News Intel's Alder Lake CPUs May Not Work With Older Games

[-Fran-]

That would actually be a win scenario for intel because alder cores are still faster than rocket cores so all games would run 1:1 but on faster cores and if that allows the OS to put the smaller cores into sleep mode and push the bigger ones higher then even better.

The loosing scenario would be for devs to use the smaller cores but do so incorrectly because then you could have performance degradation from 11th gen, games could run slower and/or have stutters.

The first part makes sense, especially for games that like fast cores more than many cores. Maybe they won't use slow cores on purpose?

About devs in general, I was thinking this earlier:

Imagine you have 1000 work items to parallelize between 8 cores. That's 125 items per core, and you get optimum performance.
But imagine you have 4 fast cores and 4 slow ones. You can't divide them evenly, otherwise performance will suffer (the fast cores finish first, but you still have to wait for the slow cores). So what is a good distribution of work? How will you do it in code? I see frameworks letting you choose that, but it also means more code to achieve good results.

I'm not mad at Intel for trying to inovate, I'm mad because it feels like a way they found to be competitive against a much better architecture. It feels like they just couldn't design something fast and cool, but their competitor has, so there is no excuse.

Funny how I was typing something along these lines.

But just to add to your reply (which I agree with), I'll also say the CPU still has a shared power budget, so no matter how efficient the small cores are, they still take away from the power budget of the overall CPU, unless Intel manages the two independently, which I doubt? This is to say, if you move load to the small cores, then the big ones will have to slow down somewhat, no? I could be wrong there, but it is still one of the things I want tested.

Other than that, I do have high expectations on the Performance cores; they do sound like being very good, but if you ask me, I would've liked more Performance cores over smaller ones and "efficiency" be damned. A 10c/20t CPU for me sounds more enticing than a 16c(8P8E)/24t CPU. But much like the iGPU discussion, Intel is not going to budge, because of OEMs and trying to entice Apple. Not enthusiasts.

Regards.

 

[TerryLaze]

Imagine you have 1000 work items to parallelize between 8 cores. That's 125 items per core, and you get optimum performance.

That's how video encoding or 3d rendering works but not games but even in this scenario any one of these threads can have stalls or wrongly predicted branches running or whatnot, let alone task switching and cache flushes, it's only optimal in theory and even then you should have one item per core or max one item per thread.

But imagine you have 4 fast cores and 4 slow ones. You can't divide them evenly, otherwise performance will suffer (the fast cores finish first, but you still have to wait for the slow cores).

Games have some small tasks, like reading the user input creating the sounds playing the music making the enemies walk around and so on and some very heavy tasks like rendering all the graphics, no matter if all your cores are the same or if they are different, the easy tasks are always going to finish much faster than the graphics rendering.
Just have a look at this forum, almost every day you will find some poor guy with a very expensive system asking why he is getting stutters.

 

[TerryLaze]

But just to add to your reply (which I agree with), I'll also say the CPU still has a shared power budget, so no matter how efficient the small cores are, they still take away from the power budget of the overall CPU, unless Intel manages the two independently, which I doubt? This is to say, if you move load to the small cores, then the big ones will have to slow down somewhat, no? I could be wrong there, but it is still one of the things I want tested.

Yup, that's why I mention this.
" and if that allows the OS to put the smaller cores into sleep mode and push the bigger ones higher then even better."
Also you can shut the small cores off from bios to get the best clocks for the bigger cores.
And obviously you will still be able to ignore the power limits, like all the benchmarks do, and pump all the power in the world into these CPUs to make all cores run at their respective highest clocks.