Intel's Next-Gen 16-core/32-thread Alder Lake Processor appears online and sporting a new "Hybrid Chip Layout / Design"

[BiG StroOnZ]

 

Alder Lake is slated to be Intel's most significant desktop CPU overhaul in nearly a decade.

 

Quote

 

 

 

 

Quote

 

Alder Lake is a departure from the expected for Intel, and we're still not entirely sure what its hybrid approach will mean for desktop CPU performance or gaming. But today we're at least that little bit closer to finding out, as Alder Lake has appeared on a benchmarking database ahead of its debut next year. An Alder Lake S chip has been spotted on the famously leaky SiSoft Sandra benchmarking database by the eagle-eyed TUM_APISAK over on Twitter. The listing describes a 16-core Alder Lake desktop chip with multiple threads per core, bringing the thread total up to 32, and accompanied by 30MB of L3 cache and 10x 1.25MB L2.

 

Alder Lake is likely to be comprised of eight Golden Cove cores (big ones) and eight Gracemont cores (little ones) in its largest configuration. It's this hybrid architecture that differentiates it from today's Comet Lake 10th Gen, which is made up of just one Skylake-derivative architecture. Alder Lake S was suspected of being the first to arrive with support for DDR5 memory, the likes of which SK Hynix just today announced was ready to go. The listing on SiSoft Sandra, however, shows only support for 'DRR4', which is sure to be a typo or some other misreporting error. It looks like we'll have to wait a little longer for confirmation of DDR5's debut. The use of the Sabrent Rocket 4.0 NVMe SSD drive does at least allude to PCIe 4.0 compatibility with Alder Lake, which would see Intel gain level-footing with AMD Ryzen in the platform bandwidth department. It has also been rumored that PCIe 5.0 support would be introduced with Alder Lake, so perhaps even this more-than-capable drive is on the slow side for the new platform. Alder Lake is set for release in 2021.

 

Source 1: https://hothardware.com/news/intel-alder-lake-s-16-core-32-thread-hybrid-cpu-leak 

Source 2: https://www.pcgamer.com/intels-next-gen-16-core32-thread-alder-lake-chip-appears-online/  

Source 3: https://www.dsogaming.com/news/12th-gen-intel-alder-lake-s-cpu-reportedly-spotted-in-sisoftware-benchmark-database/

 

 

This is certainly an interesting CPU design, to say the least. Since there are no clear-cut demonstrations of whether or not this particular chip layout will be more beneficial for desktop, I'm really looking forward to seeing more data on its effectiveness in the near future. We do somewhat know it will certainly be far more efficient, highlighting performance per watt metrics. However, like everyone else, we want to see how it performs in games. As well as whether or not Intel aims to entirely replace its desktop lineup with this new and fresh hybrid solution. 

 

Quote

 

Spoiler

 

 

 

 

 

 

[porina]

Configuration options were leaked months ago. I'm still not sure I understand the value of "slow" cores on desktop in addition to "fast" cores. I think this is mainly a cheap way to increase core counts above 8 for mainstream.

[RejZoR]

This is going to suck for the normies who are just getting on the high core count thing. Oh it's 16c/32t. But it's actually really weak.

[porina]

24 minutes ago, RejZoR said:

This is going to suck for the normies who are just getting on the high core count thing. Oh it's 16c/32t. But it's actually really weak.

It will certainly be interesting to see how this works in the real world once it is out in over half a year's time. I'd argue the scenarios where a user genuinely needs 12 or 16 "fast" cores are limited. A good test is if a user knows why they need a higher end configuration, they should get it, otherwise, it will be more because they can than any need. Never hurts to go a bit overkill right?

 

One way I see this playing out is that an Intel 16 core would cost less than an AMD 16 core (Zen 4?). The AMD CPU would be faster at high thread scaling workloads, and Intel will fight in more mixed workload areas that don't use so many threads. 

[RejZoR]

29 minutes ago, porina said:

It will certainly be interesting to see how this works in the real world once it is out in over half a year's time. I'd argue the scenarios where a user genuinely needs 12 or 16 "fast" cores are limited. A good test is if a user knows why they need a higher end configuration, they should get it, otherwise, it will be more because they can than any need. Never hurts to go a bit overkill right?

 

One way I see this playing out is that an Intel 16 core would cost less than an AMD 16 core (Zen 4?). The AMD CPU would be faster at high thread scaling workloads, and Intel will fight in more mixed workload areas that don't use so many threads. 

