(Theory but possible) Efficient cores might be the future of multi-core performance

[sunny5]

Intel is gonna announce 12th gen CPU series called Alder Lake with Big LITTLE cores just like Apple. It seems they are planning to increase cores by increasing only efficient cores. I was quite skeptical because of Big LITTLE cores but who knows?

Intel Corporation Alder Lake Client Platform - Geekbench Browser

Benchmark results for an Intel Corporation Alder Lake Client Platform with an Intel Core i9-12900K processor.

browser.geekbench.com

Someone already tested with Intel i9-12900K which has 8 performance cores and 8 efficient cores. The score is 1893, 17299.

ASUS System Product Name - Geekbench Browser

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 5950X processor.

browser.geekbench.com

On the other hand, AMD Ryzen 9 5950X which has 16 cores without Big LITTLE cores. The score is 1768, 17588.

Surprisingly, 12900K performs as good as 16 cores CPU, not 8 cores CPU. Btw, M1's multicore performance was similar to 6 cores from Geekbench despite it has 4+4 cores.

_COM1_ NBINF-M7-5R9R8N - Geekbench Browser

Benchmark results for a _COM1_ NBINF-M7-5R9R8N with an AMD Ryzen 9 5900H with Radeon Graphics processor.

browser.geekbench.com

This is mobile APU AMD 5900H with 8 cores.

So you may wonder, why not just increase the CPU core only with efficient cores? 12900K's multicore score is quite impressive especially since it's not 16 core CPU instead of 8+8 core CPU. The multi core performance is as good as 5950X but the wattage and heat might be lower than 5950X.

Intel could release more 13th-gen desktop CPUs later this year

Most Raptor Lake processors for desktops and laptops have already been announced - including one 6GHz chip - but more could be on the way

www.techadvisor.com

Intel is planning to increase only efficient cores for better multicore performance. It's 13th gen so it's not too far. Even 35W will have 24 cores (8+16). If the efficient core's performance is good for multi core tasks, then Intel's road map might be surprising as hell. So M1X might need to compete Intel with 16 cores or 24 cores.

But we might need to wait for actual test results later this year because Alder Lake is not released yet. This is why it's just a theory so dont yell at me right away. But if Intel proves it, then it would be interesting.

If the theory is true, Apple can just make Mac Pro's CPU by 8 performance cores + 120 efficient cores. It will be beneficial for multicore performance while the power consumption and temperature is lower than just 128 cores CPU without efficient cores. Both Intel Xeon and AMD EPYC already have slower clock speed anyway. But using multiple cores will increase the power consumption so it may not work with laptops I guess.



Thoughts?

 

[leman]

This strategy works for Intel, but not for Apple. Apples high-performance cores use 4-5 times less power than Intel to deliver the same performance, so they are inherently much more scalable.

 

[Andropov]

This strategy works for Intel, but not for Apple. Apples high-performance cores use 4-5 times less power than Intel to deliver the same performance, so they are inherently much more scalable.

Wouldn't Apple's efficiency cores be even more scalable than its performance cores?

 

[UBS28]

Wouldn't Apple's efficiency cores be even more scalable than its performance cores?

If you need a lot of cores that are not fast, that is what the GPU is for.

 

[sunny5]

If you need a lot of cores that are not fast, that is what the GPU is for.

You still need CPU. Fugaku uses only CPU and yet it's the world fastest super computer.

 

[Argon_]

Until Intel's posted TDPs start reflecting reality, it's hard to take them seriously for mobile applications.

 

[Analog Kid]

If you need a lot of cores that are not fast, that is what the GPU is for.

GPU is for easily parallelized algorithms with minimal branching that can run within a limited memory architecture.

As far as efficiency versus performance cores, the details certainly matter. If performance is thermal limited, as it seems to be in most architectures and certainly is in Intel's architecture, then it'll matter what the performance to power tradeoff is and what the expected workload is.

There are people who know much more about the details here, but I'd imagine that Intel has more to gain from this strategy than Apple does. An Intel performance core is likely optimized to maintain performance on some very old and ugly code that is not easy to efficiently optimize in hardware, while I'd guess Intel could design an efficiency core that is reasonably performant on modern code even if it's dog slow on some handful of legacy instructions. The Intel ethos is to ensure support for legacy code, so a performance core or 8 would do that, while they can ramp up real world performance across the more streamlined efficiency cores. Legacy code probably isn't well written for multicore performance anyway.

