What happened to Ivy Bridge?

With a process shrink to 22nm plus 3D tri-gate transistor technology we were told Ivy Bridge was going to be equivalent to a double jump.

The 3D tri-gate transistors alone were supposed to bring up to 50% less power consumption at the same performance level as 2D planar transistors. With a 22nm shrink on top of that what on earth went wrong?

Early reports of the 35W i7-3612QM are pretty dire stating that they aren't running any cooler than the 45W i7-3610QM. Also other reports of the i7-3720QM running hotter than the SB equivalent with minimal reductions in battery usage.

Surely the small hike in clock speeds, additional GPU power and increased thermal density due to the 22nm process shrink wouldn't cancel out the huge thermal and power benefits mentioned above?

Your thoughts?

 

I'm a bit disappointed. Also with HD 4000, while it can perform plugged in, it's gimped to run at 650MHz on battery which severely limits performance. I was hoping for gaming on battery with HD 4000 and then plugged in with dedicated GPU 650m. But no such luck.

 

Holy moley. Really? That sucks!

 

I think HD 3000 is limited on battery too, not sure. The CPU's always have been, but they are plenty powerful to begin with.

 

Isn't there any way to override that GPU clock limit? That sounds pretty dumb.

 

So far no way that I can tell to override.

edit: Hmm, seems if you disable speedstep (EIST) you can run at full speed.

 

There are a few simultaneous effects:

1) Much of the die shrink went into... well, shrinking the die. Ivy is 160mm^2, Sandy is 216mm^2 -- that's 26% smaller. The smaller the die, the more CPUs Intel gets per wafer and the higher their profit margin on each CPU.

2) The main improvement was in graphics. Intel is behind here so they are trying to catch up. Ivy is much closer to Trinity than Sandy was to Llano. The CPU did not get much attention this time around.

3) Ivy is oriented towards low voltages. Take a look at this review -- the CPUs of the 17W parts are fast. We have not seen most of the better Ivy parts yet.

4) Intel clearly had some problems with the new transistors and/or process -- Ivy got delayed and they had to make a choice between high and low voltage rather than having both of them perform well. This is the price paid for being on the cutting edge. There should be better chips later in the year.

 

So are you saying what we have now are the higher performing chips and Inte will release lower idle/full load chips later on? That's not what I understood IB has very close to idle wattage of SB, I don't think refreshed chips would be that much lower. I was hoping alot for Ivy Bridge for the X230, but I guess not.

 

My 630M clocks down to 250~ MHz on battery. No exception on nVIDIA I guess...

 

Well that's a good design though! The underclocking on battery is there for a reason: To save battery life. Laptops are meant to be portable, after all.

Mr. Mysterious

 

I can get the 650m to run full speed on battery though. No restrictions.

Good design is having flexibility and options. It should clock low with low use and have the option to run full speed if desired.

 

Yeah, I guess you're right. But it's so frustrating

 

Intel like to do the improvements step by step, or tick by tock as they call it.

The die shrink aka tick, is the preparation for the next big step with x86 performance, Haswell. The 3D transistor process needs to mature.

 

Totally agree +1

Mr. Mysterious

 

x240 tablet with 15 watt dual Core i5, here I come! Hopefully Haswell/Broadwell, and hopefully IGP becomes even more potent.

 

Intel have stated that these traits will not improve with later revisions of IB.

And Cloud I'm not talking about raw performance. I'm talking about the reduced heat output / improved battery life. Both of which have been pretty much non existent. From what is effectively a double process shrink.

 

I think Intel might have shot themselves in the foot with the idle frequencies. iirc SB idles at 800mhz and IB idles at 1200mhz and they consume around the same amount of power. If intel just clocked IB the same as SB then the battery life would probably be significantly better.

Or maybe their 22nm tri-gate just scales well enough at lower voltages that 1200mhz and 800mhz consume a similar amount of power. Someone should get an XM chip and test.