No, my worry is when Intel's CU will be 16c/32t with 8 high speed cores and 8 low speed ones and AMD's in same 16c/32t will all be full fat ones. People will just look at numbers and price and think they're getting best bang in the world. In reality it'll be nowhere as fast. But if it'll be marketed correctly and people will understand it right, these will be pretty interesting. Assuming load will be balanced on them correctly.

[Arika]

1 hour ago, porina said:

Configuration options were leaked months ago. I'm still not sure I understand the value of "slow" cores on desktop in addition to "fast" cores.

on desktop i agree, but on laptops, it would be amazing for battery life when doing regular tasks

[Loote]

10 minutes ago, RejZoR said:

No, my worry is when Intel's CU will be 16c/32t with 8 high speed cores and 8 low speed ones and AMD's in same 16c/32t will all be full fat ones. People will just look at numbers and price and think they're getting best bang in the world. In reality it'll be nowhere as fast. But if it'll be marketed correctly and people will understand it right, these will be pretty interesting. Assuming load will be balanced on them correctly.

The same things exists in phones and... it depends. Some just count cores, or cores and frequencies, but others only care about Snapdragon 8xx. You just need a mindset of 'AMD is faster'(if it will be) and most people will follow it.

[porina]

10 minutes ago, RejZoR said:

No, my worry is when Intel's CU will be 16c/32t with 8 high speed cores and 8 low speed ones and AMD's in same 16c/32t will all be full fat ones. People will just look at numbers and price and think they're getting best bang in the world. In reality it'll be nowhere as fast. But if it'll be marketed correctly and people will understand it right, these will be pretty interesting. Assuming load will be balanced on them correctly.

Products have been getting more complicated for a while. You can't just look solely at a Cinebench score as a measure of processor goodness as it neglects pretty much everything else in the system outside of the CPU. Where I think we'll land is that 8 "fast" Intel cores will be competitive with 8 AMD cores, but the hybrid 16 cores will not match 16 fast cores. That's pretty much a given. For many actual more complex workloads, it will be much more complicated than that. I'm still kinda annoyed we have 16 fast cores on a consumer platform, I think they deserve better as it is really unbalanced unless you run light worloads like Cinebench.

 

4 minutes ago, Arika S said:

on desktop i agree, but on laptops, it would be amazing for battery life when doing regular tasks

That's the hope anyway. Remains to be seen how much benefit that gives.

[porina]

10 minutes ago, Tedny said:

that for 2022-2023,right? 

Alder Lake has been stated by Intel as expected second half of 2021 for both desktop and mobile. That doesn't rule out the possibility of delays of course, since that is one thing they've not been short of recently.

[porina]

6 minutes ago, Tedny said:

yea, most likely 2022 chip, at that point it will be head to head with Zen4 and Zen 5 should arrived in half of the year or so 

AMD Ryzen launches are slower than one a year. We still haven't got Zen 3 yet, and Zen 4 will likely be at least a year after that. End of 2021 for Zen 4 is not impossible but seems more likely early 2022. Alder Lake might catch the tail end of Zen 3 going into Zen 4, but it will be Alder Lake successor going against Zen 5 in 2023.

 

Of course, that's best guess based on info we have today. Let's see if that needs to be revised after the Zen 3 reveal tomorrow.

[Fatih19]

15 minutes ago, porina said:

Alder Lake might catch the tail end of Zen 3 going into Zen 4,

Just like Tiger Lake, released about a year after Zen 2.

[Fatih19]

1 hour ago, RejZoR said:

No, my worry is when Intel's CU will be 16c/32t with 8 high speed cores and 8 low speed ones and AMD's in same 16c/32t will all be full fat ones. People will just look at numbers and price and think they're getting best bang in the world. In reality it'll be nowhere as fast. But if it'll be marketed correctly and people will understand it right, these will be pretty interesting. Assuming load will be balanced on them correctly.

I wonder if angry consumers would start a class-action lawsuit over this (they'll probably lost but still interesting to see).

[Rauten]

This could actually be interesting for gaming; keep the 8 little cores busy with background Windows bullcrap, discord and any browser tabs you may have open and use the 8 big cores for gaming/streaming.

If the price is right, it could be a very compelling option for gamers I think.

If the price is right. Which, this being Intel, likely won't be.

[LAwLz]

I don't see much use for power efficient cores in a desktop.

For laptops? Absolutely! But desktops? Big cores can already go fairly low in terms of idle power consumption, so we are talking about fairly minor savings here.

[porina]

55 minutes ago, Fatih19 said:

I wonder if angry consumers would start a class-action lawsuit over this (they'll probably lost but still interesting to see).

