Intel "Haswell" Speculation thread

[Arachnotronic]

So, with SNB pretty much whipping butt and taking names as a crazy efficient uArch, what's next for Intel to improve? Do you think we'll see significant IPC gains in "Haswell", or do you think we'll see kind of the same thing at lower power levels ala Ivy Bridge?

Is there anything in SNB that is glaringly flawed and needs to be improved, or will an IPC improvement from here require some really serious work?

 

[Don Karnage]

I see haswell having another 5-7% IPC jump over IB. It'll be more towards IGP improvement then anything else.

 

[Arachnotronic]

That's awfully disappointing if true, but I wouldn't doubt it. At least my Gulftown will be modern for a very long time...

 

[janas19]

I see haswell having another 5-7% IPC jump over IB. It'll be more towards IGP improvement then anything else.

With SB doing so well, what's the impetus for Intel to significantly improve performance in anything other than incremental steps?

My prediction: a substantial increase in IGP performance, only moderate improvements in TDP and processor performance.

 

[BallaTheFeared]

5-7% is less than what we're getting from 32nm to 22nm with Ivy Bridge..

If Intel can't figure out how to come up with another 17%~ or more than there is no real reason to come out with a new uArch.

I would think they'd still get the 15-20% IPC increase in, with more focus on the TDP and IGPU.

 

[janas19]

5-7% is less than what we're getting from 32nm to 22nm with Ivy Bridge..

If Intel can't figure out how to come up with another 17%~ or more than there is no real reason to come out with a new uArch.

I would think they'd still get the 15-20% IPC increase in, with more focus on the TDP and IGPU.

Yeah, you might be correct, since Haswell was in development long before Sandy Bridge launched.

My question would be, if AMD doesn't show significant competition in the desktop market going forward, will Intel really have the fire to keep making great stuff? Doubt it. Remember the Core cpus came about due to the failure of Netburst primarily. It was AMD's moves that kept them on track. Intel is primarily a corporation, who knows if they have any reason to make Skylake superior.

 

[Don Karnage]

5-7% is less than what we're getting from 32nm to 22nm with Ivy Bridge..

If Intel can't figure out how to come up with another 17%~ or more than there is no real reason to come out with a new uArch.

I would think they'd still get the 15-20% IPC increase in, with more focus on the TDP and IGPU.

From what i've seen IB ES are performing 6% better then SB clock for clock. Anyone complaining about that from just a die shrink needs their head examined.

 

[TuxDave]

http://www.dailytech.com/Intel+Says...val+Todays+Discrete+Graphics/article22002.htm

its upcoming 22 nm Haswell architecture, currently under development by the company's secret Oregon team

Secret? lol

 

[Don Karnage]

http://www.dailytech.com/Intel+Says...val+Todays+Discrete+Graphics/article22002.htm



Secret? lol

It's like Area 51. It's a secret but we all know about it

 

[janas19]

http://www.dailytech.com/Intel+Says...val+Todays+Discrete+Graphics/article22002.htm

This just goes to prove my point exactly, which is Intel just wants to be better than its competition. AMD has better IGP now, so that's what Intel wants to improve upon. Once they become the leader, their marketing department takes over R&D, they stagnate, and then their product quality starts to tank again.

 

[Lonbjerg]

This just goes to prove my point exactly, which is Intel just wants to be better than its competition. AMD has better IGP now, so that's what Intel wants to improve upon. Once they become the leader, their marketing department takes over R&D, they stagnate, and then their product quality starts to tank again.

Then blame AMD for Dulldozer...

 

[frostedflakes]

5-7% is less than what we're getting from 32nm to 22nm with Ivy Bridge..

If Intel can't figure out how to come up with another 17%~ or more than there is no real reason to come out with a new uArch.

I would think they'd still get the 15-20% IPC increase in, with more focus on the TDP and IGPU.

IIRC Haswell products are going back to 95W TDP for mainstream desktop. If it brings more cores and/or solid IPC improvements, though, I doubt anyone will complain.

 

[BallaTheFeared]

It must bring something, since IB is giving a 6%~ IPC increase with 19% reduced TDP.


Intels numbers state more than 6%, but it remains to be seen (IB).

 

[MrTeal]

Even if AMD brings nothing to the table, Intel still has incentive to produce new and more powerful chips.

