Wednesday, April 28th 2021

AMD Zen 5 "Strix Point" Processors Rumored To Feature big.LITTLE Core Design

AMD launched the 7 nm Zen 3 microarchitecture which powers Ryzen 5000 processors in late 2020, we expect AMD to follow this up with a Zen 3+ on 6 nm later this year and a 5 nm Zen 4 in 2022. We are now beginning to receive the first rumors about the 3 nm Zen 5 architecture which is expected to launch in 2024 in Ryzen 8000 series products. The architecture is reportedly known as "Strix Point" and will be manufactured on TSMC's 3 nm node with a big.LITTLE core design similar to the upcoming Intel Alder Lake and the Apple M1. The Strix Point lineup will consist exclusively of APUs and could feature up to 8 high-performance and 4 low-performance cores which would be less than what Intel plans to offer with Alder Lake. AMD has allegedly already set graphics performance targets for the processors and that they will bring significant changes to the memory subsystem but with rumors for a product 3 years away from launch take them with a healthy dose of skepticism.
Sources: MEOPC, Video Cardz
Add your own comment

70 Comments on AMD Zen 5 "Strix Point" Processors Rumored To Feature big.LITTLE Core Design

#26
Operandi
Tartaros
There has been intentions of changing into RISC in general cpus since some quite long time and having everything made in house and there is the developement of the IOS into a unified platform for a long time now. They would have switched to their M1 even if Intel delivered, thinking they would have change for AMD at some point it's just wishful thinking, they could have done that in the best scenario for AMD which has been the last 3 years and they didn't.

The RISC boat sailed a long time ago for Apple.
Oh, yeah I totally agree it wouldn't have made any sense for them to adopt AMD when they already have so much invested in a different ISA. I don't think they ever wanted to ditch PowerPC in the first place (RISC) but back then IBM was having issues of their own, Apple didn't have a CPU design of their own, and AMD was probably too small, so they had little choice but to go with Intel.

Maybe in parallel universe where Apple's own silicon isn't as strong as it is or M1 wasn't ready it could have happened and would have been interesting to see but it would have been a stop-gap effort at most.
Posted on Reply
#27
Tartaros
Operandi
Oh, yeah I totally agree it wouldn't have made any sense for them to adopt AMD when they already have so much invested in a different ISA. I don't think they ever wanted to ditch PowerPC in the first place (RISC) but back then IBM was having issues of their own, Apple didn't have a CPU design of their own, and AMD was probably too small, so they had little choice but to go with Intel.

Maybe in parallel universe where Apple's own silicon isn't as strong as it is or M1 wasn't ready it could have happened and would have been interesting to see but it would have been a stop-gap effort at most.
Rather than being small, they didn't have a unified platform like Intel had at the time. Intel was some years into the Centrino thing they started with Pentium M and offered everything a computer needed with little hassle while AMD was doing the old thing of manufacturing cpus and some chipsets that weren't very good, ironically the best AMD chipsets were from Nvidia at that time. They also didn't offer an I/O solution, storage was dodgy, mobile gpu was still a separate thing and not even today they offer networking. Intel did all of that in one package, it's obvious they were the best choice for a swift transition.

And also there is the Intel resurrection with Core Duo, AMD was on its downward spiral into Bulldozer and ATI's adquisition also was hurting. But if it was in the Pentium III/Athlon or Pentium 4/Athlon XP and 64, Apple would have gone AMD almost for sure. Even today Intel is not into a full NetBurst situation, those were hellish years for them.
Posted on Reply
#28
Wirko
dragontamer5788
So the general issue is that Windows (and Linux / Android) don't really have much to gain from little cores yet. What kind of thread/process should go on a little core? Its a complicated question: it seems that big cores are more efficient at higher GHz, while little cores really benefit if you downclock them to 200MHz or slower.

The scheduler has to not only pick the threads/processes that go on big vs little cores, but also pick what GHz (or MHz) to run the processor at. There's a bit of a "race to idle" problem. If you spend 1000ms running a task on a little core, you probably would use more power than if you had run the task on a big-core for 100ms and then slept for 900ms afterwards.

But if the sleep is interrupted after 500ms, your calculus changes yet again. So you need to predict the future and know when to turn on big cores (or turn them off), and same thing with when to clock little cores up or down.