Given half a chance there'll be some idiot in the US who will try it. The scary thing is, the legal system in the US is so broken they have some chance, like the nvidia 970 VRAM case or the HD capacity cases in the past. Both IMO were technically correct on the part of the manufacturers, so the legal battle was won by other means.

 

51 minutes ago, Rauten said:

If the price is right. Which, this being Intel, likely won't be.

Intel will likely still be able to keep a price premium over AMD, but their pricing has got closer in the recent past. People who focus purely on some (usually faulty) price/performance metric ignoring all else will probably still lean AMD.

 

39 minutes ago, LAwLz said:

I don't see much use for power efficient cores in a desktop.

For laptops? Absolutely! But desktops? Big cores can already go fairly low in terms of idle power consumption, so we are talking about fairly minor savings here.

There's one scenario where this could get interesting. There is a continuing push around the world to make our devices more efficient. Desktops are not excluded from that. If the idle power can be reduced that could help with future efficiency requirements. Individual savings might be insignificant, but scale it to every device out there, the savings add up.

[bomerr]

2 hours ago, porina said:

Products have been getting more complicated for a while. You can't just look solely at a Cinebench score as a measure of processor goodness as it neglects pretty much everything else in the system outside of the CPU. Where I think we'll land is that 8 "fast" Intel cores will be competitive with 8 AMD cores, but the hybrid 16 cores will not match 16 fast cores. That's pretty much a given. For many actual more complex workloads, it will be much more complicated than that. I'm still kinda annoyed we have 16 fast cores on a consumer platform, I think they deserve better as it is really unbalanced unless you run light worloads like Cinebench.

 

That's the hope anyway. Remains to be seen how much benefit that gives.

The smaller cores could be optimized for different types of tasks. They aren't necessary crippled cores. 

[LAwLz]

28 minutes ago, porina said:

There's one scenario where this could get interesting. There is a continuing push around the world to make our devices more efficient. Desktops are not excluded from that. If the idle power can be reduced that could help with future efficiency requirements. Individual savings might be insignificant, but scale it to every device out there, the savings add up.

Sure but then you have to factor in how much extra resources it takes to actually manufacture this.

Big cores from AMD and Intel can already scale very low (even the ones tuned for high performance) so we're talking savings of maybe 1-2 watts, and only in some situations (although a fairly common situation).

Sure, over many, many CPUs it makes a bigger impact, but we're still talking about a drop in the ocean. Makes sense for battery powered devices, but for desktops I think it has more to do with being able to put a higher # of cores on the boxes.

 

 

  

17 minutes ago, bomerr said:

The smaller cores could be optimized for different types of tasks. They aren't necessary crippled cores. 

It seems very likely that the Gracemont (low power cores) are just flat out worse than the Golden Cove (high performance) cores.

Except when it comes to efficiency, of course. 

 

 

 

 

 

  

1 hour ago, Fatih19 said:

I wonder if angry consumers would start a class-action lawsuit over this (they'll probably lost but still interesting to see).

28 minutes ago, porina said:

Given half a chance there'll be some idiot in the US who will try it. The scary thing is, the legal system in the US is so broken they have some chance, like the nvidia 970 VRAM case or the HD capacity cases in the past. Both IMO were technically correct on the part of the manufacturers, so the legal battle was won by other means.

My guess is that it will play out similarly to how things played out for AMD, which I believe were sued twice.

Once because they marketed FX CPUs as "8 cores" when it didn't have 8 full cores. They lost that lawsuit and had to pay up to 12.1 million dollars.

Then I believe there was a lawsuit (but I can't find it) about AMD using misleading marketing for their APUs. They started combining CPU and GPU cores in their marketing, so they could advertise "12 core APU" when it was like 4 CPU cores (which were 2 bulldozer modules) and 8 GPU cores.

[porina]

21 minutes ago, bomerr said:

The smaller cores could be optimized for different types of tasks. They aren't necessary crippled cores. 

They are a separate line of cores that Intel have been making over many years. The optimisation is different than we're used to on performance orientated desktop, in that they're historically simpler and balanced more for lower power running. I don't think I used the term "crippled" myself, but they are generally speaking lower feature set.

 

15 minutes ago, LAwLz said:

Sure but then you have to factor in how much extra resources it takes to actually manufacture this.

Big cores from AMD and Intel can already scale very low (even the ones tuned for high performance) so we're talking savings of maybe 1-2 watts, and only in some situations (although a fairly common situation).

Sure, over many, many CPUs it makes a bigger impact, but we're still talking about a drop in the ocean. Makes sense for battery powered devices, but for desktops I think it has more to do with being able to put a higher # of cores on the boxes.

