Will Intel Broadwell Be Any Better Than Haswell?

[Keysplayr]

Might be time for Intel to fire up the Israeli team. (E.G. Core).

Since Core/Core2, there hasn't been much in the way of drastic improvements gen to gen. Core2Duo is STILL to this day an impressive architecture and is sufficient for 90% of personal computing functions short of super high end gaming. I'm not saying Nehalem or Sandy or Ivy aren't improvements, but I am saying it was nothing like P4 to Core. That was night and day. Now, instead of night and day, we are just getting a different time of day.
I hope Broadwell brings something special, but I can't picture it the way Nehalem - Lynnfield - Sandy - Ivy - Haswell improvements have been. At this rate, one would have to wait at LEAST five years between upgrades to notice anything appreciable for the expense and effort.
Yes, I want to be wrong. LOL.

 

[NTMBK]

Might be time for Intel to fire up the Israeli team. (E.G. Core).

Since Core/Core2, there hasn't been much in the way of drastic improvements gen to gen. Core2Duo is STILL to this day an impressive architecture and is sufficient for 90% of personal computing functions short of super high end gaming. I'm not saying Nehalem or Sandy or Ivy aren't improvements, but I am saying it was nothing like P4 to Core. That was night and day. Now, instead of night and day, we are just getting a different time of day.
I hope Broadwell brings something special, but I can't picture it the way Nehalem - Lynnfield - Sandy - Ivy - Haswell improvements have been. At this rate, one would have to wait at LEAST five years between upgrades to notice anything appreciable for the expense and effort.
Yes, I want to be wrong. LOL.

But Core 2 wasn't the successor to Netburst- it was the successor to Core, which was the successor to Pentium M, which was the successor to Pentium !!!. It's not quite as insane a jump when you look at it that way- but yes, the Israel team did a really nice job on the Pentium M family.

 

[Hulk]

I don't have high hopes for significant IPC increase for Broadwell with traditional software and by traditional I mean code not optimized for some new Broadwell instruction(s). I think we'll see perhaps 2% increase due to so cache optimizations or prefetcher logic, lower power usage, and better GPU's across the board.

The one ray of hope is the slim chance that when Intel opened up the back end on Haswell they did it with the later intention of opening up the front end as well for a 5-10% increase in IPC. But even if this might be the case, this large of a change would certainly be saved for Skylake.

The "-wells," Haswell and Broadwell are going to share very similar, if not nearly identical CPU (but not GPU) architectures.

 

[Keysplayr]

But Core 2 wasn't the successor to Netburst- it was the successor to Core, which was the successor to Pentium M, which was the successor to Pentium !!!. It's not quite as insane a jump when you look at it that way- but yes, the Israel team did a really nice job on the Pentium M family.

Whatever. As an end user, I went from Pentium 4 to Core2Duo.

 

[NTMBK]

Whatever. As an end user, I went from Pentium 4 to Core2Duo.

But there was no magic improvement that Intel pulled out of a hat- Core 2 was the result of years of steady incremental improvement, the same way Haswell is. Expecting that kind of magic leap is unrealistic.

 

[Homeles]

Might be time for Intel to fire up the Israeli team. (E.G. Core).

Since Core/Core2, there hasn't been much in the way of drastic improvements gen to gen. Core2Duo is STILL to this day an impressive architecture and is sufficient for 90% of personal computing functions short of super high end gaming. I'm not saying Nehalem or Sandy or Ivy aren't improvements, but I am saying it was nothing like P4 to Core. That was night and day. Now, instead of night and day, we are just getting a different time of day.
I hope Broadwell brings something special, but I can't picture it the way Nehalem - Lynnfield - Sandy - Ivy - Haswell improvements have been. At this rate, one would have to wait at LEAST five years between upgrades to notice anything appreciable for the expense and effort.
Yes, I want to be wrong. LOL.

I don't know if the Israeli team really is "better," or if it's just coincidence. Regardless, they're in charge of Skylake, so we'll see.

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.

Intel was trying to launch such a thing back in the Pentium 4 days, but it never came into fruition. They're still working on it, though. Let me go find it.

Here, Intel's BBUL: http://www.anandtech.com/show/834

 

[Keysplayr]

But there was no magic improvement that Intel pulled out of a hat- Core 2 was the result of years of steady incremental improvement, the same way Haswell is. Expecting that kind of magic leap is unrealistic.

