Question How much compute do you need for foreground application real time user response?

[Hulk]

I'm not talking about encoding video, audio, or other compute related applications that don't require constant user intervention.

I'm talking about applications like video editing where you need real time preview, multitrack audio editing, where you need glitch-free playback, photo editing, where you don't want to wait more than a second or two for filters or other operations, gaming, etc...

I have found that the 8 Golden Coves in my 12700K provide a good foreground application user experience for my primary applications which require compute. Magix Vegas Pro 19, Presonus Studio One, Photoshop Elements being the primary suspects for me.

Intel seems to be moving in a foreground/background CPU specific direction while AMD is staying with the technique we're used to which is throw all of the compute you have at all applications.

Obviously both have advantages and disadvantages.

 

[moinmoin]

That's completely dependent on the OS and software used. For an OS example BeOS was built around giving the user instant feedback at all time, at a time computers were still single core.

The use cases you list all would profit most of hardware acceleration that in most cases would use the GPU, not the CPU. Apple with M1/M2 is going that route, that's how they can claim "up to 15x faster than Intel‑based model" (as just happened with the announced M2 Macbook Air).

 

[Markfw]

That's completely dependent on the OS and software used. For an OS example BeOS was built around giving the user instant feedback at all time, at a time computers were still single core.

The use cases you list all would profit most of hardware acceleration that in most cases would use the GPU, not the CPU. Apple with M1/M2 is going that route, that's how they can claim "up to 15x faster than Intel‑based model" (as just happened with the announced M2 Macbook Air).

I agree. Most here say my use case is not like most, but I wonder how much servers are like what I do. I know where I retired from, our servers were virtually hammered at 100% 24/7, like what I do. And I do use them also for other things. In the server case, and most of what I do, all cores need to be equal.

There are so many use cases, I doubt we could cover them all.

 

[VirtualLarry]

Intel seems to be moving in a foreground/background CPU specific direction while AMD is staying with the technique we're used to which is throw all of the compute you have at all applications.

Do you realize that the Windows NT Scheduler, was designed to give a "priority boost" to the current foreground application, to help with real-time responsiveness. This is irregardless of the presence of big.LITTLE compute hardware.

 

[Hulk]

Do you realize that the Windows NT Scheduler, was designed to give a "priority boost" to the current foreground application, to help with real-time responsiveness. This is irregardless of the presence of big.LITTLE compute hardware.

Yes of course. I was referring specifically to the current behavior of Alder Lake wherein background apps are sent to the E's and the foreground app to the P.

 

[DrMrLordX]

If you're talking about a situation where you have a background application using up a theoretically infinite amount of CPU compute time, then modern operating systems (Windows 10/11, pretty much any modern Linux kernel, and probably MacOS) can seemlessly handle a foreground UI application and something intensive in the background like media encoding or, I dunno, Prime95 or y-cruncher. It's not that hard to do. Mainly:

1). Don't run out of RAM
2). Don't swap to a slow disk if you do run out of RAM
4). Don't set process priority too high to your background application that's hammering all your cores. If you just leave it at "normal" (Windows) then you're fine.

I can see the possibility for conflict if you're trying to do something in the foreground that has some pretty intensive computational requirements, like, I dunno, running a CAD application. If you tried to do that AND Prime95 at the same time (lol) then maybe there would be a problem?

But I've run web browsers and other simple stuff on top of Prime95 before with zero issue. Windows in particular does a really good job of prioritizing foreground tasks, regardless of the CPU. It hurts performance of the background task, but let's face it, if you're trying to multitask at that level, the race-to-completion of your background application can't be that important, can it?

Alder Lake in particular isn't offering you much unique or desirable protection from resource conflict just by walling off background applications to the Gracemont cluster. All it's doing is leaving a lot of Golden Cove compute power left unutilized in situations where you might want those cores to put in work too.