Ivy Bridge models and clocks leaked

[cbn]

The question we have yet to answer as laypeople is how well do these things scale. Will 14nm 3D xtors have 50% lower power compared to 22nm 3D xtors? Or is this merely a one-time bump in performance/watt after which we are back to the same old scaling trend?

That is a really good question.

A couple of questions I have:

1. I wonder if Intel will be able to arrange the 3D xtors more vertically at 14nm? Stacking xtors on top of each other?

Maybe this would shorten various interconnects for power savings on a mobile platform.....reducing or slowing down the trend towards "dark silicon" at any given TDP when connected to battery power?

2. How could 14nm 3D xtors affect desktop or server applications where battery power is no longer a limiting factor, but heat density is? Maybe some type of new cooling arrangement for Generation 2 3D xtors?

 

[Blitzvogel]

yeah the better process/power use and better IGP will make this a really interesting mobile product.

HD 3000 is roughly equal to higher clocked 80 SP dedicated Radeons and 16 core Geforces, and in some games, quite a bit faster. If Intel really lives up to their promise of about 2x graphics performance (and not taking memory bandwidth into account), 160 SP Radeon 6450s and even upper end Llano IGPs (320 and 400 SP) will be threatened. AMD will need to find a better balance with graphics clock speeds on future APUs to counter the Intel threat since Intel's superior process tech allows them to push such high speeds on a relatively small GPU.

Interesting still to see once commonly mainstream parts like the Geforce 8600 GT finally, truly, utterly made archaic by an IGP that will be insanely commonplace and on large range of mainstream CPUs, unlike AMDs Llano APUs where the IGPs differ way too wildly.

 

[IntelUser2000]

In the original post there is one column named "FMB" with values such as 2011B, 2011C and so on. What does FMB and the column values mean?

Anytime you see Intel presentations with "FMB" in them its a desktop designation. FMB stands for Flexible MotherBoard, and its basically a guideline for power regulation and TDP.

B, C, D are just Intel only recognized names(and the partners building the motherboards) showing which thermal or power envelope its set for. 2011B would be a Year 2011 "B" guideline, to say it simply. If you know what 2011B/C/D is, than you know the significance of them on Ivy Bridge. As far as I know it just means its "electrical and pin compatible" with Sandy Bridge. Now that's a different matter from whether you need a new board or not but...

You do realize that all SB chips can OC, just not that high. I expect the same with IB.

Sandy Bridge's tying of everything to the bclk allows maybe 5% at best increase for overclocking beyond multiplier limitations. If rumors are true, Ivy Bridge is supposed to have something similar to Sandy Bridge E, meaning it might offer you coarse steps of bclk you can choose from, like back in the Pentium MMX days. Say 100/125/150MHz. Maybe + or - 5% from that.

I wonder if Intel will be able to arrange the 3D xtors more vertically at 14nm? Stacking xtors on top of each other?

What you are suggesting sounds like "3D stacking" which is fundamentally different from FinFET/3D/Multi Gate FET transistors. That will eventually get there, but are not the same thing. They do note multiple fins can be used to increase drive current, but that's a totally different thing with its own trade-offs. IDC can elaborate that for us if he wishes to do so.

Intel's superior process tech allows them to push such high speeds on a relatively small GPU.

Remember, GPU can't be thought like a CPU. The architectures are absolutely NOT comparable. AMD decided to use "Standard" clocks for the whole GPU, while Nvidia almost doubles the shader part while keeping the rest same. Intel keeps a "overall high frequency" approach.

AMD uses VLIW 4/5-wide architecture, while Nvidia uses something more akin to a scalar CPU(1-wide). Intel's GPUs are more like AMD's in architecture but uses high frequency, closer to Nvidia.

 

[Blitzvogel]

Remember, GPU can't be thought like a CPU. The architectures are absolutely NOT comparable. AMD decided to use "Standard" clocks for the whole GPU, while Nvidia almost doubles the shader part while keeping the rest same. Intel keeps a "overall high frequency" approach.

AMD uses VLIW 4/5-wide architecture, while Nvidia uses something more akin to a scalar CPU(1-wide). Intel's GPUs are more like AMD's in architecture but uses high frequency, closer to Nvidia.

