The difference between AMD cores and Intel cores

[Kuzma]

Introduction

So in my last topic talking about console optimisations and how they will effect people (you can see that here) I mentioned the difference between AMD cores and Intel cores just briefly but the topic is so large and broad that it deserves a thread to itself quite frankly so here it is.

 

What is the difference between AMD(FX) cores and Intel cores?

I'm going to start off with a very basic example for the tl;dr people and then expand as I go along for those of you that don't mind reading for ages :P . AMD use modules that contain 2 cores that share resources, Intel use cores that each have smaller individual resources. The easiest way to think of it is that AMD use rectangles that contain two squares (cores) and Intel use the squares themselves ^_^ (note: Intel processors share L3 cache so the sharing resources idea is not unique to AMD).

 

So are AMD(FX) cores true cores?

This question is asked a lot and the idea that AMD cores are not true cores is thrown around a lot slandering AMD's processors. The fact of the matter is linked to what I said above, when people think about cores they think about the traditional squares that Intel uses but AMD messed with this idea by creating the rectangle that contains two squares ^_^ , if you think about it like that you will realise that yes they are indeed true cores.

 

What're the advantages of this new design?

The advantages are pretty simple to understand actually, if the two cores in the modules are co-operating with each other and working on a task together then sharing resources is perfect because they can share answers, so to speak, and complete a task faster than if they were working on a task together while not co-operating. So if you think about it like that tasks that use an even number of cores AMD's core design is a better idea ^_^ .

 

What're the disadvantages of this new design?

The moment you get tasks using 1,3,5,7 etc. number of cores then you start to run into issues of the cores working on different things and trying to use some of the resources that the other core is already working with and the cores start to conflict reducing the efficiency of the task (which is the reason FX processors don't have an MP ratio equal to the core count).

 

Conclusion

AMD and Intel are no longer using the same nature of cores and have branched off in different directions meaning that we have to think about cores differently ^_^ there isn't really such a thing as a "true core" because cores can be very different. I hope this has helped you understand the difference between Intel cores and AMD(FX) cores, if there's something you want me to add or feel I've made a mistake feel free to PM me.

 

P.S. For those of you who understand things through visualisation there's a picture of the 8350 here and the 4770k here

[joelthezombie15]

Nice. I never actually payed it any thought. I figured. CPU is CPU

[Kuzma]

Nice. I never actually payed it any thought. I figured. CPU is CPU

Thanks and xD neither did I but when I saw someone post once that an AMD core isn't a real core I went investigated and found out why they were wrong but right at the same time ^_^ figured I'd share what I'd found

[MetroP]

In AMD processors the building block is the module, you just add however modules as you can, it's a complete and independent entity.

 

2 integer (traditional) cores and one floating point unit reside in a module, making the CPU an 8 thread CPU for integer workloads and a quad core CPU for floating point unit workloads.
However AMD skipped on the floating point units because it intends to have all floating point workloads off-loaded to the GPU.

We'll start seeing performance breakthroughs with the new APUs "kaveri" because they're the first APUs to allow for workloads to be offloaded to the GPU cores as if they were just regular CPU cores (HSA & HUMA), so programming them to do specialized tasks will be very easy and this will skyrocket the performance.

[Kuzma]

In AMD processors the building block is the module, you just add however modules as you can, it's a complete and independent entity.

 

2 integer (traditional) cores and one floating point unit reside in a module, making the CPU an 8 thread CPU for integer workloads and a quad core CPU for floating point unit workloads.

However AMD skipped on the floating point units because it intends to have all floating point workloads off-loaded to the GPU.

We'll start seeing performance breakthroughs with the new APUs "kaveri" because they're the first APUs to allow for workloads to be offloaded to the GPU cores as if they were just regular CPU cores (HSA & HUMA), so programming them to do specialized tasks will be very easy and this will skyrocket the performance.

Thank you for the technicality :P I tried to simplify everything as much as possible to make it easier for people to understand

[Emperor_Piehead]

Great write up one of the mods should pin this.  games like crysis 3 that don't use FPU benefit a lot from the modules and then the 8 core amd becomes competitive gaming wise against the 3930k.

[Kuzma]

Great write up one of the mods should pin this.  games like crysis 3 that don't use FPU benefit a lot from the modules and then the 8 core amd becomes competitive gaming wise against the 3930k.

^_^ Great real world example of where the AMD design is superior

[TechFan@ic]

Die shots for the win !

FX 8350

i7 3960x

[Kuzma]

Die shots for the win !

FX 8350

i7 4770K

I'll put those in the first post ^_^

[WoodenMarker]

Nice. I never actually payed it any thought. I figured. CPU is CPU

"This is my CPU. There are many like it but this one is mine."

[Kuzma]

"This is my CPU. There are many like it but this one is mine."

chances are there is no exact replica of your cpu as yours most likely overclocks differently

[WoodenMarker]

chances are there is no exact replica of your cpu as yours most likely overclocks differently

'like'.

[SteepGnomeKing]

It's like cuda cores and stream processors you just can't compare.

 

But nice write up. Will help a lot of people wondering why amd has 8 cores and intel has 4 cores on the consumer level.

 

I don't think of AMD cores as being true cores in the traditional sense though. It shares to many resources for it to be its own independent core for me to consider it one.

