Will Haswell and Ivy Bridge overclocking properties be similar?

[Fjodor2001]

Hi,

I just wonder if it's likely that Haswell will have the same overclocking properties that Ivy Bridge has? After all they will both use the same 22 nm process, tri-gate transistors, etc?

What do you think?

 

[SickBeast]

I hope not but probably.

 

[Don Karnage]

Yes. Same process

 

[joshhedge]

Haswell is a Tock right? So even if it overclocks as much as Ivy Bridge the performance would be greater clock for clock.

 

[Concillian]

Nobody knows. Haswell is ~a YEAR away.

There are plenty of examples where a new stepping of an existing CPU comes out and it gains 10% or more OC margin.

For all we know Ivy Bridge overclocking in 6 months won't be like Ivy Bridge overclocking now. Nobody can even pretend to know about Haswell.

 

[Fjodor2001]

But you can make an educated guess based on previous CPUs. I.e. based on previous experiences, does a Tock usually overclock about the same as the corresponding Tick that use the same manufacturing process?

 

[jpiniero]

I think there's a good chance that Haswell's overclocking will be worse. The new features, L4 and included VRM are only going to add more heat.

 

[Arzachel]

Short answer: No.

Long answer: Hell no. The process is going to be more mature, die size is going to be different, the iGPU will take up a different amount of the die, memory on a silicon interposer could change cooling properties, not even mentioning the overhauled arch that could totaly change baseline clocks.

 

[bryanW1995]

Haswell should overclock more like SB than IVB. Based upon everything I've heard about IVB, that would be a good thing.

 

[gplnpsb]

But you can make an educated guess based on previous CPUs. I.e. based on previous experiences, does a Tock usually overclock about the same as the corresponding Tick that use the same manufacturing process?

It varies. Conroe clocked significantly worse than Cedar Mill at 65nm, but it also brought substantial IPC improvements on a scale not seen since.

At 45nm, the initial C0 revision of Nehalem clocked similarly to the C0/C1 Penryn and Penryn CPUs, but perhaps a little poorer than the E0 revisions of those chips. Nehalem's D0 stepping improved clocks to the point that they met or exceeded the capabilities of many E0 Penryns. Keep in mind that Nehalem's die size was 263 mm2, vs 107 mm2 for Wolfdale, the dual core Penryn. The combined area of the two Penryn dies used for quad core Yorkfield CPUs was still smaller than Nehalem. We never saw how well Havendale - the dual core Nehalem - clocked, as it was cancelled at the B-0 stepping.

At 32nm, Sandy Bridge improved clock potential by about 500 MHz over the six core Westmere processor. (ie 4.2GHz to about 4.7 GHz)

There is truly a multitude of factors that affect clock potential, so it really is impossible to say how Haswell will clock at this point.

 

[Concillian]

But you can make an educated guess based on previous CPUs. I.e. based on previous experiences, does a Tock usually overclock about the same as the corresponding Tick that use the same manufacturing process?

Based on previous experiences, does a Tick usually clock worse than the previous Tock?

I'm not sure past experience is really something to work from in this case. Fin-fet is a little bigger departure from the norm than your typical shrink or even architecture change.

 

[Fjodor2001]

It varies. Conroe clocked significantly worse than Cedar Mill at 65nm, but it also brought substantial IPC improvements on a scale not seen since.

Comparing the jump from CedarMill->Conroe, to IvyBridge->Haswell is completely pointless.

Conroe was a completely redesigned architecture leaving behind the old Netburst based architecture which was based on high clocks and high power consumption. It was a paradigm shift.

The architechture change going from Ivy Bridge->Haswell is much more subtle. It is more fair to compare it to the Clarkdale->SandyBridge architechture change.

At 32nm, Sandy Bridge improved clock potential by about 500 MHz over the six core Westmere processor. (ie 4.2GHz to about 4.7 GHz)

But how does SandyBridge overclock compared to Clarkdale? I think that would be a better estimate of how Haswell will overclock compared to Ivy Bridge.

 

[IntelUser2000]

I think there's a good chance that Haswell's overclocking will be worse. The new features, L4 and included VRM are only going to add more heat.

