Intel Alder Lake-S (12th Gen) Hybrid ES CPU Leaks in Geekbench and SiSoftware databases: Flaunting 16 cores & DDR5-4800 memory (Updated)

[BiG StroOnZ]

Intel will introduce a hybrid CPU architecture codenamed Alder Lake that is built on a 10nm node. An apparent engineering sample (with 1.4GHz base clock) has found its way to Geekbench. While not exactly the same concept, the go-to comparison for Alder Lake is Arm's big.LITTLE design. Alder Lake will combine high-performance cores based on Intel's Golden Cove architecture, with power efficient cores based on Gracemont. In essence, Golden Cove would serve as the 'big' cores and Gracemont would serve as the 'small' cores. This could lead to some interesting configurations, whereby the number of threads will not necessarily be double the number of the total cores, depending on the SKU and arrangement.

 

Quote

According to the listing, this unnamed Alder Lake-S processor (it's identified simply as "Intel 0000") sports 16 cores and 24 threads. That may or may not be accurate, but it does reinforce the notion that threads will not necessarily be double the number of total cores. Assuming it is accurate, however, we are looking at 8 Golden Cove cores with 16 threads, paired with 8 Gracemont cores with 8 threads. The Geekbench 5 results list it with 16 cores/24 threads running at 1.4GHz, consistent with a previous sighting on the SiSoft Sandra benchmarking database back in October. The Alder Lake CPU scored 996 in single-core tests and 6931 for multi-core. Performance is nothing to get excited about, but keep in mind that this is a very early engineering sample. Intel's production die is likely to have a much higher clock speed. The CPU also features 30 MB of L3 cache and 12.5 MB of L2 cache. The iGPU has 256 shader cores at 1.15 GHz running on Intel's Xe LP graphics engine.

Source 1: https://www.tomshardware.com/news/intel-12th-generation-alder-lake-s-cpu-multi-core-performance-ryzen-5-3600x

Source 2: https://videocardz.com/newz/intel-alder-lake-s-16-core-and-24-thread-cpu-appears-on-geekbench

Source 3: https://www.techspot.com/news/88110-intel-10nm-alder-lake-s-cpu-engineering-sample.html

Source 4: https://hothardware.com/news/intel-alder-lake-s-hybrid-cpu-leaks-16-cores-24-threads

 

While the performance is nothing to get overly excited about, it is worth noting that being that it is an early ES chip, and currently clocking at only a measly 1.4GHz (and probably not much higher than that, as the reported boost clock is definitely incorrect). The single-core performance is practically the same as a mobile Ryzen 5 4600H @ 994 average points; a 6c12t CPU that boosts to 4.0GHz with a base clock of 3.0GHz (nearly 114% increase in clock speed compared to this Alder Lake ES sample). The multi-core performance is about equal to a Ryzen 5 3600X @ 6,906 average points; a 6c12t CPU that boosts to 4.4GHz with a base clock of 3.8GHz (nearly 171% increase in clock speed compared to this Alder Lake ES sample). Within that specific context, knowing that finalized clocks on the 12th Gen Alder Lake parts will be much higher, I would say that these early results/numbers are actually relatively notable, for the most part.

 

Let us for example extrapolate performance real quick: equalizing the multi-core performance by applying base clock parity between the Alder Lake ES chip and the Ryzen 5 3600X; that would push the result of the Alder Lake ES Chip to around 18,800 points. Making its multi-core performance more in line with a Ryzen 9 5950X (a 16c32t part). In actuality, the Alder Lake chip would be faster on average compared to the 5950X, and that was with only a 3.8GHz clock speed enhancement to the Alder Lake CPU (compared to the current 1.38GHz number). 

 

Minor, but interesting update to this story ~ 

 

Quote

A never-before-seen Intel (12th Gen) Alder Lake-S processor has been discovered on the SiSoftware website by @momomo_us. Along with once again confirming support for DDR5 memory (@ 4800MHz), the listing gave some details about the Xe Gen 12.2 iGPU and revealed 30 MB L3 cache. Clock rates were measured @ 1.8 GHz to 4.0 GHz.