Intel's IGPs can be seen as "half the transistors, double the clock [in turbo]."

 

[dma0991]

I'm curious as to why are there no Core i3's in the list.

 

[LoneNinja]

I'm curious as to why are there no Core i3's in the list.

Those may be the launch models for IB, I read some where that lower end parts would come in Q3 2012, so I'd expect that to include IB Pentiums and I3s. It's not a new socket, so there isn't a need to release low end chips right away when SB prices should drop for 1155.

 

[Arachnotronic]

So, couple of things:

1. Those of you holding off for Ivy probably shouldn't do so, unless >4.8GHz overclocks are needed.

2. Intel is very smart not to cannibalize their SNB-E platform with these chips @ stock.

 

[Olikan]

HD 3000 is roughly equal to higher clocked 80 SP dedicated Radeons and 16 core Geforces, and in some games, quite a bit faster. If Intel really lives up to their promise of about 2x graphics performance (and not taking memory bandwidth into account), 160 SP Radeon 6450s and even upper end Llano IGPs (320 and 400 SP) will be threatened. AMD will need to find a better balance with graphics clock speeds on future APUs to counter the Intel threat since Intel's superior process tech allows them to push such high speeds on a relatively small GPU.

is 2 times for haswell, ivy is suposed to be 60% better than sandy. (will lose to llano)

 

[zlejedi]

Those chips could have much better clockspeeds and or TDPs if it weren't for sucky IGP taking space.

I'd much prefer if they kept GPU integrated in 2 cores models which go to cheap desktops and laptops but actually make smaller quad core chip for desktop without huge part of die space.

 

[Idontcare]

is 2 times for haswell, ivy is suposed to be 60% better than sandy. (will lose to llano)

The 6 people who have been able to buy a Llano will be laughing at IB's inferiority.

 

[Sweepr]

http://img440.imageshack.us/img440/6830/intelivy.png

 

[exar333]

is 2 times for haswell, ivy is suposed to be 60% better than sandy. (will lose to llano)

60% better will play a lot of 'console'-class games at lower resolutions (think laptops here). Sure, not as good as Llano for GPU, but much better power consumption and superior CPU performance. Discrete mobile GPUs will start to be really only for high-end laptops...

 

[wlee15]

You do realize that all SB chips can OC, just not that high. I expect the same with IB.

Not the Pentium and Celeron SB since they don't have any turbo multipliers.

 

[psolord]

http://img440.imageshack.us/img440/6830/intelivy.png

Larger cache, as well as faster? Ok I get the faster part, but both are 8MB L3 chips. Unless they mention the L2 or L1, of which I am not familiar if Ivy will indeed larger caches for these.

I love those accounts that are dormant for years and jump out just to throw an important looking slide in our face, lol.

 

[piesquared]

Larger cache, as well as faster? Ok I get the faster part, but both are 8MB L3 chips. Unless they mention the L2 or L1, of which I am not familiar if Ivy will indeed larger caches for these.

I love those accounts that are dormant for years and jump out just to throw an important looking slide in our face, lol.

It also says architectural improvements AND turbo boost 2.o increase performance, so part of those improvements shown in the graph come from higher clocks. How much improvement in IPC then? 1%, 2%? Not to mention it's best case scenario provided by IntEl's own labs.

 

[aphelion02]

It also says architectural improvements AND turbo boost 2.o increase performance, so part of those improvements shown in the graph come from higher clocks. How much improvement in IPC then? 1%, 2%? Not to mention it's best case scenario provided by IntEl's own labs.

There could be zero improvement from Sandy Bridge but enthusiasts will still have no choice but to buy Intel. In that light it becomes even clearer just how much of a failure Bulldozer is: Intel can essentially coast for a generation without worries on the high-end, because frankly Bulldozer is about 2-3 generations behind Intel in terms of performance. Truly a pathetic situation.

 

[piesquared]

There could be zero improvement from Sandy Bridge but enthusiasts will still have no choice but to buy Intel. In that light it becomes even clearer just how much of a failure Bulldozer is: Intel can essentially coast for a generation without worries on the high-end, because frankly Bulldozer is about 2-3 generations behind Intel in terms of performance. Truly a pathetic situation.

