Intel confirms Alder Lake-S and Alder Lake-P specs, pushes out optimization guide for...

nanoguy

Posts: 1,021   +14
Staff member
In brief: Intel revealed it its Alder Lake lineup would land sometime in the coming months and seemingly confirmed specifications that have been leaked over the past few months. The company also provided an optimization guide for game developers looking to make the most out of the new hybrid CPU architecture.

On Friday, Intel released its Alder Lake Developer Guide, an in-depth look at how game developers can leverage the new architecture found in the company’s 12th generation Alder Lake processors. In doing so, it also confirmed several details about the upcoming CPU lineup that have leaked over the past few months.

By now, it’s no secret that some Alder Lake CPUs will come with a hybrid architecture that combines P-cores (performance) dubbed Golden Cove and E-cores (efficiency) dubbed Gracemont. However, we also know that Intel will use this configuration on all Alder Lake laptop and desktop parts except on entry-level and mainstream desktop SKUs like the Core i5-12400, which could turn out to be the next value king.

Alder Lake-S processors are aimed at enthusiasts and will feature up to 8 P-cores and 8 E-cores, while mainstream and lower-end variants will only have P-cores. Either way, they’ll all come with an Intel Xe integrated graphics engine with 32 execution units.

Meanwhile, Alder Lake-P parts will include all SKUs that used to be in the U and H subfamilies. Higher-end mobile parts will have 6 P-cores and 8 E-cores, while low voltage parts will combine two P-cores and 8 E-cores. All Alder Lake-P processors will be paired with an Intel Xe graphics engine with 96 execution units.

Only the L3 cache is shared across P-cores and E-cores to allow them to operate separately depending on the workload that’s thrown at them. Intel has chosen to keep the instruction sets all the same for both types of cores, with the notable exception of E-cores, which won’t support AVX512 workloads due to their current architecture. This also means that if a laptop or desktop PC manufacturer chooses to enable E-cores, AVX512 will be disabled on P-cores.

Taking a page from Apple’s book, Intel’s Tread Director (ITD), and Microsoft’s software optimizations in Windows 11 will allow app and game developers to target specific cores depending on the task that needs executing. Hybrid cores will appear functionally identical to the operating system, but the ITD and the OS will govern task allocation to the suitable cores based on special classes that indicate relative performance and efficiency levels.

Game developers will also be able to use Intel’s tools for manual interaction with ITD if they find the OS scheduler does a poor job of discriminating between light and heavy tasks. In fact, there will be a host of optimizations that they can do, such as using E-cores for multi-threaded workloads to prevent maxing out P-core usage or parking E-cores to give more power to the E-cores.

At the same time, setting thread priority and affinity for things like game rendering and audio will be a bit more complicated than on a non-hybrid architecture. This is because E-cores have two logical processors and P-cores do not, and the OS might decide to assign a high-priority task to an E-core simply because it was immediately available. In other words, some apps and games could run slower on systems equipped with an Alder Lake CPU if they don’t have the necessary optimizations in place, even though the OS scheduler should do a decent job in most cases.

"Analysis of games on hybrid architectures has shown the majority of games perform well, with older, or less demanding, games favoring the Performance cores," The authors of the guide point out. "Games that were already built to heavily utilize multithreading, and that can scale to double-digit core counts, were found to benefit from hybrid architecture due to better throughput. However, there are inevitable performance inversions, attributed to either poor multithreading game architectures, poor OS scheduling, or increased threading overhead."

Intel also notes that Microsoft added partial support for ITD in Windows 10, but you’ll have to update to the latest version, 21H2. This means that if you buy or build a new PC and Windows 11 isn’t to your liking, you’ll still enjoy some of the benefits of Intel’s latest CPU architecture.

Permalink to story.

 

Dimitriid

Posts: 1,338   +2,625
I think the part I like the most is Iris Xe on all SKUs. This could actually push into the 5600g and 5700g as a 900p 30 FPS option for gamers while prices remain unreasonable (Which might still be like a full year if not more)

Not that this is really all that useful for anything about the i5 or i7 SKU mind you so not sure why they needed it on "all" probably more to do with the fab process and cut down on faulty chips but well that might be a tiny advantage over the 5950x if you need 16 cores but don't need a GPU and don't mind the 8 slow ones on Alder: I don't expect it to beat the 5950x on all test but it might match the 5900x if they price it competitively.
 

Nanochip

Posts: 50   +58
The developer guide was a good read. Seems to me that Intel wants the game industry to think on and innovate new approaches to multi threading their engines, figuring out what types of tasks are prime for P cores, and what other tasks (e.g., physics, sound, etc.) are prime to be assigned to E cores. In short, Intel seems to be leading the industry towards a paradigm shift in how they think about splitting tasks into threads.