AMD has modules and Intel has cores. That's the difference to me. You can't and shouldn't compare them on an number basis.

Each has it's own benefits though and along with that are negatives. It's all situational.

[Kuzma]

'like'.

True

[Kuzma]

It's like cuda cores and stream processors you just can't compare. But nice write up.

 

I don't think of AMD cores as being true cores in the traditional sense though. It shares to many resources for it to be its own independent core for me to consider it one.

 

AMD has modules and Intel has cores. That's the difference to me. You can't and shouldn't compare them on an number basis.

 

Each has it's own benefits though and along with that are negatives. It's all situational.

:p The traditional sense is no longer a thing, both Intel and AMD have branched off from the original idea of a core; look at 10 years ago, a processor with an inbuilt GPU that does some calculations and process' (like QuickSync) would have sounded insane!

[leipero]

Yes, that is good explanation. The difference is not so big tho, they share L2 cache and some instructions. I always had hard time to explain to people that those are real cores, they only work in different manner compared to older architectures or current (and older) Intel architecture.

[PCBudgetSolutions]

Despite me not really being here I feel obligated to add to this just to benefit the community.  As stated an AMD FX-8350 has 8 cores, 2 cores on each modules totally 4 modules.  Each module shares L2 cache unlike Intel cores.  The reason why FX cores drop in productivity is because each module has 1 floating integer for execution and this creates the bottleneck.  The theory is, FX cores are 100% efficient at 50% load, but lose HALF it's efficiency from 50-100% load.  That is assuming, you can max out 1 core on each module up to 50%.  This effectively makes an FX-8350 about 6 cores worth of power give or take.

 

What hurts the FX cores even more is they have I believe 32 pipelines and HALF the IPC's as Intel, where Intel has I believe 12 pipelines.  Think of this like ECC ram.  Data sets have to go down the pipelines to be executed, Often times data sets go down the wrong pipeline and have to try other pipelines.  Increased IPC's makes this process more efficient.  So having MORE pipelines and far fewer IPC's is the reason why a Pentium G860 can execute a quad threaded task more efficiently than an FX-4100.  The architecture improvements in Piledriver increased IPC's and along with an unlocked multiplier (which helps on the execution level), they can offset some of this deficiency.  They also improved cache latency as well with Piledriver which helps a bit for gaming especially on the second level.  This is the more difficult side of the explanation.

[GrievingJosh]

I'm just hoping AMD comes back with something epic, and just messes Intel's stuff up. They've done it before ... I'm far fetched to believe they couldn't again. Which is awesome for consumers, bring back the competition!

[WoodenMarker]

I'm just hoping AMD comes back with something epic, and just messes Intel's stuff up. They've done it before ... I'm far fetched to believe they couldn't again. Which is awesome for consumers, bring back the competition!

Something like a dual/quad socket motherboard would be an enthusiast's dream. 

Cheap and powerful with lots of cores totally fits AMD.

[Kuzma]

Something like a dual/quad socket motherboard would be an enthusiast's dream. 

Cheap and powerful with lots of cores totally fits AMD.

That sounds crazy O.o quad socket 8350s; the heat... omg... watercooling loop would be imminent but I can imagine that ._. with quad 7970 6GB.

[GrievingJosh]

That sounds crazy O.o quad socket 8350s; the heat... omg... watercooling loop would be imminent but I can imagine that ._. with quad 7970 6GB.

Something that is fairly cheap but at the same time ridiculously powerful, I think that's AMD's sweet spot.

[IzzuddinFX]

I agree with you. The way AMD's core work is not like the way a lot of benchmark works. That's why the result is lower than the Intel's processor. I think 8350 is probably like a 'hidden monster' :P.

[Elite14]

so why are intel so much better then amd? if intel share the L3 chash were as amd has its own L2 chash and has shared L3 chash?

[Kuzma]

so why are intel so much better then amd? if intel share the L3 chash were as amd has its own L2 chash and has shared L3 chash?

Intel are so much better firstly because of the IPC and secondly because they share a lot less ^_^

[Stephie_Girl]

Despite me not really being here I feel obligated to add to this just to benefit the community.  As stated an AMD FX-8350 has 8 cores, 2 cores on each modules totally 4 modules.  Each module shares L2 cache unlike Intel cores.  The reason why FX cores drop in productivity is because each module has 1 floating integer for execution and this creates the bottleneck.  The theory is, FX cores are 100% efficient at 50% load, but lose HALF it's efficiency from 50-100% load.  That is assuming, you can max out 1 core on each module up to 50%.  This effectively makes an FX-8350 about 6 cores worth of power give or take.

 

What hurts the FX cores even more is they have I believe 32 pipelines and HALF the IPC's as Intel, where Intel has I believe 12 pipelines.  Think of this like ECC ram.  Data sets have to go down the pipelines to be executed, Often times data sets go down the wrong pipeline and have to try other pipelines.  Increased IPC's makes this process more efficient.  So having MORE pipelines and far fewer IPC's is the reason why a Pentium G860 can execute a quad threaded task more efficiently than an FX-4100.  The architecture improvements in Piledriver increased IPC's and along with an unlocked multiplier (which helps on the execution level), they can offset some of this deficiency.  They also improved cache latency as well with Piledriver which helps a bit for gaming especially on the second level.  This is the more difficult side of the explanation.

That's a mouth full, took me like 3 hours to understand it