 

[JMacHack]

I figure it’s coming whether or not it’s for performance. AMD might be the last to do big.little, but they’re best poised to do it with their chiplet design.

 

[MrGunny94]

big.Little has always been the future, despite that Intel really needs to up their game with Alder Lake... However from what I have seen around the inside news, it seems like Lunar Lake will be the big revolution as a complete new arch and redefining the node processing materials.

I will be upgrading my desktop rig from Intel Coffee Lake to Alder Lake later this year using 8 + 8 and then might upgrade to the 8 + 16 in Raptor Lake.

For Laptops I'll continue to use my Mac M1 and might even upgrade to either the Pro or the Air. Nevertheless I'm looking forward to see what happens next in the Windows world with how the M1 behaves and how Apple will continue to expand upon it

 

[Argon_]

big.Little has always been the future, despite that Intel really needs to up their game with Alder Lake... However from what I have seen around the inside news, it seems like Lunar Lake will be the big revolution as a complete new arch and redefining the node processing materials.

I will be upgrading my desktop rig from Intel Coffee Lake to Alder Lake later this year using 8 + 8 and then might upgrade to the 8 + 16 in Raptor Lake.

For Laptops I'll continue to use my Mac M1 and might even upgrade to either the Pro or the Air. Nevertheless I'm looking forward to see what happens next in the Windows world with how the M1 behaves and how Apple will continue to expand upon it

The IPC improvement from Arm V9 coupled with new TSMC nodes might improve Apple's lead.

 

[JMacHack]

However from what I have seen around the inside news, it seems like Lunar Lake will be the big revolution

It seems like the current Intel cycle is “the next gen will be really good!” I heard it for whiskey lake, ice lake, coffee lake, alder lake and now for lunar lake.

 

[MrGunny94]

It seems like the current Intel cycle is “the next gen will be really good!” I heard it for whiskey lake, ice lake, coffee lake, alder lake and now for lunar lake.

Well the 14nm++++ train was insane, Heck I upgraded desktop from the 2600k to the 8700k and only now I'm possibly considering the 12th gen due to DDR5

 

[leman]

Wouldn't Apple's efficiency cores be even more scalable than its performance cores?

If you want to build a humongous server chip, maybe. But up to the „normal“ workstation size, using performance cores is probably better way for Apple, as the resulting chip will be smaller and more flexible. For Intel, there is no other path than using efficiency cores, given how hot the performance cores run. But let’s not forget that their new Gracemont efficiency cores will probably consume as much power as Apples performance cores while being significantly slower.

 

[dmccloud]

The IPC improvement from Arm V9 coupled with new TSMC nodes might improve Apple's lead.

The problem there is that Intel jumped ahead of Apple and others and apparently locked up 3nm capacity from TSMC. Granted, that's not necessarily for consumer products, but that still throws a wrench in Apple's move to 3nm. TSMC has been working on 2nm as well, so maybe Apple leapfrogs Intel and jumps from 2nm to 3nm.

Intel to utilize TSMC 3nm technology as it struggles to move to 7nm process, says report

Intel will reportedly utilize TSMC's 3nm technology while it works to improve its own processors.

www.windowscentral.com

 

[leman]

The IPC improvement from Arm V9 coupled with new TSMC nodes might improve Apple's lead.

There are no IPC improvements from ARM V9. It’s the same Aarch64 Apple uses right now with a bunch of security and vector related extensions. Any IPC improvements have to come from Apple‘s design. Personally, I have little doubt that Apple next gen cores with match - or even outperform Golden Cove while retaining much better energy efficiency. Apple is years ahead in this field and they can scale IPC easier.

The problem there is that Intel jumped ahead of Apple and others and apparently locked up 3nm capacity from TSMC. Granted, that's not necessarily for consumer products, but that still throws a wrench in Apple's move to 3nm. TSMC has been working on 2nm as well, so maybe Apple leapfrogs Intel and jumps from 2nm to 3nm.

Intel to utilize TSMC 3nm technology as it struggles to move to 7nm process, says report

Intel will reportedly utilize TSMC's 3nm technology while it works to improve its own processors.

www.windowscentral.com

Im not sure how reliable these reports are. I remember seeing some later news that Apple has also booked 3nm capacity.

