When will we see a completely new Intel uarch (similar to Netburst->Core jump)?

[Fjodor2001]

The Intel Conroe CPUs were introduced in 2006. By that Intel moved from the old Netburst to the new Core uarch generation. It was a revolutionary change in the design of their CPUs.

Since then we've basically only seen incremental improvements of the Core uarch generation (going from Conroe->...->Haswell).

So now I just wonder when it's likely we'll see a new revolutionary jump in uarch design from Intel like when transitioning from Netburst->Core? How long can they continue just making incremental updates to the now 7 year old Core architecture? Isn't a complete redesign needed at some point to get any further major increases in performance (IPC, CPU clock frequency, etc)?

 

[frozentundra123456]

The Intel Conroe CPUs were introduced in 2006. By that Intel moved from the old Netburst to the new Core uarch generation. It was a revolutionary change in the design of their CPUs.

Since then we've basically only seen incremental improvements of the Core uarch generation (going from Conroe->...->Haswell).

So now I just wonder when it's likely we'll see a new revolutionary jump in uarch design from Intel like when transitioning from Netburst->Core? How long can they continue just making incremental updates to the now 7 year old Core architecture? Isn't a complete redesign needed at some point to get any further major increases in performance (IPC, CPU clock frequency, etc)?

Well, unfortunately ipc and clockspeed aren't a priority anymore. Low power and better igp are the goals now. Seems to me the easiest way Intel could increase performance would be to release a mainstream hex core.

 

[smakme7757]

I'd imagine they have something in the works. They are constantly looking to improve their technology. We might get a few more years with incremental improvements, but I'd imagine they are always looking for a new more efficient design in their R&D department.

As to when?

That's a question for the R&D team at Intel.

 

[Lepton87]

You're wrong, C2D is just a refinement of Yonah, that is Core Duo. It's not a completely new uArch.

Read this

http://www.anandtech.com/show/2056

Revolutionary performance leap? Hardly, just 10-15% IPC bump as we are used to see.

 

[Torn Mind]

Netburst sacrificed a lot in IPC scaling--that is, increase in IPS every time you increase clockspeed 100Mhz--to increase clockspeed easily. But even so, AMD beat it with superior IPC, less power consumption, and without the need for huge clockspeeds.

If there is something that will destroy "Core", it will be Atom, since that is the only other uarch Intel is working on now, and it will take a lot for that to happen, such as everyone suddenly ditcheing PCs completely, which will not happen.

Also, it was the bringing out the Pentium M, which itself was dervied fromt he P6 uarch, that was the first step for the Core 2 Duos to make it to the market years later.

The Intel Core Microarchitecture was designed from the ground up, but is similar to the Pentium M microarchitecture in design philosophy. The Penryn pipeline is 12–14 stages long[3] — less than half of Prescott's, a signature feature of wide order execution cores. Penryn's successor, Nehalem has 16 pipeline stages.[3] Core's execution unit is 4 issues wide, compared to the 3-issue cores of P6, Pentium M, and 2-issue cores of NetBurst microarchitectures. The new architecture is a dual core design with linked L1 cache and shared L2 cache engineered for maximum performance per watt and improved scalability.

https://en.wikipedia.org/wiki/Intel_Core_(microarchitecture)#Technology

 

[jones377]

If your definition of completely new arch is some arbitrary perf/watt or IPC measurement, on legacy code no less, then I'd say never. If we look at the archs themselves then we've already had 2 radical changes since C2D with Sandy Bridge and Haswell. Nehalem brought some pretty big changes too, but mostly to the system architecture.

 

[Pandamonia]

When AMD catches up and over takes intel you will see intel wake up.

 

[Arachnotronic]

When AMD catches up and over takes intel you will see intel wake up.

I would say Haswell-ULT shows that Intel is very much awake. Intel is aiming at where the money is - mobile - and not at further extending the lead on platforms in which their lead is so huge to begin with.

 

[inf64]

Mobile is the fastest growing segment which has HUGE potential. I see intel focusing more and more on their high volume products and this will mean high perf. low power products on lower nodes (think new Atom and its successors; these will be probably used in micro servers and other markets where applicable).

 

[nonameo]

You're wrong, C2D is just a refinement of Yonah, that is Core Duo. It's not a completely new uArch.

Read this

http://www.anandtech.com/show/2056

Revolutionary performance leap? Hardly, just 10-15% IPC bump as we are used to see.

and yonah is an updated dual core pentium M, which is just a revision of the PIII, which is derived from the p-pro, which is just an x86 interpreter for a buncha RISC innards

 

[aigomorla]

wouldnt netburst be theoretically possible with today's cpu speeds?
Or is it still outside the reach?

i know the problem with the old netburst was intel could not solve a problem with heat and power in relationship to the speed required for netburst to be benifical.

However looking at how processors are today, and how they scale... wouldnt it be possible?

 

[ShintaiDK]

You are most likely never gonna see a Netburst->Core increase again with x86 again in legacy code.