Not saying it was magic. What makes you certain that there hasn't been another avenue of architecture being worked on alongside Haswell, or Skylake just like work went on as you say with PIII?

The leap from Pentium 4 to Core was enormous. No matter how long it was worked on. Do you understand what I am talking about......? There are no buts.
It matters not how many milleniuum it took to get there. Right now, we are being dribbled performance increases of as little as 5% with these new gen releases. From P4 to Core was huge.

 

[IntelUser2000]

I don't know if the Israeli team really is "better," or if it's just coincidence. Regardless, they're in charge of Skylake, so we'll see.

The way I view it:

Haifa team: CPU core
Oregon team: Platform bottlenecks

They both are very important to Intel in the long term. Nehalem itself wasn't significant either. Single thread IPC was only 5% or so(plus Turbo). But Hyperthreading, integrated memory controller, and QPI was a big thing for servers.

Haswell is similar. It brought radically lower idle power use, FiVR for fine grained power management, which enhanced notebooks(well actually for the U and Y chips) greatly. Again, not something enthusiasts are looking towards to.

What makes you certain that there hasn't been another avenue of architecture being worked on alongside Haswell, or Skylake just like work went on as you say with PIII?

Pentium 4 sucked though. Almost horrible?

Core 2 brought ~20% improvement over Core Duo. That's not something "magical". But because Pentium 4 derivatives were so bad, it looked that much better.

Again, to make it "seem" like a big improvement, Intel can do the same thing. Cap the cores and do improvements all at once. What's the point though? Results are same.

Remember the comparisons are always made versus the BEST. Core 2 in reality succeeded Athlon 64-based chips.

 

[witeken]

Might be time for Intel to fire up the Israeli team. (E.G. Core).

Since Core/Core2, there hasn't been much in the way of drastic improvements gen to gen. Core2Duo is STILL to this day an impressive architecture and is sufficient for 90% of personal computing functions short of super high end gaming. I'm not saying Nehalem or Sandy or Ivy aren't improvements, but I am saying it was nothing like P4 to Core. That was night and day. Now, instead of night and day, we are just getting a different time of day.
I hope Broadwell brings something special, but I can't picture it the way Nehalem - Lynnfield - Sandy - Ivy - Haswell improvements have been. At this rate, one would have to wait at LEAST five years between upgrades to notice anything appreciable for the expense and effort.
Yes, I want to be wrong. LOL.

There's only so much you can do, from an architectural standpoint. All of the easy performance improvement tricks are already implemented.

 

[Nothingness]

There's only so much you can do, from an architectural standpoint. All of the easy performance improvement tricks are already implemented.

Indeed, and I'm impressed by how long Intel has been able to improve things, Haswell included. Going from i7 920 to 4770K some of my number crunching code got twice faster. OK that's 2x faster for 4.5 years, but that still is very good IMHO.

OTOH I guess the current micro-architecture is reaching its limits, but I have no doubt Intel is working on some brand new micro-arch to start from.

 

[Lisathedinosaur]

I bought a i7-4770K in the summer, and IMO I do not want to buy a Broadwell.

 

[tweakboy]

I say go with Macy's when it comes out.

 

[podspi]

Not saying it was magic. What makes you certain that there hasn't been another avenue of architecture being worked on alongside Haswell, or Skylake just like work went on as you say with PIII?

The leap from Pentium 4 to Core was enormous. No matter how long it was worked on. Do you understand what I am talking about......? There are no buts.
It matters not how many milleniuum it took to get there. Right now, we are being dribbled performance increases of as little as 5% with these new gen releases. From P4 to Core was huge.

I think the thing you're missing is that it wasn't a jump from P4 to Core 2. It was Core to Core 2. It's just that nobody used Core on the desktop (though IIRC there were some motherboards starting to cater to that crowd).


Imagine this: If AMD continued developing Bulldozer and Bobcat cores, and eventually the Bobcat-derived cores became faster than the Bulldozer-derived cores, a lot faster, and AMD just decided to give up and switch everything over to the Bobcat-derived family, it might look like we had a huge jump (and in terms of market products we did) -- but from a technical standpoint it isn't really a huge leap.

 

[Lisathedinosaur]

I highly doubt Broadwell will be worth it over Haswell. From Sandy Bridge, maybe.

 

[Keysplayr]

I think the thing you're missing is that it wasn't a jump from P4 to Core 2. It was Core to Core 2. It's just that nobody used Core on the desktop (though IIRC there were some motherboards starting to cater to that crowd).