 

I don't think idle can improve much further, but yes, the chips destined for ultrabooks are not out yet.

Do you have a link?

 

Ivy Bridge (microarchitecture) - Wikipedia, the free encyclopedia

 

That's for desktop processors I believe.

 

Yes, that's for desktop processors only. It's also not what I was talking about -- I meant the manufacturing process itself should improve with later revisions (they're up to E2 or something of the sort already).

 

The part about the TIM under the IHS is yes but the rest is generic info about the IV architecture.

Intel have stated that the increased heat output of IB will likely not improve with future revisions.

It does sound as though this 3D tri-gate tech is not as advantageous as they expected.

 

I get the feeling that the engineers who designed Ivy Bridge optimized it for Intel's profit margins, not power consumption, performance, or heat.

I suppose this is just the beginning of what we should expect to see over the next few years, with AMD having no trump cards left for the foreseeable future.

 

It's disappointing and expensive.

I'd like to compare temps at full load and idle in a particular chipset 7 notebook between say a core i7 2720qm and a core i7 3610qm...Could be interesting

 

Clearly you haven't seen the benches then. Even though IVB runs a tad hotter on the desktop side, they DO orient themselves towards lower voltages. Plus their performance is outstanding. The ULV i5 IVB models handily trump the Sandy Bridge ULV models.

Even if they do run hotter I still wouldn't put this down to Intel. Notebook OEM's need to actually test their cooling solutions and compensate if they're running hot. But that would clearly cost them money, so it's unlikely. They shouldn't rely solely on TDP to determine what a notebook needs to cool a chip.

 

That is true, have seen some reports that the change in cooling paste/pad is one of the factors causing the extra heat in Ivy Bridge. Still, the heat problems is surprising considering the slightly lower power draw

 

The rest is about overclocking and overvolting which you can't do on most laptops. As I said, the architecture is clearly oriented towards lower voltages -- if you overvolt, you will have problems, but how many people actually do that on a laptop?

Again, where have they stated it and what exactly have they stated? If you mean overvolting will continue to cause problems, then yes, it's probably true -- but I don't think they've said anything about normal operations. And they will definitely release the ULV chips later on (should be quite soon now).

I think the idea itself is sound -- there doesn't seem to be a more practical way of shrinking the transistors further. The execution could use some work though.

 

Clearly you've not seen my OP as I'm not disputing that. What I am saying is where are the lower temps and super duper battery life from all these enhancements. An effective double jump/shrink should be massive.

Surely lower voltages should translate into greater battery life, or is the increased heat output cancelling out the benefits?

 

Those heat issues with the heat spreader are from desktops only. There are no heatspreaders on the mobile chips.

 

I think you guys got it wrong.

The tri-gate was never meant to reduce power consumption of the whole chip while operating. It is designed to reduce power consumption while going from "on" to "off" and reduce leakage in this transision.

You won`t see reduced power consumption until Haswell is here. Thats when we will see the increased performance/clock come in handy. We saw just that when we went from Clarksfield to Sandy Bridge.

 

You are a bit confused cloud, there are a few reasons why a transistor uses power. When "off" it leaks, when "on" it has a voltage drop and when in the middle of switching it has a higher resistance and current is flowing.

A perfect transistor would have no resistance, no leakage and would switch instantly.

A new design would have to tackle all of these and the switching power is usually the largest reason for power consumption simply due to the frequency it occurs. This is why when you "load" your cpu, more transistors are switching and more power is used than any other state.

 

Uhm, ok. I thought the "on" and "off" state was just CPU active and CPU idle and only big areas of the chip and many transistors involved, example EUs of the IGP is shifting state from "on" to "off" while cores are active.

 

You do have power gating that can switch off areas of he core with special low leakage transistors. This is the same as sandy bridge however. The main targets are leakage and switching power, but without knowing the profile of standard 22nm transistors its hard to judge how good trigate is. We always knew it was biased to lower voltages though.