Manufacturing is not a significant barrier. We already have chiplets in the mainstream with Zen 2.

 

Outside of more intensive use cases like rendering or gaming, a system that is "in use" is still pretty much idle most of the time waiting for the user to do something. Even if it is "1-2 watts" multiply that by the number of systems out there, it adds up. It doesn't matter the individual gains are small, any gain is still a gain for the world.

 

15 minutes ago, LAwLz said:

It seems very likely that the Gracemont (low power cores) are just flat out worse than the Golden Cove (high performance) cores.

Except when it comes to efficiency, of course. 

Look at the Lakefield launch, on the overall performance/power curve the small cores fill in nicely on the low end and the big cores on the higher end. The difference in area is not insignificant either. On lakefield the 4 small cores in total take up not much more area than a single big core.

[Energycore]

This has got to be for mobile only. I don't know why Intel would want their "16 Core" part to be 30 or 40% slower than AMD's 16 core part. I feel like that would look horrible to the press and consumers that at least look at one benchmark.

[Silentprototipe]

Whille neat tech for Mobile I have no clue why this is being put on a desktop platform. Unless its a mobile CPU which then that would be a cheeky way of Intel saying they have 16 cores in a laptop with it really just being a 8 core with 8 extra gracemont cores 

[porina]

1 hour ago, Energycore said:

This has got to be for mobile only. I don't know why Intel would want their "16 Core" part to be 30 or 40% slower than AMD's 16 core part. I feel like that would look horrible to the press and consumers that at least look at one benchmark.

Only if you run cinebench or similar, in which case just buy a threadripper already. IMO 16 "big" cores really belongs on HEDT not consumer where it crippled by the platform. Given the small cores really are much smaller than the big ones, I'd expect the 8+8 retail price to come in well below the 16 core AMD of the time, although still at a premium over the 8 core AMD. That's where I think things will get really interesting.

 

For 16 big Intel cores, I'd look forward to the next HEDT platform. The current iteration has shown some indication of price matching with AMD, and I think that makes a lot more sense once you are looking at 16 cores. The only problem is that Ice Lake server was due out by end of this year, but I think I saw somewhere it was delayed again... like in past that'll likely be the basis of whatever HEDT offering is next.

[Maticks]

This makes sense for laptop design little and big core. Since in a low power profile you could tell the cpu to run off little cores only. Resulting in low battery usage and heat. Curious how’s much power saving that is.. but I don’t understand this design on desktop.

 Would have thought with desktop they’d have two cores that could boost to 6Ghz while the rest were stuck at 5.2Ghz Max since intel can’t break that barrier.

 But usually there is one or two cores on a die that can clock higher. If intel could identify which ones could they might be able to do this out of the box.

 That could help with low thread loads being pushed onto the higher clocked core.

On desktop why would you want a low power core? While your modern graphics card is sucking down 300Watts? 

[Unixsystem]

5 hours ago, porina said:

Given the small cores really are much smaller than the big ones, I'd expect the 8+8 retail price to come in well below the 16 core AMD of the time, although still at a premium over the 8 core AMD. That's where I think things will get really interesting.

I don't know that we can assume that automatically. AMD is able to churn out 8 core chiplet at an absurd rate on a well established, high capabity node. No matter how small the small cores are, you're still presumably looking at a monolithic die on a node that has been a complete trainwreck in terms of yields. There's a decent probability that an Intel 8+8 core could wind up costing more than an AMD 16 core where the production cost is split across the two chiplet + IO die.

 

Maybe they'll be willing to take a cut to profits and sell at near-cost just to keep themselves in the game while they bang their heads against 7nm for another indeterminate number of years. 

[porina]

3 minutes ago, Unixsystem said:

No matter how small the small cores are, you're still presumably looking at a monolithic die on a node that has been a complete trainwreck in terms of yields. There's a decent probability that an Intel 8+8 core could wind up costing more than an AMD 16 core where the production cost is split across the two chiplet + IO die.

Intel are going their version of chiplet. How exactly they do the split on future CPUs remains to be seen. Lakefield is the first example with a "core" die and an "IO" die. Sound familiar? 

[Unixsystem]

3 minutes ago, porina said:

Intel are going their version of chiplet. How exactly they do the split on future CPUs remains to be seen. Lakefield is the first example with a "core" die and an "IO" die. Sound familiar? 

I'm specifically referring to Alder Lake. I may have misinterpreted, but I was under the impression that Alder was still monolithic.