Which is why I said P4 to Core2. Core was of no perceivable significance. Not until Conroe.
I'm not missing anything guys. We have not really been wowed since then.
Incremental meh is all we've seen since the north bridge went bye bye. Not saying that is the cause mind you, just the time frame.

 

[Bryf50]

Not saying it was magic. What makes you certain that there hasn't been another avenue of architecture being worked on alongside Haswell, or Skylake just like work went on as you say with PIII?

The leap from Pentium 4 to Core was enormous. No matter how long it was worked on. Do you understand what I am talking about......? There are no buts.
It matters not how many milleniuum it took to get there. Right now, we are being dribbled performance increases of as little as 5% with these new gen releases. From P4 to Core was huge.

Looking at the AT bench, the leap from a Core 2 Quad to a 4570k is greater than the leap from a Pentium D to a Core 2 Duo. The issue is that most users don't need anything faster than a Core 2 not that there hasn't been significant performance improvement.

 

[Keysplayr]

There's only so much you can do, from an architectural standpoint. All of the easy performance improvement tricks are already implemented.

I agree with this. At least the possibility of it. Has all that could be done, been done in this architecture? All the tweaks that can be tweaked, tweaked?
Granted, there is plenty of CPU power currently so it isn't such a big deal like it used to be. But when we utilize two, three or four high end GPU's at INSANE resolutions across multiple monitors, I have to wonder if even the fastest Intel CPU can keep the GPUs fed.

 

[MongGrel]

Westmere !!!!!!

Just kidding a bit, but still liking the new X5650 I plunked in this old thing the other day, have it pretty stable at 4.3.

Is going to be interesting to watch what they do, but the power usage dropping for more mobile devices overall does seem to be more the trend.

I did like my old Q6600, had it running 4G in the day, my sister in law still using the thing for general stuff.

After getting the X5650 in this ah heck, I still need to transplant the L5639 into the 920 rig.

The wife still running one of my old Q9650 setups, I keep trying to get her to use the newer ones but she seems happy with it.

 

[Lisathedinosaur]

I am thinking that the Broadwell will have excellent graphics; however.

 

[OCGuy]

Yes

 

[jhu]

Remember the comparisons are always made versus the BEST. Core 2 in reality succeeded Athlon 64-based chips.

What I don't understand is how Core 2 was able to do this without an integrated memory controller.

 

[Arachnotronic]

What I don't understand is how Core 2 was able to do this without an integrated memory controller.

Excellent cache design ;-)

 

[Borealis7]

I know a couple of guys at Intel Haifa (Israel) and they indeed told me that they're working on Skylake, in the past they weren't so much working on the CPU core but rather on the iGPU for instance on Haswell.

What they also tell me is perhaps more interesting, that Intel these days isn't so much about pushing the envelope and innovating, but rather slow and incremental improvements.
they say that innovative ideas get bogged down by bureaucracy and that managers won't allocate time for researching better solutions and prefer sticking with the original plan with predictable results, which is understandable.

All that makes me believe that we won't see radical jumps in performance and that the average 8% SB->IVB or IVB->HSW is going to be the norm in the near future.

 

[NTMBK]

I know a couple of guys at Intel Haifa (Israel) and they indeed told me that they're working on Skylake, in the past they weren't so much working on the CPU core but rather on the iGPU for instance on Haswell.

What they also tell me is perhaps more interesting, that Intel these days isn't so much about pushing the envelope and innovating, but rather slow and incremental improvements.
they say that innovative ideas get bogged down by bureaucracy and that managers won't allocate time for researching better solutions and prefer sticking with the original plan with predictable results, which is understandable.

All that makes me believe that we won't see radical jumps in performance and that the average 8% SB->IVB or IVB->HSW is going to be the norm in the near future.

The last time Intel let its engineers have a free hand to reinvent their CPU from the ground up and chase innovative new ideas, they got Itanium and Netburst. You can imagine how that fear of radical change became ingrained.

 

[Nothingness]

The last time Intel let its engineers have a free hand to reinvent their CPU from the ground up and chase innovative new ideas, they got Itanium and Netburst. You can imagine how that fear of radical change became ingrained.

Silvermont is brand new and, though not great, not that bad (at least not as bad as Netburst or Itanium [which BTW was more an HP design than an Intel one]).