---------

Answering these scheduling problems adequately for sizable gains in efficiency is not very easy. Its a lot easier when you have all the same kind of core: the answer in that case is simple. Just estimate the power usage per core and keep the clock as low as reasonable.
True, and it gets more complicated when you combine cores with different capabilities, like the presence or absence of multithreading or AVX-512.

When you need to predict the future, you use predictors based on statistics from recent past, like branch predictors. For scheduling, I can imagine a solution that's based on both HW and SW. There would have to be some dedicated hardware on the CPU that collects some statistics about program execution. For example, how much time is spent executing/emulating AVX, or how much time is spent waiting for I/O or memory while the core is gobbling up power, or how much time is spent waiting because of the other thread on the same core is using some shared resource. The scheduler would then use these statistics to determine if the execution is optimal and move it to another core if it isn't.
The executable code itself could contain some metadata, provided by the compiler or manually, for a whole DLL/library or more detailed, and the scheduler would use that data as a hint when picking the best core for that code.
As it's based on statistics, it would be called "AI scheduler", of course.
Posted on Reply
#29
CheapMeat
What if the OS see's patterns of usage for certain background tasks and services, so makes sure those run on the little cores more often, as they don't need high performance but instead consistency?
Posted on Reply
#30
Vanny
Did AMD make love with ASUS? "Strix Point"?
Posted on Reply
#31
Operandi
Vanny
Did AMD make love with ASUS? "Strix Point"?
Nah.
  1. The strix (plural striges or strixes), in the mythology of classical antiquity, was a bird of ill omen, the product of metamorphosis, that fed on human flesh and blood. It also referred to witches and related malevolent folkloric beings.
Sounds like AMD is about get midevil on Intel's ass, lol.
Posted on Reply
#32
Tartaros
Wirko
True, and it gets more complicated when you combine cores with different capabilities, like the presence or absence of multithreading or AVX-512.

When you need to predict the future, you use predictors based on statistics from recent past, like branch predictors. For scheduling, I can imagine a solution that's based on both HW and SW. There would have to be some dedicated hardware on the CPU that collects some statistics about program execution. For example, how much time is spent executing/emulating AVX, or how much time is spent waiting for I/O or memory while the core is gobbling up power, or how much time is spent waiting because of the other thread on the same core is using some shared resource. The scheduler would then use these statistics to determine if the execution is optimal and move it to another core if it isn't.
The executable code itself could contain some metadata, provided by the compiler or manually, for a whole DLL/library or more detailed, and the scheduler would use that data as a hint when picking the best core for that code.
As it's based on statistics, it would be called "AI scheduler", of course.
I'm not versed in these matters, so maybe you can help me to understand this: it depends just on the OS maker or the software programmer can influence how their program might take care advantage of the little cores, like identifying them and telling them to use the little ones? And if they can, there is a situation that the software requires more horsepower, can the programmer tell the software to switch to a better core on the fly?
Posted on Reply
#33
Vanny
Operandi
Nah.
  1. The strix (plural striges or strixes), in the mythology of classical antiquity, was a bird of ill omen, the product of metamorphosis, that fed on human flesh and blood. It also referred to witches and related malevolent folkloric beings.
Sounds like AMD is about get midevil on Intel's ass, lol.
Yeah, they ran them into the ground so much they decided to also destroy them in the dumbass pricing department
Posted on Reply
#34
Operandi
Vanny
Yeah, they ran them into the ground so much they decided to also destroy them in the dumbass pricing department
Not sure what that means. There isn't anyone in the industry that can hold a candle to Intel's pricing over the last 10+ years.
Posted on Reply
#35
Vanny
Operandi
Not sure what that means. There isn't anyone in the industry that can hold a candle to Intel's pricing over the last 10+ years.
They showed them who's the boss when it comes to CPUs and market disruption, now they showed them, and will continue to show them (unless blue boys do something about it) who's the boss when it comes to dumb pricing.
Posted on Reply
#36
Operandi
Vanny
Yeah, they ran them into the ground so much they decided to also destroy them in the dumbass pricing department
Not sure what that means. There isn't anyone in the industry that can hold a candle to Intel's pricing over the last 10 years.
Tartaros
I'm not versed in these matters, so maybe you can help me to understand this: it depends just on the OS maker or the software programmer can influence how their program might take care advantage of the little cores, like identifying them and telling them to use the little ones? And if they can, there is a situation that the software requires more horsepower, can the programmer tell the software to switch to a better core on the fly?
I'm not expert either but I know there are and should be optimizations that should be done on both the OS and software level.