Will be interesting to see how or if the various approaches to threading change over time, and what the performance per watt benefits of a hybrid architecture will be.

Seems like Intel is trying to spark a software innovation in game engine design. Let’s see how it plays out.
 

0dium

Posts: 283   +338
The developer guide was a good read. Seems to me that Intel wants the game industry to think on and innovate new approaches to multi threading their engines, figuring out what types of tasks are prime for P cores, and what other tasks (e.g., physics, sound, etc.) are prime to be assigned to E cores. In short, Intel seems to be leading the industry towards a paradigm shift in how they think about splitting tasks into threads.

Will be interesting to see how or if the various approaches to threading change over time, and what the performance per watt benefits of a hybrid architecture will be.

Seems like Intel is trying to spark a software innovation in game engine design. Let’s see how it plays out.
What? You could multithread your tasks even before that. Moreover didn't need to think which cores to use for which task, all cores were "p" cores on Ryzen CPUs. Where's innovation in making half of your cores perform worse than the others?
 

Kosmoz

Posts: 462   +814
The developer guide was a good read. Seems to me that Intel wants the game industry to think on and innovate new approaches to multi threading their engines, figuring out what types of tasks are prime for P cores, and what other tasks (e.g., physics, sound, etc.) are prime to be assigned to E cores. In short, Intel seems to be leading the industry towards a paradigm shift in how they think about splitting tasks into threads.

Will be interesting to see how or if the various approaches to threading change over time, and what the performance per watt benefits of a hybrid architecture will be.

Seems like Intel is trying to spark a software innovation in game engine design. Let’s see how it plays out.
You can already do that with the CPUs we have now with only big cores, the E-cores introduce just more headaches and a yet unproven energy saving advantage.

If all the leaks are true, Alder Lake might beat Zen3 in gaming, but not actually be more efficient, despite their claims and using half smaller cores... and that would be really bad for them.

1 year later, half the cores are small and have worse efficiency than Zen3, barely beating it... yeah, if that's the case it's a pretty crappy scene for intel.
 

R00sT3R

Posts: 536   +1,517
However, there are inevitable performance inversions, attributed to either poor multithreading game architectures

Kind of concerning, seeing as most game engines older than a few years are exactly this.

I don't think many Ubisoft games, which are still atrociously CPU bound, are going to like this new core arrangement at all.
 

Dimitrios

Posts: 951   +760
Sure notice WINDOWS 11 is cuddling INTEL architecture. AMD's was hampered 2 times with performance the last month from MICROSOFT "mistake".

We should call it "MICROSOFT INSIDE" and have the INTEL's logo with it.

I called this months ago W11 was made to cater INTEL's new architecture and to surpass AMD.
 

Irata

Posts: 1,812   +3,053
What? You could multithread your tasks even before that. Moreover didn't need to think which cores to use for which task, all cores were "p" cores on Ryzen CPUs. Where's innovation in making half of your cores perform worse than the others?
This is something I don‘t get - why would a hybrid core architecture be / perform better than several identical smt capable cores ?

What I could see potentially is if Intel says this allows budget / lower power multi core solutions vs using full cores, I.e. a 10C16T 12600K offers the same gaming performance at lower price and power consumption vs an 8C16T 5800X.
 

hahahanoobs

Posts: 3,869   +1,920
@Irata
Why even develop desktop Alder Lake chips with efficiency cores,? “You’re absolutely right—battery life doesn’t matter in desktop,” said Stephen Robinson, a CPU architect and Intel fellow. “But the thermals do. Fans, cooling power—at some point, you have a limit.”

And that’s why Intel designed its new efficiency cores.

The new E-Core also delivers 40 percent more performance than Skylake. If you put four E-Cores against a dual core Skylake system using four threads, you’d still get 80 percent more performance with less power, Robinson said. “We exceed Skylake Core performance by consuming less power in a smaller footprint,” he added.
 

Nanochip

Posts: 50   +58
You can already do that with the CPUs we have now with only big cores, the E-cores introduce just more headaches and a yet unproven energy saving advantage.

If all the leaks are true, Alder Lake might beat Zen3 in gaming, but not actually be more efficient, despite their claims and using half smaller cores... and that would be really bad for them.

1 year later, half the cores are small and have worse efficiency than Zen3, barely beating it... yeah, if that's the case it's a pretty crappy scene for intel.

