AMD CTO confirms it is adopting hybrid architecture for its consumer CPUs

DragonSlayer101

Posts: 373   +2
Staff
What just happened? AMD has confirmed that it will use hybrid architectures in consumer processors in the future, thereby replacing uniform CPU cores with a combination of performance cores, efficiency cores, and accelerators, like Intel. The company did not offer a definitive timeline to make the change, but confirmed that the new design would bring variations not only in core density but also in the type of core and how they are configured.

Speaking to Tom's Hardware's deputy managing editor Paul Alcorn on the sidelines of the ITF World conference in Antwerp, Belgium, AMD CTO Mark Papermaster said that the change will introduce "high-performance cores mixed with power-efficient cores mixed with acceleration." According to the exec, the hybrid designs will be based on specific application needs, as core count alone does not effectively serve customers with varied requirements.

The confirmation comes after AMD debuted a hybrid structure for its Ryzen 7040 'Phoenix' APUs with the inclusion of AI acceleration, but they still use generic uniform cores like other consumer CPUs in the company's portfolio.

Hybrid core architectures have been around for a couple of years in the x86 space, having been debuted by Intel with its Alder Lake processors in 2021. Alder Lake was the first consumer x86 CPU lineup to come with a combination of performance and efficiency cores to help improve power consumption. While the performance cores come with high clock speeds and hyperthreading, the efficiency cores consume less power and occupy less chip space than the more powerful cores.

The same idea is also used by ARM to design smartphone SoCs, where hybrid architectures help the processors offer high-end performance while reducing power consumption. While the performance cores in these chips power the apps and services that run in the foreground, the efficiency cores are used for background tasks that do not typically need a great deal of processing power.

Meanwhile, Papermaster also discussed AMD's plans to integrate AI into the company's chip design, testing, and verification processes, and said that the company already uses artificial intelligence to improve performance and lower energy consumption. Papermaster also confirmed that the company plans to use generative AI more extensively for chip design in the future.

Answering a question about whether generative AI could one day design microarchitectures, Papermaster said it is plausible that it could "build upon existing designs" in the future. However, he looked to allay fears that artificial intelligence could once day make chip designers redundant. According to him, "It (AI) won't replace designers, but I think it has a tremendous capability to speed design."

Permalink to story.

 
Fingers crossed this doesn't infect the chiplets in the future. Make it for APU and laptop designs only.

I love that about the chiplets as they are now. All the USB, storage controllers, PCIe, integrated GPU and what not (security processor) are all separated from the CPU cores so all that junk doesn't take any of the valuable CPU silicon area.
 
If they have to say AI won't replace people then it will. AMD is america based and one of our biggest hobbies is screwing over the citizenry.
 
I wonder what led AMD to change its tune? When alder lake first came out, amd made it seem like an all big core design was better. Yet amd is following in intel’s footsteps. Why?

 
I wonder what led AMD to change its tune? When alder lake first came out, amd made it seem like an all big core design was better. Yet amd is following in intel’s footsteps. Why?

I assume now that process node shrinks are slowing down power leakage and usage might hinder improvements and one way to mitigate that is the Big.Small core method like Intel and Arm uses. Its a way to continue to incrementally improve performance while keeping power envelopes within reason. Just my 2 cents.
 
Makes sense for mobile but is nonsensical for desktops. Workstations have many cores, servers have many high performing cores, regular consumers don't need either.
 
I wonder what led AMD to change its tune? When alder lake first came out, amd made it seem like an all big core design was better. Yet amd is following in intel’s footsteps. Why?

Why? Becuase firstly AMD's e-cores are still very strong, much stronger than Intel's, they are power efficient versions of regular P cores with less cache and lower clocks, but still 70-80% the performance. AMD's is getting beaten soundly in productivity now, just look at how 13700K beats 7900X in many MTing apps other than a few cases. I greatly welcome this move if it means we get say a 8850X with 8 Zen 5 cores, 4-8 Zen 5c cores.
 
If they can make the scheduling work a 8 core + 3D vcache chiplet plus a 16 core dense chiplet can make sense for high end desktops.

As long as they keep the single chiplet option for gamers.
 
AMD's e-cores are still very strong, much stronger than Intel's, they are power efficient versions of regular P cores with less cache and lower clocks, but still 70-80% the performance
Given that there aren't any products on the market currently using AMD's Zen 4c design, it's too early to tell how well it compares to Intel's Gracement (or Zen 4, for that matter). There are no details about the specific nature of the cache changes -- it could be a reduction of just L3, a reduction in L2 and L1, a reduction in data path widths, or a combination of all of these. AMD has just used the phrase "density-optimized cache" which could mean almost anything.
 
I think amd will still use uniform cores for desktop since it use same chiplet as epyc.

Hybrid architecture is suitable for laptop use cases which actually doesn't need avx512
 
I wonder what led AMD to change its tune? When alder lake first came out, amd made it seem like an all big core design was better. Yet amd is following in intel’s footsteps. Why?
As written in the article, efficiency cores need less circuit.
Hence it has lower production cost and higher profit
 
Makes sense for mobile but is nonsensical for desktops. Workstations have many cores, servers have many high performing cores, regular consumers don't need either.
It's suitable for desktop too because usually there is only 1 active application while other apps runs in idle mode.
And avx512 is not useful for most consumer user cases
 
If they can make the scheduling work a 8 core + 3D vcache chiplet plus a 16 core dense chiplet can make sense for high end desktops.

As long as they keep the single chiplet option for gamers.

AFAIK Windows 11 has a big/little aware scheduler for Intel CPUs, and I'd assume that it shouldn't be too hard to make it work with whatever AMD produces. Of course that means that both manufactures will eventually be locking the majority of their user base into Win11. Kind of makes a person wonder if there wasn't a bit of M$ arm twisting involved in the decision... Just saying.
 
Why? Becuase firstly AMD's e-cores are still very strong, much stronger than Intel's, they are power efficient versions of regular P cores with less cache and lower clocks, but still 70-80% the performance. AMD's is getting beaten soundly in productivity now, just look at how 13700K beats 7900X in many MTing apps other than a few cases. I greatly welcome this move if it means we get say a 8850X with 8 Zen 5 cores, 4-8 Zen 5c cores.
So then there is some merit in a hybrid design… performance per watt per die area savings.
It's suitable for desktop too because usually there is only 1 active application while other apps runs in idle mode.
And avx512 is not useful for most consumer user cases
Avx512 is good for emulators.
 
Why run emulated old games while there are many better native games
Nostalgia. There is no modern equivalent to many of the older games on ps3 Super Nintendo etc. no modern equivalent to super Mario world. I don’t personally emulate ps3 games (the ps3 emulator can benefit from avx512) but I do have a collection of Super Nintendo games that I play from time to time.
 
Back