Will Intel Broadwell Be Any Better Than Haswell?

[Lisathedinosaur]

Intel Broadwell is supposedly coming out in 2014-2015. But, will be any improvment performance wise over Haswell. The only things I can think of that will make it better is power consumption and built in graphics. Will Broadwell be "any" major change?

 

[NTMBK]

Better power consumption and improved graphics.

 

[Lisathedinosaur]

Better power consumption and improved graphics.

I know . I said that in my description. I was talking about performance wise.

 

[Homeles]

I'm expecting a return to Sandy Bridge-level overclocks (upper 4GHz to ~5GHz), although it's too early to say at this point. We don't know much/any of the inter-core changes, so it's not possible to comment on IPC improvements.

We'll hopefully be filled in a bit more next month at IDF, although I was hoping for them to give that debrief last fall IDF, and the spring IDF before that...

 

[jpiniero]

There's no IPC improvement. The mobile parts will probably get a couple hundred Mhz increase (at the same TDP) while Broadwell-K will get the addition of the L4 cache which should help a couple percent.

 

[ShintaiDK]

IPC might improve anyway. All the shrinks have had minor improvements. Broadwell also contains a handfuld of new instructions.

 

[witeken]

We don't know any of the architectural changes, so we can't really comment on the IPC improvements, can we?

 

[TuxDave]

We don't know any of the architectural changes, so we can't really comment on the IPC improvements, can we?

That never stopped anyone in the past.

In fact, before Haswell uArch was released, a bunch of super-nerds scoured the Intel programming reference guide to find some interesting tidbits.

 

[Magic Carpet]

I'm expecting a return to Sandy Bridge-level overclocks (upper 4GHz to ~5GHz), although it's too early to say at this point.

Indium solder will be making a comeback to mainstream, you reckon?

 

[Nothingness]

We'll hopefully be filled in a bit more next month at IDF, although I was hoping for them to give that debrief last fall IDF, and the spring IDF before that...

I couldn't find anything related to broadwell in the IDF program, but there's an Haswell session

 

[jpiniero]

I don't think we will hear much about Broadwell. They'll probably wait until the Fall IDF.

 

[escrow4]

Any increase will be pointless for Haswell users, Sandy may as well upgrade seeing as there are some large gains in some workloads (x264 encoding) and all the IPC and structural improvements combined with new extensions.

A 4770 is still freakishly fast though ***caresses his 4770's***

 

[Phynaz]

I know . I said that in my description. I was talking about performance wise.

Lower power consumption = higher performance in today's form factors.

 

[bunnyfubbles]

won't care about Broadwell, will be too busy with Haswell-E

Skylake looking like the more enticing option to return to double dip with mainstream, DDR4 + PCIe4. Although I suppose DDR4 could be a problem, as unless they increase the memory controller from dual channel to triple or quad, that means we're going to be limited to 2 DIMMs...not that 99+% of mainstream rigs will need more than 16GB that an 8GB pair (would assume to be relatively common by then) would provide, but I know its already a minor damper on X99 having "only" 4 slots for quad channel DDR4 after having up to 6 with X58 and then up to 8 with X78.

Indium solder will be making a comeback to mainstream, you reckon?

I was under the impression they got away from solder because they could, ie TDP was low enough, so unless Broadwell gets TDP up there (not sure how with a die shrink, outside of Intel going totally nuts on beefing up the iGPU) I'd think not? Although I suppose temps could also just start to get out of control at current stock clock speeds on the new process with the sub-par TIM paste to merit going back to solder despite TDP being low...

 

[Bateluer]

Too early for even AT's armchair engineers to have any kind of accuracy with their guesswork.

 

[Homeles]

That never stopped anyone in the past.

In fact, before Haswell uArch was released, a bunch of super-nerds scoured the Intel programming reference guide to find some interesting tidbits.

I'd be one of those people for Broadwell, if such a guide were released to the public. AFAIK, there isn't one.

Indium solder will be making a comeback to mainstream, you reckon?

Doubtful. My guesswork is based off traditional scaling.

 

[NTMBK]

Doubtful. My guesswork is based off traditional scaling.

Traditional scaling is dead. Intel aren't targeting their processes at higher clock speeds, they're targeting lower power consumption. I'm expecting the same overclocking story we had at 32->22- no higher clock speeds, and more difficulty dissipating heat due to thermal density.

The only market segment that cares about higher clock speeds is us, and we just don't matter in Intel's grand plan. They want lower power consumption for tablets and thin and light laptops, and they want more cores on servers and workstations. Ivy Bridge was a big yawn on the desktop, but on workstations it added 50% more cores- expect more of the same from Broadwell.

 

[witeken]

I don't think we will hear much about Broadwell. They'll probably wait until the Fall IDF.

What about Computex?

 

[mikeymikec]

I've lost track, is Broadwell a 'tick' or 'tock' generation?

- edit - answered my own question:
http://en.wikipedia.org/wiki/Intel_Tick-Tock

"Every "tick" is a shrinking of process technology of the previous microarchitecture (and sometimes introducing new instructions as with Broadwell)"

So no, I don't think Broadwell is intended to bring anything significantly new/better to the table, just some minor improvements to the existing tech.

 

[Charlie98]

The only market segment that cares about higher clock speeds is us, and we just don't matter in Intel's grand plan. They want lower power consumption for tablets and thin and light laptops, and they want more cores on servers and workstations. Ivy Bridge was a big yawn on the desktop, but on workstations it added 50% more cores- expect more of the same from Broadwell.

^^^ :\

 

[NTMBK]

^^^ :\

lol, sorry to be a downer.

 

[Homeles]

Traditional scaling is dead.

Cliched phrase...

Intel aren't targeting their processes at higher clock speeds, they're targeting lower power consumption. I'm expecting the same overclocking story we had at 32->22- no higher clock speeds, and more difficulty dissipating heat due to thermal density.

Thermal density is a red herring.

As far as clock speed scaling goes, shrinking transistors DOES give us increased performance today. However, it's not just from shrinking. Reformulating HKMG, strain, FinFETs... there's plenty of room for improvement.

The only market segment that cares about higher clock speeds is us, and we just don't matter in Intel's grand plan. They want lower power consumption for tablets and thin and light laptops, and they want more cores on servers and workstations. Ivy Bridge was a big yawn on the desktop, but on workstations it added 50% more cores- expect more of the same from Broadwell.

So? We're still along for the ride.

 

[NTMBK]

As far as clock speed scaling goes, shrinking transistors DOES give us increased performance today. However, it's not just from shrinking. Reformulating HKMG, strain, FinFETs... there's plenty of room for improvement.

There's room for improvement, sure, but that requires Intel improving it. I'm just pessimistic that Intel will target improved maximum clock speeds for a new process; I would expect them to hold maximum clock speeds constant while improving energy consumption, because that aligns with their current goals.

So? We're still along for the ride.

Sure we're along for the ride. Unfortunately this bus is going to "Lower Power Consumption Station", not "Higher Clock Speed Central".

 

[Homeles]

They're not mutually exclusive.

 

[know of fence]

Indium solder will be making a comeback to mainstream, you reckon?

Indium is just too precious to be wasted as a TIM. What we need is a new bare die design with a metal shim or ridge, just like AMD GPUs. There is no better opportunity to make it happen than the release of Broadwell-K parts, targeting the high end.