 

 

 

The new sample that was discovered today has a higher clock speed though. With a base clock of 1.8 GHz and a boost of 4.0 GHz, we are looking at a faster CPU. The hybrid chip seemingly sports the same 16-core, 32-thread configuration. SiSoftware detects the Alder Lake-S part as a 16-core part, meaning there are eight Golden Cove cores and eight Gracemont cores onboard. The new Alder Lake-S sample reportedly features a 1.8 GHz base clock, 400 MHz higher than the previous leaked sample. It's uncertain if the two are the same processor or if Intel has managed to improve the previous chip's base clock. On this occasion, the software was able to pick up the processor's boost clock speed that's apparently configured to 4 GHz. The cache configuration remains unaltered: We still see the 12.5MB of L2 cache and 30MB of L3 cache. 

 

Surprisingly, the 1.4 GHz Alder Lake-S chip delivers a 21% higher score on the processor multi-media test than the 1.8 GHz sample. It's plausible that SiSoftware's Sandra benchmark software isn't yet fully optimized for Alder Lake-S. We shouldn't lose sight of the fact that Alder Lake-S arrives with a hybrid setup, so the whole high-performance and high-efficiency core arrangement can trip some software up. The 1.8 GHz Alder Lake-S part does score better in some of the other tests though. The memory bandwidth test is one of the benchmarks that stand out the most. There's an improvement up to a whopping 153.6%. The logical reasoning behind the substantial uplift is that the previous Alder Lake-S was paired with DDR4 memory, while the new sample purportedly runs with DDR5 memory.

 

Source 5: https://www.notebookcheck.net/16-core-Intel-Alder-Lake-S-desktop-processor-s-SiSoftware-entry-reveals-specs-Xe-Gen-12-2-iGPU-details-30-MB-L3-cache-and-DDR5-memory-support.515770.0.html

 

Source 6: https://videocardz.com/newz/intel-alder-lake-s-processor-with-16-cores-at-4-ghz-and-ddr5-4800-memory-spotted

 

Source 7: https://www.tomshardware.com/news/alder-lake-s-cpu-hits-4-ghz-with-ddr5-memory

[TempestCatto]

Competition makes my insides all fuzzy. Can't wait to see what's next! 

[thechinchinsong]

Cautiously optimistic, but I'm willing to bet this isn't gonna be relevant until Zen 4 is out.

[captain_to_fire]

Inb4 Apple launches desktops with 16 cores on the base and 32 cores for the top of the line 

[saltycaramel]

MacMini M1 late 2020 (4 big + 4 little)

single: 1700

multi: 7400

 

Alder Lake-S late 2021 (8big + 8 little)

single: 1000

multi: 7000

 

 

 

[leadeater]

 

Oh really now? Hmmmm 

[leadeater]

7 minutes ago, saltycaramel said:

MacMini M1 late 2020 (4 big + 4 little)

single: 1700

multi: 7400

 

Alder Lake-S late 2021 (8big + 8 little)

single: 1000

multi: 7000

 

Quote

While the performance is nothing to get overly excited about, it is worth noting that being that it is an early ES chip, and currently clocking at only a measly 1.4GHz

 

Probably want to hold your judgement when dealing with ES samples, especially one that is clearly very early and not a QS sample.

[WhitetailAni]

Instead of "Genuine Intel (R) CPU 0000", they are "We promise it's an Intel (R) CPU".

 

 

[porina]

The big danger about using results like this to extrapolate performance is you need to know at least two things pretty well.

1, how does the benchmark scale? - this could be tested and understood now. I think I might have done similar with an earlier version as part of a benching competition, but doubt I'll find it again.

2, what clock is it really running? If you don't know this, errors and uncertainty will likely be far greater than any difference we're looking for.

[CarlBar]

7 hours ago, leadeater said:

 

 

Probably want to hold your judgement when dealing with ES samples, especially one that is clearly very early and not a QS sample.

 

Not just that, at best this is going to be on intel 10nm which is still only equivalent to TSMC 7nm which is two whole generations behind the node apple is using. the M1 for example has 16 billion transistors, a 4800h only has 9.8 billion. We don';t know how many this Intel sample has, but it probably isn't comparable to the M1. Thats going to be another important metric to keep in mind. Whilst performance won't scale linearly with transistor count, it does provide a hard physical limit on scaling and can probably be used for some very broad and loose predictions. If Intel gets this chip even within shouting distance of the M1, they're going to be doing pretty freaking amazing.