I know certain game engines would completely fall apart on on Bulldozer if the game wasn't made aware of the clustered nature of threads. I'm sure something similar will have to be done with big/little cores as you can't have a rendering thread for example go from a high performance big core to little one with half the performance without drastic performance issues.
Vanny
They showed them who's the boss when it comes to CPUs and market disruption, now they showed them, and will continue to show them (unless blue boys do something about it) who's the boss when it comes to dumb pricing.
Yeah, AMD could go lower for sure but thats not how you run a company. Competition is weak but performance is pretty in line with price, especially when you consider that Intel has nothing in their stack that can do what AMDs top end parts do, nothing about AMD's pricing is abusive. The same cannot be said for Intel over the years with halo HEDT CPUs costing orders of magnitude more than the desktop units of the same family with absolutely no performance to justify it.
Posted on Reply
#37
Vanny
Operandi
Yeah, competition is weak but performance is pretty in line with price.
Not really when I can get a 11400f and a capable cheap B560 board for it for the price of a single 5600X and have comparable performance. I'm going to find it hard to recommend people Zen 3 CPUs with their pricing and availability, when there are 10850Ks going for like $350. And it's why I want Alder Lake to be good, so that AMD comes back to their senses. They're clearly taking advantage of that dominant market position while they can.
Posted on Reply
#38
Operandi
Vanny
Not really when I can get a 11400f and a capable cheap B560 board for it for the price of a single 5600X and have comparable performance. I'm going to find it hard to recommend people Zen 3 CPUs with their pricing and availability, when there are 10850Ks going for like $350. And it's why I want Alder Lake to be good, so that AMD comes back to their senses. They're clearly taking advantage of that dominant market position while they can.
Yeah the 11400 / 600 are good deals but competing on price is what you do when you don't have the technology lead. The other thing to keep in mind is AMD is supply limited and will have no problem selling every CPU they make. Also, TSMC yields are really good on 7nm so most what is being produced is going to fully enabled 8 core dies which further puts a supply limit on 6 core 5600x CPUs.

So yeah... AMD is being sensible.
Posted on Reply
#39
Wirko
Tartaros
I'm not versed in these matters, so maybe you can help me to understand this: it depends just on the OS maker or the software programmer can influence how their program might take care advantage of the little cores, like identifying them and telling them to use the little ones? And if they can, there is a situation that the software requires more horsepower, can the programmer tell the software to switch to a better core on the fly?
Hey, I'm just waving a napkin with some ideas handwritten on both sides. I don't think that the application programmer should have absolute control over these details. The program needs to run efficiently on a wide range of processors with various number of big and little cores, the user can choose one of several power plans, and the programmer can't predict any of that. But the programmer could include some additional info that says, for example, "this DLL/this procedure usually doesn't benefit at all from multithreading". The scheduler would use this, as well as other information: statistics that I mentioned before, number of cores, total CPU load, power plan, process priority etc. to choose the most appropriate core.
1d10t
Which box? Haswell is already long overdue.
Only if Intel decides to reverse the flow of time, which at this point isn't completely out of question, and isn't illegal under Moore's Law. But then you'd need to wait for Comet Lake, Coffee Lake Refresh, Coffee Lake, Kaby Lake, Skylake and Broadwell to come first. All of them still on 14 nm, mind you.
Posted on Reply
#40
dragontamer5788
CheapMeat
What if the OS see's patterns of usage for certain background tasks and services, so makes sure those run on the little cores more often, as they don't need high performance but instead consistency?
Most background tasks are like 'check I/O for virus signatures', which only runs when you read or write to disk.

What this really looks like is a 5ms (for the hard drive to respond) and then 4096 bytes (one sector of the hard drive loaded). A millisecond is very slow for a computer (4 GHz is 0.25 nanoseconds. 5ms is 5000000 nanoseconds or 20 million cycles).