 

Yeah never mind what I said earlier
Intel's Mark Bohr Explains Tri-Gate Transistors - YouTube

 

A good basic grounding of the idea =) it is just a fancy switch. 0 or 1.

 

Yes Ivy Bridge does seem underwhelming, but I think we're biased because it comes after Sandy which was both a new microarchitecture and a die-shrink (remember only Arrandale/Clarkdale were on 32nm). The IPC gain of 10-15% was not great but still very decent, combined with turbo 2.0 and improved idle consumption it was bound to be a great leap forward.

I remember at the time Sandy bridge was released and reviewed many people were downplaying it (nvm that mobile quad-cores showed +60/70% improvements ) while overstating how awesome Ivy was going to be following Intel's hype of their 22nm process and their whole "cutting standard TDP on mobile chips in half, starting with Ivy" claptrap (though I do admit, for quite a while I thought Ivy was going to bring dual cores TDP to 25W and quad cores to 35W).

But the Sandy -> Ivy transition should not be compared with Nehalem -> Sandy, but rather with Conroe -> Penryn, and then it looks decent. The problem is transistor density is way up from Sandy Bridge while consumption hasn't experienced a similar decrease, resulting in a hot chip (the fact that intel's turbo seems more agressive than ever doesn't help ; thinner cases and poor cooling designs are also to be blamed). Also they clearly concentrated their efforts on the IGP and many people who complain about Ivy don't care about it.

I think in a way Intel's grandiloquent marketing & PR is partly responsible for the overall Ivy Bridge disappointment.

 

I wish they would make proper dies with and without the integrated graphics.

 

What does it matter?

 

Like the 2011 chips, with no built in gpu circuitry.

 

They are OR gates

"This allows up to 37% higher speed, OR a power consumption at under 50% of the previous type of transistors used by Intel"

By our "logic" we all wanted AND gates?

 

Bad pun =`(

 

But it doesn't affect ANYTHING if you're not using it, so I don't see the point.

 

Intel can put higher clocks on the chips if they don't have to factor in an IGP.

Sent from my SGH-T959 using Tapatalk 2

 

Well that would only matter on the top end, the models you can overclock anyway.

 

Not really. The core i7 640M runs at 2.8 GHz and includes an IGP. If they had removed or disabled the IGP, it could have run at 3.0GHz. Obviously, I'm making up a few numbers here, but you get the point. There is more thermal room when the IGP isn't present.

Sent from my SGH-T959 using Tapatalk 2

 

What makes you think they would bump up the clock speeds of everything? Anything that is not the fastest processor in the series is obviously not being held back by TDP in the slightest. Since most processors are not being limited by their TDP anyway, why would Intel increase the frequencies if there was slightly more thermal headroom? And the fastest processors, the only ones that could possibly benefit from more thermal headroom, can have the IGP's turned off and be overclocked, making it a moot argument. So not including the IGP is an absolutely pointless move, even when you completely forget about the benefits that it provides.

 

If you don't use the IGP it's gated and consumes zero or close to zero. Therefore raising the clock speed would definitely mean higher consumption and a hotter chip. Without the IGP they could use the extra space for a bunch of stuff though.

 

I'm talking more cores, more cache, lower price.

 

I don't know how it would be lower price. It's all based on number of chips off the die, and even then Intel just artificially inflates all their pricing like crazy. Sure they could add more cores and more cache, but how much will that really help? I'm sure the marketing geniuses at Intel figure out that their sales would be significantly better if they include a decent igp than without. I'm sure there would be a specialty market for it, and surprised they don't run BOTH types of chips though to satisfy server market and what not. If only you could harness the igp for some calculations when not used.

 

That would be interesting. Remove the IGP and use the extra TDP to bump up the clocks for the people who don`t need an IGP. For example do it with the XM chips and create bigger overclock headroom.