[NunoLava1998]

..wait, why does it have 24 threads and not 32 on a 16-core processor? 

[AlexGoesHigh]

15 minutes ago, NunoLava1998 said:

..wait, why does it have 24 threads and not 32 on a 16-core processor? 

Just taking a guess but the smaller cores could have hyperthreading off

[Bombastinator]

11 hours ago, TempestCatto said:

Competition makes my insides all fuzzy. Can't wait to see what's next! 

Sounds fatal.  I do not want fuzzy insides. 

[porina]

2 hours ago, CarlBar said:

at best this is going to be on intel 10nm which is still only equivalent to TSMC 7nm which is two whole generations behind the node apple is using.

It's still one node give or take, unless you're counting improvements within a node as a generation. By that count, how many 14nm generations have Intel had  

 

2 hours ago, CarlBar said:

If Intel gets this chip even within shouting distance of the M1, they're going to be doing pretty freaking amazing.

The M1 is not a competitor unless you limit it to the scenarios the M1 is good at. In other words, the M1 is best at doing what the M1 does. I don't feel Alder Lake was designed with that in mind.

 

55 minutes ago, NunoLava1998 said:

..wait, why does it have 24 threads and not 32 on a 16-core processor? 

The small cores do not have HT.

[saltycaramel]

2 hours ago, porina said:

The M1 is not a competitor unless you limit it to the scenarios the M1 is good at. In other words, the M1 is best at doing what the M1 does. I don't feel Alder Lake was designed with that in mind.

 

 

And those scenarios would be? I thought we were past “ARM is good only at..”..

 

I’ll add that by the second half of 2021 Apple will have M2 desktop CPUs on the 5nm+ node out...these early Alder Lake samples may not be representatives of the final CPUs but that could be a big gap to fill...

 

Also, supposedly Alder Lake S parts are targeting a 80-125W TDP...the M1 is a mobile part with much less than that...what Apple will be able to accomplish in an 80-125W desktop thermal envelope could be earth shattering by the look of it..

 

[porina]

4 minutes ago, saltycaramel said:

And those scenarios would be? I thought we were past “ARM is good only at..”..

If you look at my other posts on the topic you'll find my position is I do see ARM as being in a good position to take over the desktop space also. Doesn't mean it will, but certainly this may be the break out of mobile space.

 

I think Apple with M1 are in a position similar to where AMD were with original Ryzen launch. On the surface it looks great, but once you dig into deeper it is less clear. I wouldn't say AMD passed Intel Skylake architecture overall until Zen 2, as original Zen had significant weaknesses in many areas. With the M1 it is much harder to do comparisons. The ever misused Cinebench is there. Apple have certainly got some good performance out of it in demonstrated use cases, but until more software is natively coded for it we wont have a good picture of what it will really do.

 

4 minutes ago, saltycaramel said:

Also, supposedly Alder Lake S parts are targeting a 80-125W TDP...the M1 is a mobile part with much less than that...what Apple will be able to accomplish in an 80-125W desktop thermal envelope could be earth shattering by the look of it..

Desktop CPUs are not really power restricted, so power efficiency hasn't been the priority. Power scaling is far from linear, unless you do it by going wide at increased silicon cost, and take the associated potential problems with software scaling capability. Certainly it will fail much faster if you are doing it via clock. They will have to balance both as and when they go further up.

[leadeater]

6 hours ago, CarlBar said:

 

Not just that, at best this is going to be on intel 10nm which is still only equivalent to TSMC 7nm which is two whole generations behind the node apple is using. the M1 for example has 16 billion transistors, a 4800h only has 9.8 billion. We don';t know how many this Intel sample has, but it probably isn't comparable to the M1. Thats going to be another important metric to keep in mind. Whilst performance won't scale linearly with transistor count, it does provide a hard physical limit on scaling and can probably be used for some very broad and loose predictions. If Intel gets this chip even within shouting distance of the M1, they're going to be doing pretty freaking amazing.

M1 has a reasonably big GPU and ML engine in it, I wouldn't look at transistor count much here. It's not like current Intel CPUs are that far off so it'll be no shocker at all if single thread is on par and then everything else faster due to more cores.