 

[Falhófnir]

I like the road analogy for this - the single core is the speed limit, the multi core is the number of lanes at that speed limit - for something lighter like running the system, it benefits from getting to the end of the road faster more than it does from having more lanes available to move the load along simultaneously.

Apple devices feel so ‘snappy’ in part because they have superlative single core performance (Also because of fast storage and large caches etc) but giving up that really high single core performance for more multi core is something that will be detrimental to a lot of users and useful to relatively few.

 

[leman]

I like the road analogy for this - the single core is the speed limit, the multi core is the number of lanes at that speed limit - for something lighter like running the system, it benefits from getting to the end of the road faster more than it does from having more lanes available to move the load along simultaneously.

Apple devices feel so ‘snappy’ in part because they have superlative single core performance (Also because of fast storage and large caches etc) but giving up that really high single core performance for more multi core is something that will be detrimental to a lot of users and useful to relatively few.

Pretty much this. Using many weaker cores to improve multi-core performance only works for applications that can efficiently split their work in many parallel packages. In the end, it’s a balance between power, chip size/cost and programming complexity. Intel is betting on efficiency cores because their performance tech is extremely power-hungry. But overall, designs with fewer faster cores will be more flexible, because they can adapt better for various workload types.

 

[jdb8167]

The problem there is that Intel jumped ahead of Apple and others and apparently locked up 3nm capacity from TSMC. Granted, that's not necessarily for consumer products, but that still throws a wrench in Apple's move to 3nm. TSMC has been working on 2nm as well, so maybe Apple leapfrogs Intel and jumps from 2nm to 3nm.

Intel to utilize TSMC 3nm technology as it struggles to move to 7nm process, says report

Intel will reportedly utilize TSMC's 3nm technology while it works to improve its own processors.

www.windowscentral.com

The article says it is both Intel and Apple that have contracted to use TSMC's 3 nm. How does that mean Intel jumped ahead? The last I heard, Intel was only going to use TSMC for GPUs anyway. That is a threat to Nvidia and AMD not Apple. Also, Nikkei is about the least reliable of rumor sources there is.

 

[dmccloud]

The article says it is both Intel and Apple that have contracted to use TSMC's 3 nm. How does that mean Intel jumped ahead? The last I heard, Intel was only going to use TSMC for GPUs anyway. That is a threat to Nvidia and AMD not Apple. Also, Nikkei is about the least reliable of rumor sources there is.

This article goes into more specifics. The biggest takeaway is that by locking in all remaining 3nm capacity, Intel has boxed everyone else out on that process node. That means neither Apple nor AMD could ramp up production if demand outstrips supply of 3nm parts.

Intel locks down all remaining TSMC 3nm production capacity, boxing out AMD and Apple

Intel's 3nm orders include a new GPU and three server chips

www.techradar.com

 

[jdb8167]

This article goes into more specifics. The biggest takeaway is that by locking in all remaining 3nm capacity, Intel has boxed everyone else out on that process node. That means neither Apple nor AMD could ramp up production if demand outstrips supply of 3nm parts.

Intel locks down all remaining TSMC 3nm production capacity, boxing out AMD and Apple

Intel's 3nm orders include a new GPU and three server chips

www.techradar.com

There is no way that TSMC will anger Apple. Intel (at least) claims this is a temporary measure until they can catch up. Do you really think TSMC is willing to piss off their largest customer for a one-time payment. It doesn't pass the most basic logic test.

 

[l0stl0rd]

Pretty much this. Using many weaker cores to improve multi-core performance only works for applications that can efficiently split their work in many parallel packages. In the end, it’s a balance between power, chip size/cost and programming complexity. Intel is betting on efficiency cores because their performance tech is extremely power-hungry. But overall, designs with fewer faster cores will be more flexible, because they can adapt better for various workload types.

I think it will be interesting to see how they work under windows 11 with the new scheduler. This will either make the great it not.

 

[l0stl0rd]

Intel is gonna announce 12th gen CPU series called Alder Lake with Big LITTLE cores just like Apple. It seems they are planning to increase cores by increasing only efficient cores. I was quite skeptical because of Big LITTLE cores but who knows?

Intel Corporation Alder Lake Client Platform - Geekbench Browser

Benchmark results for an Intel Corporation Alder Lake Client Platform with an Intel Core i9-12900K processor.

browser.geekbench.com

Someone already tested with Intel i9-12900K which has 8 performance cores and 8 efficient cores. The score is 1893, 17299.

