Intel microarchitecture: Nehalem v. Skylake

[destrekor]

And, just because, let's add the original Core series, like Conroe, to the mix.


No, I am not interesting in comparing performance or features, that is obviously futile.

What I am most interested in, is whether the microarchitecture has changed radically over the generations?

What I've heard is that Intel hasn't truly re-engineered the architecture since Pentium M/Yonah and Nehalem (the latter borrowed from both the P6-derived Yonah and Netburst architecture), at least not to the same degree. I understand they have iterated and enhanced the architecture, but... can the entire recent "Core" series be said to be related to the original Nehalem architecture?

I guess my main question is whether one could say Intel could radically re-engineer the architecture to be a totally new design, and not one that is iterative of past designs?

 

[hrga225]

Well,if you look at processor floorplans,I would say that Intel's designs haven't changed radically in 20 years.Or to recap: Enough steps in evolution after some time look like revolution.

 

[frozentundra123456]

Yea, I wonder about that too. Is it time for a total redesign, or is the core architecture just so good that a total redesign would not improve that much, even after all these years. They certainly seem to be getting less and less improvements gen to gen with die shrinks and tweaking the architecture.

 

[ShintaiDK]

Its a bit like saying cars haven't changed since the first one

 

[Edrick]

Its a bit like saying cars haven't changed since the first one

Exactly.

The OP needs to define "re-engineered". I can argue that they have not re-engineered the x86 core since 1979, but rather a series of enhancements.

 

[frozentundra123456]

Its a bit like saying cars haven't changed since the first one

Huh???? Nobody is saying skylake has not changed from original core or Pentium M. The question is whether a new architecture could give much bigger improvements than continuing to refine an architecture that has been tweaked over and over.

Lets switch to a sports analogy. Is having your current quarterback (cpu) get just a bit better every year enough, or do you need an entirely new one?

 

[coercitiv]

Lets switch to a sports analogy. Is having your current quarterback (cpu) get just a bit better every year enough, or do you need an entirely new one?

Yeah, let's equate a quarterback to a finite-state machine and see if we need to replace it.

 

[ShintaiDK]

Huh???? Nobody is saying skylake has not changed from original core or Pentium M. The question is whether a new architecture could give much bigger improvements than continuing to refine an architecture that has been tweaked over and over.

Lets switch to a sports analogy. Is having your current quarterback (cpu) get just a bit better every year enough, or do you need an entirely new one?

The point is that the uarchs compared are radically different. To say they are the same is exactly like looking at an object with 4 wheels and an engine and say its a car while disregarding everything else.

A completely new fresh architecture isn't going to do anything besides trade a few pros and cons at best. Unless you go radically different like the IA64.

 

[Edrick]

The question is whether a new architecture could give much bigger improvements than continuing to refine an architecture that has been tweaked over and over.

Intel already tried that.....it failed.

In fact, many companies have tried that and failed.

 

[jhu]

Intel already tried that.....it failed.

Depends on how you look at it. The argument that Itanium was a resounding success can be easily made.

 

[deasd]

Cache/buffer coherency hasn't been change since Conroe. Nehalem has extra registers to handle another thread, but since then there was hardly any enhancement in ILP, even later generations have larger ROB and multiple buffer to help pre-R/W, and much wider machine(more ports, longer SIMD), it cannot reduce latency significantly(OTOH it exchange more performance with longer SIMD but draw much power than normal situation).
I believe this architecture already hit a wall, even if there will have longer SIMD and wider decode in the future but it's diminishing return.

 

[carop]

I understand they have iterated and enhanced the architecture, but... can the entire recent "Core" series be said to be related to the original Nehalem architecture?

In terms of superscalar hardware data structures, Nehalem was a Data-in-ROB (ReOrder Buffer) design, whereas Sandy Bridge (re)introduced the Unified-PRF (Physical Register File) design.

- Data-in-ROB: P6 up to Nehalem and Westmere
- Unified-PRF: P4, Sandy Bridge up to Skylake

The “Sandy Bridge” microarchitecture introduces the concept of Physical Register File (PRF) to replace Nehalem’s centralized Retirement Register File (RRF).

This approach is substantially more power efficient because it keeps a single copy of every data and eliminates the movement of data values after calculation. PRF is a key enabler of making Out of Order unit (OOO) larger and for Intel Advanced Vector Extensions (Intel AVX). PRF is enabled in the Sandy Bridge processor by default.

http://software.intel.com/en-us/art...e5-26004600-product-family-technical-overview

 

[el etro]

I think the OP said how much the architecture changed in a very high level overwiew.