But how efficient are current approaches ? What if some programmers are leaving performance on the table by using inefficient threading algorithms? Multithreading is hard because it’s not obvious at programming time how the threads will be scheduled, dispatched, interrupted, etc during real time. How to properly load balance so as to not starve or over load the cpu? It takes a lot of profiling and debugging work to get it right. A lot of colleges teaching computer science barely even touch on optimal multi threading approaches (at least that’s how it was when I was still in college, I don’t know if that has changed).

My point is it’s good that Intel is having a conversation with developers on optimizing their engines. You may or may not agree and that’s fine. But I ask, are optimizations of current threading approaches possible, and if so, how much performance is being left on the table? The only way to know is to have the conversation, investigate, and then innovate. If AMD and not intel did this, I suspect there wouldn’t be as much hatred or skepticism.
 

HardReset

Posts: 1,323   +985
@Irata
Why even develop desktop Alder Lake chips with efficiency cores,? “You’re absolutely right—battery life doesn’t matter in desktop,” said Stephen Robinson, a CPU architect and Intel fellow. “But the thermals do. Fans, cooling power—at some point, you have a limit.”

And that’s why Intel designed its new efficiency cores.
Intel designed new efficiency cores because Alder Lake power consumption is so high that Intel could only match Zen3 thread count with smaller cores. That's about only reason.
The new E-Core also delivers 40 percent more performance than Skylake. If you put four E-Cores against a dual core Skylake system using four threads, you’d still get 80 percent more performance with less power, Robinson said. “We exceed Skylake Core performance by consuming less power in a smaller footprint,” he added.
Yeah right if you take 2015 Skylake with 14nm without any +++++++...
 

defaultluser

Posts: 280   +243
@Irata
And that’s why Intel designed its new efficiency cores.

The new E-Core also delivers 40 percent more performance than Skylake. If you put four E-Cores against a dual core Skylake system using four threads, you’d still get 80 percent more performance with less power, Robinson said. “We exceed Skylake Core performance by consuming less power in a smaller footprint,” he added.

They only created this monstrosity because unlike AMD, they have no idea to make the big cores sip power (the cores have ti be factory-overclocked to have any chance of exceeding AMD, which means double the idle)!

But yes, these messy cores will soon launch p mobile

And now Microsoft is permanently suckered-into making all this magically work (when it doesn't even work that well on Android Systems!)

If this ever catches on, AMD could do this easily by adding undercloced cores from the previous gen (with hyperthreading disabled, to cut power cache required). Zen 2 4-core single ccx >> the performance of 8 Tremont hybrid cores
 
Last edited:

quadibloc

Posts: 324   +211
I am very distressed to hear that the only way to enable AVX-512 is to disable the E cores. Of course, that has to be true for any operating system that doesn't support marking those tasks which use AVX-512, so that the scheduler can avoid assigning them to the E cores - but I would have expected that feature to be included in an update once the specs for these chips were released.
 

MaxSmarties

Posts: 507   +295
Leaked benchmarks put the E core-less i5 12400 ahead of the 5600X. It's good to see the E-Cores aren't just a crutch.
Things are heating up!
"leaked" by Intel in an highly controlled way... We already saw this with Rocket Lake. Things were different when the CPUs reached the first reviews.
 

MaxSmarties

Posts: 507   +295
I am very distressed to hear that the only way to enable AVX-512 is to disable the E cores. Of course, that has to be true for any operating system that doesn't support marking those tasks which use AVX-512, so that the scheduler can avoid assigning them to the E cores - but I would have expected that feature to be included in an update once the specs for these chips were released.
the real question is: are AVX512 really relevant today, or just a marketing gimmick ? And I'm not entirely sure about the answer
 

hahahanoobs

Posts: 3,869   +1,920
They only created this monstrosity because unlike AMD, they have no idea to make the big cores sip power (the cores have ti be factory-overclocked to have any chance of exceeding AMD, which means double the idle)!

But yes, these messy cores will soon launch p mobile

And now Microsoft is permanently suckered-into making all this magically work (when it doesn't even work that well on Android Systems!)

If this ever catches on, AMD could do this easily by adding undercloced cores from the previous gen (with hyperthreading disabled, to cut power cache required). Zen 2 4-core single ccx >> the performance of 8 Tremont hybrid cores
AMD is no leader in sipping power. Oh you mean TSMC 7nm vs Intel 14nm? Bring that up again after ADL and then Meteor Lake when it's more of a fair fight. Competition remember? Intel isn't done. Far from it.

AMD Zen has no counter to ADL E-cores or even include an IGP in their main desktop part for that matter. Remember Fusion? Still waiting....