ASUS System Product Name - Geekbench Browser

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 5950X processor.

browser.geekbench.com

On the other hand, AMD Ryzen 9 5950X which has 16 cores without Big LITTLE cores. The score is 1768, 17588.

Surprisingly, 12900K performs as good as 16 cores CPU, not 8 cores CPU. Btw, M1's multicore performance was similar to 6 cores from Geekbench despite it has 4+4 cores.

_COM1_ NBINF-M7-5R9R8N - Geekbench Browser

Benchmark results for a _COM1_ NBINF-M7-5R9R8N with an AMD Ryzen 9 5900H with Radeon Graphics processor.

browser.geekbench.com

This is mobile APU AMD 5900H with 8 cores.

So you may wonder, why not just increase the CPU core only with efficient cores? 12900K's multicore score is quite impressive especially since it's not 16 core CPU instead of 8+8 core CPU. The multi core performance is as good as 5950X but the wattage and heat might be lower than 5950X.

Intel could release more 13th-gen desktop CPUs later this year

Most Raptor Lake processors for desktops and laptops have already been announced - including one 6GHz chip - but more could be on the way

www.techadvisor.com

View attachment 1823405
Intel is planning to increase only efficient cores for better multicore performance. It's 13th gen so it's not too far. Even 35W will have 24 cores (8+16). If the efficient core's performance is good for multi core tasks, then Intel's road map might be surprising as hell. So M1X might need to compete Intel with 16 cores or 24 cores.

But we might need to wait for actual test results later this year because Alder Lake is not released yet. This is why it's just a theory so dont yell at me right away. But if Intel proves it, then it would be interesting.

View attachment 1823410
If the theory is true, Apple can just make Mac Pro's CPU by 8 performance cores + 120 efficient cores. It will be beneficial for multicore performance while the power consumption and temperature is lower than just 128 cores CPU without efficient cores. Both Intel Xeon and AMD EPYC already have slower clock speed anyway. But using multiple cores will increase the power consumption so it may not work with laptops I guess.



Thoughts?

As for the Amd and Intel numbers… seems nice but I think the M1x will not be far off, might even beat them. Especially if they went from 5nm to 5nm+

 

[Andropov]

Pretty much this. Using many weaker cores to improve multi-core performance only works for applications that can efficiently split their work in many parallel packages. In the end, it’s a balance between power, chip size/cost and programming complexity. Intel is betting on efficiency cores because their performance tech is extremely power-hungry. But overall, designs with fewer faster cores will be more flexible, because they can adapt better for various workload types.

Agreed. However, I'd argue that what Intel has used for the 12900K (8+8 cores) is a good compromise. If the workload can make use of more than 8 performance cores, it's likely that the workload is easily parallelizable to an arbitrary number of cores. I can't think of that many tasks that can use, let's say, 9 cores but not 10.

 

[Argon_]

Pretty much this. Using many weaker cores to improve multi-core performance only works for applications that can efficiently split their work in many parallel packages. In the end, it’s a balance between power, chip size/cost and programming complexity. Intel is betting on efficiency cores because their performance tech is extremely power-hungry. But overall, designs with fewer faster cores will be more flexible, because they can adapt better for various workload types.

Do you think they'll get it under control, or keep up the factory overclocks to chase a number?

It would be nice to build a desktop that doesn't require adding a second boiler to the home powerplant.

 

[dmccloud]

There is no way that TSMC will anger Apple. Intel (at least) claims this is a temporary measure until they can catch up. Do you really think TSMC is willing to piss off their largest customer for a one-time payment. It doesn't pass the most basic logic test.

As I stated earlier, this only affects 3nm processes. It's not like Apple faces a significant disadvantage on the CPU/SoC front with the 5nm process, especially given the scalability and modularity inherent to Apple's silicon designs. Apple could probably stay on 5nm/5nm+ for now and just jump to 2nm when it's ready, which would again put Intel behind the rest of the industry. AMD is also a major customer of TSMC, so perhaps Apple and AMD work together to outmaneuver Intel on the move to 2nm. What doesn't pass the logic test is trying to address Intel's move to lock up all remaining 3nm production in a vacuum without taking into account the options available to both Apple and AMD regarding TSMC capabilities and processes.