Once i'm only a enthuasiast, i think that from P6 to Pentium 4 was the biggest redesign, far ahead of the Conroe to Nehalem transition. Third one would be Nehalem to Sandy Bridge because it give a hige efficiency boost in most TDP tiers.

 

[MrTeal]

Depends on how you look at it. The argument that Itanium was a resounding success can be easily made.

The argument could be made that Itanium was a resounding success, or that the Itanium project was a resounding success?

The chip itself seems have been a pretty complete failure.

 

[NTMBK]

A completely new fresh architecture isn't going to do anything besides trade a few pros and cons at best. Unless you go radically different like the IA64.

Because that train wreck went oh so well!

 

[NTMBK]

Once i'm only a enthuasiast, i think that from P6 to Pentium 4 was the biggest redesign, far ahead of the Conroe to Nehalem transition. Third one would be Nehalem to Sandy Bridge because it give a hige efficiency boost in most TDP tiers.

Conroe to Nehalem was a massive redesign for everything but the CPU core. P!!! to P4 maintained the same basic FSB architecture, but Nehalem changed the way things had been done since the Pentium 1 days.

 

[Tuna-Fish]

What I am most interested in, is whether the microarchitecture has changed radically over the generations?

Yes.

What I've heard is that Intel hasn't truly re-engineered the architecture since Pentium M/Yonah and Nehalem

You've heard wrong.

Sandy Bridge was a total redesign of the organization of the core. The parts used inside stayed the same, but the mechanism used for OoO and thus the glue that holds it all together was completely new. In a way, the OoO system of Sandy Bridge is closer to Pentium 4 than it is to Nehalem.

 

[Dresdenboy]

@OP: For starters, let me quote one of my older posts here. It's more about the core layout, but the given link will provide you information about the microarchitecture.

This core heritage goes back to the P6. Since Banias the basic core layout (with a bit shuffling) didn't change dramatically. Well, even the PIII shows roughly the same layout. It's FP/SIMD units sit in one corner, the RS below, I$ diagonally located from FP/SIMD, D$ on the same row. It depends on orientation and mirroring, but after normalizing for this, it fits. When the smaller L2 reentered the rectangular core area, it did that on the D$ side, while it was on the I$ side in PIII.
PIII:




On that Hungarian site you can also find the architectural evolution from P6 to Skylake:
http://prohardver.hu/teszt/intel_architekturak_nehalemtol_skylake-ig/nyomtatobarat/teljes.html
(print view, easier to watch and scoll)

 

[destrekor]

I guess I'm just curious if Intel has had something as drastic of an overhaul as AMD switching to the Zen architecture, which is radically different compared to the architecture they've been using post-Athlon XP.

 

[destrekor]

And I might also add a drastic enough change to create a large performance delta between generations. Think about the jump from Nehalem to Sandy Bridge, even... has there been that much of a performance increase since then? And IIRC Nehalem from Core 2 was nearly as good, while the Prescott -> Banias > Yonah was equally as radical and impressive.

While there have been performance and power use improvements since then, I just don't recall any drastic jumps in performance compared to previous efforts.

 

[NTMBK]

I guess I'm just curious if Intel has had something as drastic of an overhaul as AMD switching to the Zen architecture, which is radically different compared to the architecture they've been using post-Athlon XP.

AMD already made a dramatic jump when they went from K10 to Bulldozer.

 

[TheRyuu]

Well at the very least you can look at how Intel has had to change things to allow AVX/AVX2 to function (which required the doubling of some things like register widths). Now they're introducing AVX-512 which once again will double the register size among other things.

 

[William Gaatjes]

The point is that the uarchs compared are radically different. To say they are the same is exactly like looking at an object with 4 wheels and an engine and say its a car while disregarding everything else.

A completely new fresh architecture isn't going to do anything besides trade a few pros and cons at best. Unless you go radically different like the IA64.

I agree, There are improvements like super scalar, out of order execution, branch prediction, register renaming, execution trace cache, clock gating... Comparing old architectures to modern architectures is not easy. It really looks like a whole new different species. for example Pentium vs Pentium pro was very different. Some would even say radically different.

 

[Ken g6]

I guess I'm just curious if Intel has had something as drastic of an overhaul as AMD switching to the Zen architecture, which is radically different compared to the architecture they've been using post-Athlon XP.

Well, there's Atom.

 

[destrekor]

Well, there's Atom.

True! lol

That has been coming a long way as of late, and the next generation Atom (especially for micro servers) should be leaps and bounds ahead of older generations.

Between the Atom series and the Xeon D series, Intel is seriously gunning for the small low-power server, especially for those customers who utilize large-scale clustering of low-power chips, such as Facebook. They got beaten to the punch by ARM licensees, and are trying to right that ship.