And the only ones that benefit are intel and their investors. Kinda makes you wonder about the motivation behind some of these forum posts, no?

 

[Nemesis 1]

That is a really good question.

A couple of questions I have:

1. I wonder if Intel will be able to arrange the 3D xtors more vertically at 14nm? Stacking xtors on top of each other?

Maybe this would shorten various interconnects for power savings on a mobile platform.....reducing or slowing down the trend towards "dark silicon" at any given TDP when connected to battery power?

2. How could 14nm 3D xtors affect desktop or server applications where battery power is no longer a limiting factor, but heat density is? Maybe some type of new cooling arrangement for Generation 2 3D xtors?

When intel introduced the 3D gate there web site showed the next step in Intel evolution quantuam well. So I suspect will see that @ 14nm maybe 9 nm

 

[Nemesis 1]

And the only ones that benefit are intel and their investors. Kinda makes you wonder about the motivation behind some of these forum posts, no?

NO!

 

[aphelion02]

And the only ones that benefit are intel and their investors. Kinda makes you wonder about the motivation behind some of these forum posts, no?

If forum posts could affect sales, you would have sold a few thousand Bulldozers already.

 

[TuxDave]

That is a really good question.

1. I wonder if Intel will be able to arrange the 3D xtors more vertically at 14nm? Stacking xtors on top of each other?

If you're talking about die stacking or "3D chips" where chip density is the top concern, it doesn't matter if it uses planar transistors or 3D transistors. You're not stacking transistors on top of each other, you're stacking chips. The challenge that I know of is:

1) Packaging, basically how do you communicate from the top chip in the stack to the world. Currently many people just bond wires directly to the top and bottom. The next venture is to have wires burrow through the lower layers to the board. And if you choose to burrow through, the lower levels need to be aware of that since you can't plow through a bunch of logic so you need to rethink how we do floorplanning.

2) And of course there's always power because if you keep the same TDP but quadruple the chip density, you just quadrupled your problems.

 

[Sweepr]

It also says architectural improvements AND turbo boost 2.o increase performance, so part of those improvements shown in the graph come from higher clocks. How much improvement in IPC then? 1%, 2%? Not to mention it's best case scenario provided by IntEl's own labs.

Yes, there is a huge 100MHz extra bin with one core active, thats i7 3770's advantage over 2600K when it comes to clock speed. There are some nice improvements to the core, specially when it comes to FP (Cinebench and other video encoding apps will benefit). Dont you find it funny how Intel managed to squeeze more IPC from a die shrink than AMD did with a brand new "revolutionary" microarcuitetucture?

 

[StinkyPinky]

Considering I don't care about the HD Graphics nor the TDP then Ivy Bridge is a pointless upgrade path for me. As expected.

It makes sense for Intel to focus on the IGP as that is where AMD have them beat.

 

[Nemesis 1]

Well Intel has a had a few problems . But according to Robert thats now undercontrol . MB makers need a bit better bios also .

http://img440.imageshack.us/img440/6830/intelivy.png

 

[IntelUser2000]

Yes, there is a huge 100MHz extra bin with one core active, thats i7 3770's advantage over 2600K when it comes to clock speed.

IPC gains look pretty good for media applications. The Turbo Mode for 2600/2600K is already pretty good, so clock speed gains are limited.

2600/2600K

Base-3.4
4 cores-3.5
3 cores-3.6
2 cores-3.7
1 core-3.8

Last time I looked, for ones like Cinebench, cache/clock normalized performance gains for Penryn was 7-8%. Doubling divider performance is similar to what was brought in Penryn.

Not the Pentium and Celeron SB since they don't have any turbo multipliers.

If you can up the bclk, they will still be overclockable. In Sandy Bridge that's limited to under 5%, if you can choose 125MHz with Ivy Bridge, that will be potential 25%.

Those chips could have much better clockspeeds and or TDPs if it weren't for sucky IGP taking space.

That's not an issue at all. Any cost is absorbed by the huge volume of chips sold that are actually using the graphics. In that case, disabling the graphics is a better idea. For discrete graphics, power use is zero when its disabled.