Sure, short term people might be retiring old servers and computers and would buy whatever is the best value for them, but if they don't continue raising the bar even relative to themselves there becomes no incentive to upgrade. Even if they have the best chips now and for the next year with SB/IB, if Haswell isn't a step forward vs those chips there's no real reason to purchase them.

 

[ThatsABigOne]

I am really hoping for Haswell to be just the right chip to retire my Core i7 960.

 

[MaxPayne63]

Amusing how, the way some people tell it, Bulldozer managed to be the final nail in the coffin for AMD and Intel.

 

[janas19]

Amusing how, the way some people tell it, Bulldozer managed to be the final nail in the coffin for AMD and Intel.

You think differently?

 

[gevorg]

I think Haswell will offer very little IPC improvement over Ivy Bridge, mainly because its not a die shrink and its mostly concentrated on polishing the architecture with new features like DirectX 11.1, better IGP, better power handling, etc. Haswell will mainly benefit laptops.

Broadwell (14nm) on the other hand should make a big step up in IPC like Sandy Bridge did.

 

[TuxDave]

This just goes to prove my point exactly, which is Intel just wants to be better than its competition. AMD has better IGP now, so that's what Intel wants to improve upon. Once they become the leader, their marketing department takes over R&D, they stagnate, and then their product quality starts to tank again.

No doubt IGP improvement is on the agenda but if you go to that page I linked and clicked the next link....

http://software.intel.com/en-us/blogs/2011/06/13/haswell-new-instruction-descriptions-now-available/

These are all new core instructions, not IGP. I guess none of it sounds exciting to you?

(and btw, in all my career, I never had a person from marketing tell me to stop doing stuff )

 

[janas19]

No doubt IGP improvement is on the agenda but if you go to that page I linked and clicked the next link....

http://software.intel.com/en-us/blogs/2011/06/13/haswell-new-instruction-descriptions-now-available/

These are all new core instructions, not IGP. I guess none of it sounds exciting to you?

(and btw, in all my career, I never had a person from marketing tell me to stop doing stuff )

Meh. Like someone else mentioned they are just polishing IB architecture, not rebuilding from the ground up.

 

[TuxDave]

Meh. Like someone else mentioned they are just polishing IB architecture, not rebuilding from the ground up.

I don't know what % of the design is built from the ground up but it's definitely not 100%. On the other hand, from a project management perspective it's impossible to have Haswell be a polished version of Ivybridge. (you figure it out)

 

[janas19]

On the other hand, from a project management perspective it's impossible to have Haswell be a polished version of Ivybridge. (you figure it out)

Not totally, I agree. I was just referring to the instruction in the link. Was very informative link though, thank you.

 

[BallaTheFeared]

IB is just SB with some new toys and tricks... I thought Haswell was far more of a redesign/tweak than SB>IB was...

Haswell will have a 2011 type chip/socket as well without a IGPU just like SB does/did.

SB had something like a 17% lead per clock per core over 32nm Westmere, with better power to performance and improved clock speeds.

Compared to core2 it seems reasonable to assume generally the same thing will take place as has been taking place since core2, that is 15-20% IPC increases, better power to performance ratios, and generally new instruction sets that will further increase it's performance potential in real world applications.

It has been reported that Nehalem has a focus on performance, thus the increased core size.[4] Compared to Penryn, Nehalem has:
10-25% more single-threaded performance / 20-100% more multithreaded performance at the same power level
30% lower power usage for the same performance
Nehalem provides a 15–20% clock-for-clock increase in performance per core(average)

AVX2 which is basically AVX for Integer work loads should improve performance over SSE, or at least I would think... AVX was a bust for most of us since FP isn't very common, nor is it typically the limiting factor in performance. Of course that's the same forward thinking Bulldozer incorporated, however just like bulldozer the new instructions won't mean anything for legacy and even current/future software until it's coded.

 

[Maximilian]

6-8 physical cores
Higher clockspeed
Better IPC
Higher overclock potential
HT on all CPUs
Better IGP capable of crysis on medium/high
All CPU's unlocked
Lower load & idle power consumption

All for the low low price of £140!

This is what i expect from Haswell. Should my expectations be dashed i will look like this guy --> D:

 

[BallaTheFeared]

Eww, HT... Do not want.