[CarlBar]

6 hours ago, porina said:

It's still one node give or take, unless you're counting improvements within a node as a generation. By that count, how many 14nm generations have Intel had  

 

 

Intels latest 14nm nodes are actually ahead of their planned 10nm performance. I'd have to dig around to be sure if my memory is acurratte but i seem to recall 7nm+ is supposed to have about the same perf per watt advantage over 7nm as 5nm does over 7nm+.I belive it works out at a cumulative 60% perf per watt between 7nm and 5nm.

 

6 hours ago, porina said:

The M1 is not a competitor unless you limit it to the scenarios the M1 is good at. In other words, the M1 is best at doing what the M1 does. I don't feel Alder Lake was designed with that in mind.

 

I know, my point is that without a similar transistor count, (or an apple architecture that makes very bad use of their available transistors), it's unrealistic top expect this to matchup even without process node factors.

 

3 hours ago, saltycaramel said:

Also, supposedly Alder Lake S parts are targeting a 80-125W TDP...the M1 is a mobile part with much less than that...what Apple will be able to accomplish in an 80-125W desktop thermal envelope could be earth shattering by the look of it..

 

Not really. @leadeater pointed this out in the original 1 thread but having so little IO, such limited memory options, and no need for internal addon devices beyond that, (like a DGPU), saves apple a bunch on the power usage. Also Desktop vs laptop typically have very different silicon quality level to desktops. And that very strongly affects the performance per watt. Also mobile parts, and the M1 is no different, generally use lower clocks to get further power savings. And the effect can be very significant for a very small performance hit. Don't expect apple to crank up power usage 6 fold and get 6 fold the performance out of it. In fact some architectures don;t scale up that well. That was AMD's issue with Polaris. The frequencies it worked at best gave very middling performance, so they had to crank it up to keep relevant and that destroyed the power efficiency.

 

2 hours ago, leadeater said:

M1 has a reasonably big GPU and ML engine in it, I wouldn't look at transistor count much here. It's not like current Intel CPUs are that far off so it'll be no shocker at all if single thread is on par and then everything else faster due to more cores.

 

Well the 4700H has a iGPU too. I do know the issues of comparison though, thats why i carefully hedged my bets in the wording, there's a lot of factors and without a detailed breakdown of numbers it's hard to get an exact comparison.

 

As an aside where are you getting the inel trasistor counts from. I just did a search and got nadda.

[saltycaramel]

Let’s go back to why Apple Silicon is being brought up here in the first place: Alder Lake is the first (after Lakefield) serious hybrid CPU family by Intel, seen by many as an answer to ARM big.LITTLE CPUs. 

 

Apples to apples (in the same thermal envelope), all the evidence we have so far points to the upcoming Apple desktop CPUs beating Alder Lake S. Of course, it’s just speculation at this point.

 

Of one thing I’m pretty sure: Apple will not, under any circumstance, release an Apple Silicon Mac that doesn’t run circles around its direct Intel-based predecessor. That’s a given, that’s how this transition works. So:

- the M1T (probably 12 big + 4 little) in the spring 2021 iMac will run circles around the current 10-core i9-10910 Comet Lake in the 2020 iMac

- the M2T (probably 16 big + 4 little) in the fall 2021 iMac Pro will run circles around the 18-core Xeon W-2109B Skylake-W in the 2017 iMac Pro

 

Make of that what you wish, in comparison to Alder Lake S.

[leadeater]

5 hours ago, CarlBar said:

Also Desktop vs laptop typically have very different silicon quality level to desktops.

That and personally, and I know this applies to the majority as well, power usages is largely a non factor for the CPU. Unless it's excessively high, higher than 10900K, then it doesn't even come in to the decision making equation at all, it really is a superfluous variable on that platform. Like that doesn't take away from technical achievements of architectures and nodes being more power efficient than others and I love those discussions, but I don't bring that thinking or factors over to my purchasing decisions when buying a desktop CPU.

 

When it comes to laptops and mobile devices power is a much larger factor, in a way more so than performance. That's why I don't particularly care for those extreme laptops that have desktop CPUs in them, highly unpractical. Like it's sort of cool you can do that, but a Ryzen 4900HS makes it almost redundant unless you're putting a 3950X/5950X in there and all I can say to that is good luck using it as a laptop.