 

[NTMBK]

and yonah is an updated dual core pentium M, which is just a revision of the PIII, which is derived from the p-pro, which is just an x86 interpreter for a buncha RISC innards

You missed Pentium II

 

[Arachnotronic]

You missed Pentium II

Wasn't Pentium II just a warmed over Pentium Pro (P6)? Or were there some non-trivial improvements there?

 

[Lepton87]

and yonah is an updated dual core pentium M, which is just a revision of the PIII, which is derived from the p-pro, which is just an x86 interpreter for a buncha RISC innards

That's not to say you can't you do what the OP asks, see BullDozer. Can you trace it to earlier architectures? People act like C2D was some sort of a miracle architecture when in fact it was just improved Core Duo hence Core 2 Duo.
ps. There is some evidence that BD has much in common with Nehalem but the canceled one, unfortunately nobody at the time knew that so AMD has what they do.

 

[BallaTheFeared]

Haswell is a continued push forward in efficiency at no cost to performance, Intel isn't slowing down they're just shifting focus away from "MHz" to perf/w and increased capability in lower power target devices.

Haswell is a great improvement over Sandy Bridge.

As far as the OP goes, if Intel had a better uarch they'd be using it. Instead they're finding ways to increase IPC each gen, as well as coupled with shrinks lower consumption... Which is what you'd want from a new uarch, if they have one or one comes along the same thing will occur, refinements until something new and better is discovered through R&D. I'm sure they're trying to find the next big thing, but until then all you can do is push forward with what you have.

 

[Lepton87]

I'm sure they're trying to find the next big thing, but until then all you can do is push forward with what you have.

BS! You can design BD.

 

[crashtech]

Balla, I'm glad you edited your post. Because Haswell works better than SB, but "destroy" it? Not so much.

 

[BallaTheFeared]

Balla, I'm glad you edited your post. Because Haswell works better than SB, but "destroy" it? Not so much.

Depends on how you look at it.

Performance wise it's clearly better, no argument there of course even at lower MHz.

However looking at it from a perf/w perspective, "destroys" is exactly the right word.

 

[Idontcare]

The Intel Conroe CPUs were introduced in 2006. By that Intel moved from the old Netburst to the new Core uarch generation. It was a revolutionary change in the design of their CPUs.

Since then we've basically only seen incremental improvements of the Core uarch generation (going from Conroe->...->Haswell).

So now I just wonder when it's likely we'll see a new revolutionary jump in uarch design from Intel like when transitioning from Netburst->Core? How long can they continue just making incremental updates to the now 7 year old Core architecture? Isn't a complete redesign needed at some point to get any further major increases in performance (IPC, CPU clock frequency, etc)?

We'd see it if we jumped to Itanium.

 

[aigomorla]

We'd see it if we jumped to Itanium.

oh god i can count less then one hand the people i know whose even seen a real itanium processor.

 

[cytg111]

Something like xeon phi problary, in an unrealistic future where is as easy to parallelize code as it is not to.. the next big jump is moar cores, or rather the software that runs on them.

 

[aigomorla]

Something like xeon phi problary, in an unrealistic future where is as easy to parallelize code as it is not to.. the next big jump is moar cores, or rather the software that runs on them.

its to get some form of AI, that can take ANY code universally, and then split it up efficiently without the software programer telling it how to split it up.



then yes... MOAR CORES!!
or highly dedicated and specialized one.

 

[cytg111]

its to get some form of AI, that can take ANY code universally, and then split it up efficiently without the software programer telling it how to split it up.



then yes... MOAR CORES!!
or highly dedicated and specialized one.

The singularity .. and if/when that happens we will all be invested in canned food and shotgun shells more than how many cores an intel cpu used to sport haha

 

[Hulk]

If your definition of completely new arch is some arbitrary perf/watt or IPC measurement, on legacy code no less, then I'd say never. If we look at the archs themselves then we've already had 2 radical changes since C2D with Sandy Bridge and Haswell. Nehalem brought some pretty big changes too, but mostly to the system architecture.

This is spot on.

Imagine a factory assembly line building cars. You can't speed up the line but you need to increase productivity.

You can create "substations" to build pieces of the car and get them ready. For example, you can build the engines really fast, but you have to have cars to put them in. There is only so much Out-Of-Order work that can be done. It's the same for processors especially when you are talking about legacy code.

As many around here have accurately stated there is only so much instruction level parallelism in software. Current processors are executing every bit of data as fast as it gets to the execution ports. So fast that there is unused capacity that is being used as logical "Hyperthreaded" core.

The easiest way to increase performance across the board would be to increase clockspeed. But to date that create thermal/power problems Intel feels aren't worth the effort.

The best "bang for the buck" method for increasing IPC is to do what Intel is currently doing, create new instructions (AVX, AVX2) and new ways to optimize code for multiple cores (TSX). But Intel can only do so much, the software developers have to take the ball and run with it once Intel hands it off to them.

In my opinion, when you get right down to it there are only a handful of applications most people use that could really use more performance. Software developers should devote their time on specific apps and use every hardware "trick" Intel provides. For example, how about if some software company came out with a video editor that, without GPU assist (or with only integrated GPU) was like 4 times faster at preview and render than pretty much everything out there through the aggressive use of AVX, SSE, TSX, etc...? I think if something like that were priced correctly it would sell.