Even if done on a slow 200MHz little core, this kind of background task almost certainly will be 'run and sleep'. Sleeping faster (by executing on a big core) could very well be the more efficient decision.
Posted on Reply
#41
Caring1
NuCore
Some of the commenters are hypocrites. That's why I made a meme that describes it well ;)
I'm glad you said some.
If AMD adopt the same methods of implementation then the meme is relevant, but if their path differs to Intel's, then it's not.
So basically it's not what they do, but how they do it that matters.
Oh and welcome.
Posted on Reply
#42
MentalAcetylide
Vanny
They showed them who's the boss when it comes to CPUs and market disruption, now they showed them, and will continue to show them (unless blue boys do something about it) who's the boss when it comes to dumb pricing.
What were those blue characters suppose to be, some kind of grotesque humanoid version of hatless smurfs? Melancholy paves the way for Intel? Just my opinion, but I think Intel could've chosen a better color.
Posted on Reply
#43
Minus Infinity
I thought AMD had some arrangement with ARM in the past, so I would not be surprised to see ARM architecture cores for the little. Maybe they will resurrect the concept behind project Skybridge from around 2015.

Anyway does this story imply there will be no more 12 and 16 core desktop, as it said all Zen5 will be APU design.
Posted on Reply
#44
Operandi
Minus Infinity
I thought AMD had some arrangement with ARM in the past, so I would not be surprised to see ARM architecture cores for the little. Maybe they will resurrect the concept behind project Skybridge from around 2015.

Anyway does this story imply there will be no more 12 and 16 core desktop, as it said all Zen5 will be APU design.
Zen is x86; you can't just throw different ISAs around in the same CPU and it expect it to work as was previously pointed out. AMD was working on something ARM before all the resources got dumped into Zen and I think they licensed the ARM ISA so it was a from scratch core similar to what Apple is doing so yeah, maybe they'll pick it back up but it won't be a Zen CPU.

And this is Zen 5, supposedly on 3nm. At that point a 16 core APU will be nothing, particularly if use multiple chiplets.
Posted on Reply
#45
Mussels
Moderprator
Vya Domus
Why would you even want to have two cluster with entirely different ISAs ?
purely a theoretical thing because of the way the world is moving with apple causing a shift over to arm

ARM hardware is a lot more power efficient so what if the OS could run on 15W of ARM hardware while the x86 cores slept?
Posted on Reply
#46
Operandi
Mussels
purely a theoretical thing because of the way the world is moving with apple causing a shift over to arm

ARM hardware is a lot more power efficient so what if the OS could run on 15W of ARM hardware while the x86 cores slept?
You don't have the slightest idea of how CPUs execute code. You can't mix and match ISAs in one CPU simply because you think one is better at one thing or another (which is false btw). You might as well say you want a CPU that runs on 8 big steam punk mechanical horse cores and 16 little gnome with slide rulers cores, makes about the same level of sense...
Posted on Reply
#47
ratirt
Why would AMD use Big.little? They still have the power efficiency advantage and the cores are pretty fast.
Posted on Reply
#48
Caring1
Mussels
purely a theoretical thing because of the way the world is moving with apple causing a shift over to arm

ARM hardware is a lot more power efficient so what if the OS could run on 15W of ARM hardware while the x86 cores slept?
More likely to use 15W x86 cores in association with full sized x 86 cores
Posted on Reply
#49
dragontamer5788
ratirt
Why would AMD use Big.little? They still have the power efficiency advantage and the cores are pretty fast.
big cores are surprisingly energy efficient these days. LITTLE cores manage to win in some applications however: low and slow. In particular, very low MHz (like 200) and much lower voltages with non-CPU heavy tasks. If the CPU is the bottleneck, you probably want the big-core. DDR4 uses a good chunk of power, including the L3, L2, L1 caches and memory controller. As long as there's work to do, a bigger core can beat LITTLE cores in efficiency.

If you're doing render farms or other "big" tasks, a bigger core at a lower frequency (think EPYC) is the best bet. But if you're streaming data from a hard drive out of a NIC into the Internet... LITTLE cores probably win (very low CPU requirements). "Schedulers aren't smart enough" to make these decisions. Heck, I don't think anyone is really smart enough to figure out the problem right now.
Posted on Reply
#50
R0H1T
medi01
Apple is doing it out of greed, so gotta be cool. #3.5mmAudioJacks
They aren't doing one with x86 & the other with ARM, so no.
Mussels
ARM hardware is a lot more power efficient so what if the OS could run on 15W of ARM hardware while the x86 cores slept?
No one's doing that, at least in the immediate future. There was a leak(?) that AMD might do something like that IIRC with the K12 or something but that's it?
Posted on Reply
Add your own comment