 

5 hours ago, CarlBar said:

As an aside where are you getting the inel trasistor counts from. I just did a search and got nadda.

Think you might have misunderstood? My comment about that was single core performance. Intel has some larger gains in the pipeline and while I don't know how close it will be on mobile variants at lower power it'll be better than what they have now and for desktop they do what all desktop CPUs do, raise the package power limit to near the stability cut off because there is no reason not to. Competition requires that they do this, limiting performance to stay within node optimal efficiency just is not attractive for desktop parts.

 

And this topic is about Alder Lake-S, -S which is the codename suffix for desktop means we are talking about desktop CPUs with desktop power budgets so we can all for sure know power efficiency is not top of the list. That's why I wont be surprise to see it above the M1 for single core and it's pretty much guaranteed to be higher for multi core (otherwise Intel is regressing).

 

Back to that whole power thing, 4700U 15W (Zen 2) does CB R20 Single/Multi @ ~464/~2500 and a 3600X 128W PPT (Zen 2) does ~497/~3700. It should be very obvious from this 50% ish gain for 8.5 times the power (for multi) just how power scaling works, this is a node trait more than anything else, and how little it matters. So Apple is not immune to this either, a 120W Apple CPU will equally suffer these poor performance gains (for single core) for the power increase, like every other x86, Power and ARM CPU have. Unless Apple find a way to defy physics then that will indeed be earthshattering, otherwise expect the expected.

 

Personally I think Apple is happy with the clocks and single core performance as it is and will not touch that, adding more cores is what I expect to see. So it should be fairly trivial to math out large core count configurations (CB R23):

• 8 Core: 13921 (~48W)
• 12: 20009 (72W)

Ryzen 5950X is ~28500 (~120W) for reference, 5900X is ~21800 (~130W). An Apple CPU with ~120W would be 20 cores with CB R23 Multi around 32,200. At least I do not think 13% better performance for the same power is earthshattering, maybe that's just me. It's better than what Intel usually gives per generation though lol.

[Fnige]

19 hours ago, leadeater said:

 

Oh really now? Hmmmm 

[CarlBar]

@leadeater Yeah i think there was a misunderstanding on the Intel transistor count thing DoH. My main reason personally for looking at power saving per node side of things was mainly because in desktop space thermal constrains tend to get in your way a lot. So having a more power efficient node leaves you more potential in the bag, though ofc ideally you use the node to run the existing spec cooler and then use the extra transistor count headroom in the same power budget to make your cores wider and/or more numerous. 

[Dwagon]

Intel here with their current specs have 8 single threaded cores and 8 hyper-threaded cores. Hence, the 16c/24t. What Intel has done is adapt their Lakefield architecture which is used for their mobile chips. Once again Intel is taking one of their mobile chips/architecture and is adapting it for desktop. *cough* Pentium 4 Netburst failure *cough*

 

https://www.pcgamer.com/pc-tech-2021/#section-intel-alder-lake

[Vishera]

On 12/30/2020 at 3:50 AM, BiG StroOnZ said:

and currently clocking at only a measly 1.4GHz

Does it boost to 17.6 GHz as well?

In all honestly it's too early to get any meaningful information about the performance of the final product.

Anyone here has a itch for a new motherboard?,because Alder Lake is LGA 1700,

I think that socketable chipsets would be significantly more efficient than making motherboards obsolete for no reason.

Think about the environment and E-waste,pretty sure that there are many 5 years old and "younger" motherboards fully capable of running those chips.

[Falkentyne]

39 minutes ago, Dwagon said:

Intel here with their current specs have 8 single threaded cores and 8 hyper-threaded cores. Hence, the 16c/24t. What Intel has done is adapt their Lakefield architecture which is used for their mobile chips. Once again Intel is taking one of their mobile chips/architecture and is adapting it for desktop. *cough* Pentium 4 Netburst failure *cough*

 

https://www.pcgamer.com/pc-tech-2021/#section-intel-alder-lake

I believe you're getting your facts wrong.

They did that with *Conroe*, which was first based off the mobile offering (Don't remember the name)  and gave us Core and Core 2 Duo, which blew the socks off everyone and destroyed AMD's offerings.  Core 2 had double the IPC of a Pentium 4 at the same clock, IIRC.

 

Netburst was before this.