why you get a benefit.
The question is, what do you mean by CPU. Is it just the chip you put into a mainboard, or is it the number of cores this chip has? If you have a high load of threads, what do you think who is going to run these threads?
The reason Intel, AMD and Arm(Apple) are adding more cores to the CPUs is to allow more parallel processing, pure CPU power is limited by physical boundaries.
Not trying to be a smartass, but actually trying to understand here. But if that’s true what you’re saying… how did they do it in ‘the old days’ with just single core processors? They had multiple tracks too.
The process on a single core machine is pretty much the same, except plugins were way less cpu intensive back then and you could use fewer of them
Very simplified explanation:
Say you have four tracks each with a few plugins on. Now cubase will create four compute threads, one for each track, and each is responsible for computing the plugins on the track (sequentially).
Now if you have a CPU with four cores, the operating system can distribute those four threads evenly on those four cores. If one of those cores saturates though, because of too many demanding plugins in a row (or other tasks), you have a problem.
If you only have one CPU, the operating system will schedule the four Cubase threads on just that one CPU, but of course not in parallel as it could do with four cores, just alternating… Given the same set of plugins, you will run into dropouts way earlier than with four CPUs, of course.
And what about this official post from Steinberg
If I am not mistaken this should be an issue also in the 13th Gen.
Did you need to deactivate any of the efficiency cores?
The title has it all, it says “or newer”, which means 12000 or higher. All new Intel processors use this architecture.
Hi… No… Everything is running stock… Except for the cpu which I am OC’ing… P-Cores running at 5.6, E-cores running at 4.3. Aside from that the entire stack from mobo on up is stock.
Im using 12700K in Cubase 11 Pro since March this year and I dont find any performance issues at all - in heavy projects with hundreds of plugins, synths and over 150 tracks all works smooth and fine. I use Windows Maximum Performance power plan.
What Windows are you using? Apparently it’s the thread director in Windows 11 that’s causing this according to the article. (This matters deeply to me, since two weeks ago bought my new mobo and an i9 13900, so as soon as I read this got pretty concerned at first)
I use Windows 11 Pro 22H2. Did not found any issues regarding new Intel Thread Director. All works smooth and great performance as expected in Cubase.
Me too except my build is windows 11 Home and my cpu is a 13900k. Evidently there was an upgrade in one of the recent Windows 11 updates which goes a long way to remedy p-core/e-core issues.
She’s a bit of a looker!!
I’m awaiting new Z790 and i9 13900K. Upgrading from Z390 and i9 9900k and brimming with excitement and anticipation for what should be a huge performance boost in Cubase. Recently bought an RME HDSPe AIO Pro and discovered my ageing system was the real bottleneck. Hopefully the 13900k will be a good match for the AIO Pro and Cubase will allow me to run a lot of plugins at ultra low latency. I now choose all my plugins very much based on lower latency and the dream is getting to final mix (and even master!) stage without increasing my audio buffer more than 256-512 and ultra low latency fx chains (without ASIO-guard active) so everything is super responsive right up until the last mastering chain is active.
Has Steinberg updated their guidance on stability of intel processors with E cores? My impression was that 11th gen intel processors are as far as they recommend because of the E core thing.
Well… my new build/upgrade certainly did not disappoint! Performance is off the chart. MASSIVE improvement over my I9 9900K.
64gb corsair dominator 5600mhz cl40
Asus Prime z790 P
3 x NVMe M2 fast storage
My RME HDSPe AIO Pro is now unleashed! Super low latency and running projects previously crippling my system (choppy even at 2048 buffer and ASIO Guard high) now at 256 buffer and no ASIO guard. The difference is just night and day.
Here you can see the load balancing across all logical processors during this project playback (not a massive project in terms of tracks and VSTi numbers but some fairly hefty buss processing going on)… Cores 1 and 4 seem to max out but this doesn’t cause audible glitches. As I say, previous system didn’t have a chance of running this even with ASIO Guard high and max buffer.
The benchmarks show 75% increase in memory bandwidth (read/write) over my last system’s 3600mhz CL16 RAM and numerous cpu tests/benchmark tests up to 6x performance (multi core tests blown away) !
I’m so pleased with this system. In nearly 30 years of building and regularly upgrading my PCs I’ve never seen such a dramatic increase in performance.
Happy to give more details if anyone is interested.
Yes, more details please.
What is the OS ?
Not familiar with the bechmark program shown. What is it ?
Are E-cores disabled ?
Is Hyperthreading disabled ?
Could you describe the tracks (synths, effects) a little more?
I just finished a very similar build upgrading from a 9900k - I to have never seen such a dramatic performance jump.
I have a simple UR44 for my audio interface but it’s managed to drop latency by half and still be stable with other programs running.
I did have a slight perf glitch caused by the NVIDA graphics driver (multi purpose machine has an RTX 4090). that was fixed by disabling G-Sync in Windowed mode (full screen is fine).
I’m looking at trying process lasso to auto change power plans and shutoff core parking just for that little bit of extra (and I like turning the knobs on the PC - My i9 is hitting 6.2GHz on 3 cores under light loads!)
I still need to try out my full orchestral template but it’s all looking very good indeed.
That would be Windows 11 and the Task Manager (not a benchmarking program).E-cores are enabled and in use (the lower 1/2 will be the e-cores).
Hyperthreading is active - hence the top half has 16 “cores”
Please post the results when you try your orchestral template. I do mostly large HSO and Iconica projects and I am trying to decide what my next upgrade will encompass…AMD vs Intel and so on.