Zen flagships were the lower clocked vs their x600 parts and the 2800(X) never came to be. AMD could not raise core clocks and that's why Zen and Zen+ were meh with just a bunch of cores. 4 fast cores will satisfy more people than having 8 slow ones. To make up for the poor clock for clock perf, everyone started talking about what you could do with more cores. lol Trash defense. For the majority that were buying Intel 4C/8T parts, they had no use for more cores for what they were doing with their computers. More cores didn't help in gaming or day to day use and if they really wanted more cores there was the HEDT lineup at the time.

AMD is treated like a kid with a disability. Nothing they do is wrong enough for someone to say hardly anything negative. Even when they do the smallest thing well, they get sooooo much praise for it! It's insane. And it bothers me greatly.

That's the only reason I publicly defend Intel.
 

hahahanoobs

Posts: 3,869   +1,920
"leaked" by Intel in an highly controlled way... We already saw this with Rocket Lake. Things were different when the CPUs reached the first reviews.
Can I see the RKL leaks you're talking about? I don't remember seeing the same level of RKL leaks you're referring to.
 

Aryassen

Posts: 188   +222
I think it was right here a few months back, while Steve and Tim investigated how the number of CPU cores affect gaming performance, when it was said that there is no such thing as a certain task dedicated to a certain core or thread: what you actually need is a certain amount of CPU power, and if a given CPU is reaching that level, it doesn't really matter how many cores it needs to achieve that feat.

I still don't understand why complicating things with multiple core tiers is good on desktop (in a laptop, it may make some sense, though even there I'm not sold on the idea). Developers have well-established methods already to spread workload, they don't need a new type of cores to do that..
 

jpuroila

Posts: 389   +236
AMD is no leader in sipping power. Oh you mean TSMC 7nm vs Intel 14nm? Bring that up again after ADL and then Meteor Lake when it's more of a fair fight. Competition remember? Intel isn't done. Far from it.

AMD Zen has no counter to ADL E-cores or even include an IGP in their main desktop part for that matter. Remember Fusion? Still waiting....

Zen flagships were the lower clocked vs their x600 parts and the 2800(X) never came to be. AMD could not raise core clocks and that's why Zen and Zen+ were meh with just a bunch of cores. 4 fast cores will satisfy more people than having 8 slow ones. To make up for the poor clock for clock perf, everyone started talking about what you could do with more cores. lol Trash defense. For the majority that were buying Intel 4C/8T parts, they had no use for more cores for what they were doing with their computers. More cores didn't help in gaming or day to day use and if they really wanted more cores there was the HEDT lineup at the time.

AMD is treated like a kid with a disability. Nothing they do is wrong enough for someone to say hardly anything negative. Even when they do the smallest thing well, they get sooooo much praise for it! It's insane. And it bothers me greatly.

That's the only reason I publicly defend Intel.
5800u outperforms anything Intel has below desktop-class, and that's with 25w TDP (and unlike Intel, AMD's TDP actually means something). AMD doesn't NEED anything to counter ADL E-cores.

Let's take a look at Zen flagships, shall we? 1800x vs 1600x: 3.6/4.0 vs 3.6/4.0. 2700x vs 2600x: 3.7/4.3 vs 3.7/4.2. 3800x vs 3600x: 3.9/4.5 vs 3.8/4.4. And so on.

AMD isn't "treated like a kid with a disability", Intel just really IS that bad and you're full of ****.
 

hahahanoobs

Posts: 3,869   +1,920
5800u outperforms anything Intel has below desktop-class, and that's with 25w TDP (and unlike Intel, AMD's TDP actually means something). AMD doesn't NEED anything to counter ADL E-cores.

Let's take a look at Zen flagships, shall we? 1800x vs 1600x: 3.6/4.0 vs 3.6/4.0. 2700x vs 2600x: 3.7/4.3 vs 3.7/4.2. 3800x vs 3600x: 3.9/4.5 vs 3.8/4.4. And so on.

AMD isn't "treated like a kid with a disability", Intel just really IS that bad and you're full of ****.
No 2800X....
Early Zen was backwards with flagships having lowest clocks because more cores plus high clocks was a no go.

25w isn't desktop. Stay on topic.
5800X is over $200 more than 5600X in Canada. $150ish in the states? 11400 is budget king. AMD has nothing to counter. Overpriced APUs. Mobile flagships were a mess. Launched products no one could buy (1600AF, 3300X, mobile flagships).

AMD's "longer lasting" sockets aren't better. That's a myth. Their chipsets and upgrades have been weak. They are on par with Intel at best. You don't need to upgrade Intel as fast you do with AMD. Maybe things will change with Zen 4 and beyond.

Zen 3 is good, but I personally think AMD took too long while Intel struggled with 10nm. Now the game is on and AMD has a strong Intel to deal with. Good luck to both sides!