I doubt you will see the on package DRAM on desktop parts.

VRM will lower power usage, and new architecture might lower it further. The real question with overclocking is how the integrated VRM will handle overclocking current.

The architechture change going from Ivy Bridge->Haswell is much more subtle. It is more fair to compare it to the Clarkdale->SandyBridge architechture change.

I don't think there's any architecture I can think of that was big of a change as going to/going away from Netburst.

Ivy Bridge to Haswell will probably be like Penryn to Nehalem. That's even smaller core change, but big platform change.

 

[jpiniero]

VRM will lower power usage, and new architecture might lower it further. The real question with overclocking is how the integrated VRM will handle overclocking current.

System power usage, sure, but there's got to be a cost to the CPU for moving the VRM in there.

It sure looks like Intel is moving to SoC for all of it's processors.

 

[Don Karnage]

Not sure why everyone is so unhappy with ivy. Yes it runs hot when overclocked but the 3770 should be a multithreaded monster.

Buy a 120 dollar rasa kit and a 3570K and be happy

 

[Edrick]

By the time Haswell comes out, the 22nm (tri gate) process will be much more mature. All the kinks should be worked out by then.

And until we start seeing ES leaks, we can only assume.

 

[Fjodor2001]

Haswell will also add stuff that increases the amount of heat: L4 cache, much larger IGP, etc. So chances are it will overclock worse than Ivy Bridge. We'll see...

 

[Borealis7]

I asked my cousin about IVB, he was on the Sandy Bridge IGP team at Intel. He said that any node shrink ("ticks") would not be good for overclocking because of die area reduction and node not yet being mature enough (even though operating voltage decreases with each node) but rather offer bug fixes to existing architecture and slight tweaks to IPC in the range of 5-10 percent.
However, "Tocks" will offer much more in terms of overclocking as the process matures. so Haswell may still be another step forward in terms of clockspeed.

that being said, if he offers to get me a 3570k with intel's employee discount i will not say no...

 

[BallaTheFeared]

Overclocking headroom isn't a kink though...

I still plan to get one because I bought a $120 rasa kit like Don said, and added a 360 rad, and two full cover gpu blocks... Great value in that kit, one of the best things I bought.

 

[Tuna-Fish]

Hi,

I just wonder if it's likely that Haswell will have the same overclocking properties that Ivy Bridge has? After all they will both use the same 22 nm process, tri-gate transistors, etc?

Processes are not typically the same from inception until EOL. Given how dramatically Intel's 22nm departs from earlier ones, I find it entirely believable that it will gain a lot from a year or so of tweaking and testing in volume production.

Of course, this does not lead to miracles.

 

[Tuna-Fish]

System power usage, sure, but there's got to be a cost to the CPU for moving the VRM in there.

I find it entirely believable that going from power distribution at 100A to one at 10A will net reduce power use on chip, even if the VRM losses are shifted on-chip.

As IntelUser2000 said, the big question is how much safety margin Intel will give that on-chip VRM, and how much of it can be reclaimed for OC use.

 

[SickBeast]

So say Haswell eeks out 500mhz more from the 22nm process. Will you guys be happy with 4.8-5.3ghz? I find it a little disappointing.

 

[BallaTheFeared]

Well since Hasell is a new uarch there is really no way to know what kind of clocks we'll see or if +/- what we're getting now will means anything.

Don't get caught up on clock speeds.

 

[Tuna-Fish]

So say Haswell eeks out 500mhz more from the 22nm process. Will you guys be happy with 4.8-5.3ghz? I find it a little disappointing.

As a programmer, I am a lot more interested in the new instructions set extensions supported by Haswell than on improvements on legacy code. As far as I'm concerned, it could degrade performance on old apps and it would still be the best thing since sliced bread.

 

[Don Karnage]

So say Haswell eeks out 500mhz more from the 22nm process. Will you guys be happy with 4.8-5.3ghz? I find it a little disappointing.

Why is everyone expecting Haswell to overclock better? Same process but an enhanced IGP and VRM's on Die. It'll probably overclock worse.

If Haswell will add another 8-10% IPC i won't even care if it can hit 4.8Ghz