Intel Broadwell Thread

[Homeles]

In a few days
"Lenovo unveils ThinkPad Helix hybrid with Intel Core M chip"

"The Windows hybrid will be available to buy in October, with prices starting at $999."

http://www.itpro.co.uk/laptops/23043/lenovo-unveils-thinkpad-helix-hybrid-with-intel-core-m-chip

Nice. No need to worry about them making holiday season, then.

 

[frozentundra123456]

Nice. No need to worry about them making holiday season, then.

That looks nice, but expensive. I would only get a convertible device like this if the "tablet" part completely detaches, which it looks like it does in this one. At first I thought all the convertibles were great, but as long as the display is attached to the keyboard, at least the models I have handled, it just does not work very well as a tablet. You still have the problem though of size. Anything that is small enough to work as a tablet, is a bit small for a laptop, and vice/versa.

With all the hundred to one fifty dollar win tablets coming out now, I am still not sure it isnt a better deal to just get a separate laptop and a cheap tablet for light on the go use.

 

[Phynaz]

Yeah, convertibles are DOA for me.

 

[Fjodor2001]

Yes, agreed - no convertibles please. This is what I need:

* Fanless
* Broadwell-M 1.1/2.6 GHz
* 8 GB RAM
* 256-512 GB SSD
* Matte 1080p+ display, with high contrast, good viewing angles, 16:10 aspect ratio
* Not convertible, just a good plain Ultrabook
* 13"
* Good connectivity:
-2-3x USB3, and ideally at least one USB 3.1 Type C connector too
-Gigabit Ethernet LAN
-Full size HDMI 2.0/DVI/VGA connectors (VGA needed for office projectors unfortunately). Optionally VGA can be skipped if chassis would be too big. Can be added through USB instead then, although not ideal to carry around USB accessories.
-Card reader slot, including SD card
* Good WLAN connectivity, including 802.11ac @ 1300+ mbit/s
* Battery life 10+ h at standard light weight use cases
* Weight 1.1 kg or less
* Nice design, preferably metal chassis, but no fingerprint magnet
* Very high quality keyboard and touchpad
* Price: Ideally $1000-1500, max $2000

Did I miss anything? Anything anyone would like to change on your Broadwell Ultrabook?

 

[IntelUser2000]

it just does not work very well as a tablet. You still have the problem though of size. Anything that is small enough to work as a tablet, is a bit small for a laptop, and vice/versa.

At least weight and thickness wise, Broadwell Convertibles should be more of an ideal than ever before. The *COMBINED* weight of the UX305 is said to be noticeably less than 1kg, probably around 900g. Likely a true convertible(not a detachable) can be made lighter than this because that was the case in the Haswell generation(detachables with keyboards are bulkier than pure convertibles).

Pricing on the Helix is FAR better than on the first version. The first version started at $1399 or so. You can get to $1700 pretty easily. I hope Intel's talks about "premium" devices like the UX305 coming in the $799 price range comes true.

 

[Enigmoid]

Interesting fabrication and process density comparison.

https://www.semiwiki.com/forum/content/3884-who-will-lead-10nm.html

Edit: Looks kinda funky.

 

[witeken]

Interesting fabrication and process density comparison.

https://www.semiwiki.com/forum/content/3884-who-will-lead-10nm.html

Edit: Looks kinda funky.

"The same data is also plotted below as the now infamous Intel density comparison:"

Nice one.

 

[Arachnotronic]

"The same data is also plotted below as the now infamous Intel density comparison:"

Nice one.

The problem with this article is that that TSMC/Samsung gate pitch * metal pitch gets worse from 20nm -> 14/16nm in his graph.

This does not line up with reality.

It's also probably helpful to offset these in time since Intel 22nm competed with TSMC/Samsung 28nm, and Intel 14nm looks to go up against 20nm for quite some time.

Just my two cents though.

 

[witeken]

The problem with this article is that that TSMC/Samsung gate pitch * metal pitch gets worse from 20nm -> 14/16nm in his graph.

I guess the graph is from an earlier version of the article:

This table has been updated since the original post based on measured TSMC 28nm and 20nm pitches from Chipworks.

But the graph probably hasn't been updated.

 

[sm625]

Mark Bohr has officially confirmed the 1.3B transistor count (1.9B was a typo).

No way. So Intel thinks they are going to get away with charging $200-$400 for a 1.3 billion transistor part, when Apple is currently paying TSMC and Samsung probably 1/10th as much per transistor. And all the other ARM competitors will be selling 1+ billion transistor chips for even less. No way is that going to continue forever.

 

[witeken]

No way. So Intel thinks they are going to get away with charging $200-$400 for a 1.3 billion transistor part, when Apple is currently paying TSMC and Samsung probably 1/10th as much per transistor. And all the other ARM competitors will be selling 1+ billion transistor chips for even less. No way is that going to continue forever.

Yes. Here the source, and title can also apply to you: Stop obsessing over transistor counts: It’s a terrible way of comparing chips

http://www.extremetech.com/computin...ble-way-of-comparing-chips#comment-1611709112:

Yes. And yet at IDF later on, Intel revealed different numbers. I have asked for clarification on this from Bohr himself.
[...]
Finally heard back from Intel on this point. The Core M 1.9B was indeed a typo. I have edited the story to reflect this change.

BTW, your $400 number is a huge exaggeration.

 

[Phynaz]

No way. So Intel thinks they are going to get away with charging $200-$400 for a 1.3 billion transistor part, when Apple is currently paying TSMC and Samsung probably 1/10th as much per transistor. And all the other ARM competitors will be selling 1+ billion transistor chips for even less. No way is that going to continue forever.

If you do the math transistors have been "free" for a long time.

 

[witeken]

If you do the math transistors have been "free" for a long time.

What math?

 

[Kallogan]

Core M are beasts considering the tdp, they obliterate Tegra K1 and all.

But early benchmarks can be a bit fake cause of turbo, question is can it sustain turbo longer than a benchmark duration ?

It's still cr*p though like all convertibles, 2 in 1, and tablets as soon as you need to do real stuff. Just my opinion. I never use those low power devices for noobs. As soon as u want to play real games and do 1080p encoding...I have a tablet and never used it lol

For about the same price i can get a laptop with 850M that can run Star Citizen...and still have a decent battery life for web browsing. So no thanks.

That Asus UX305 is sexy though and fanless is nice. I would definetely buy it for grandma or wife. You know people who can't handle a real computer.

 

[Roland00Address]

No way. So Intel thinks they are going to get away with charging $200-$400 for a 1.3 billion transistor part, when Apple is currently paying TSMC and Samsung probably 1/10th as much per transistor. And all the other ARM competitors will be selling 1+ billion transistor chips for even less. No way is that going to continue forever.

Sigh with that logic the $107 tray price intel haswell 2955u (celeron) that goes in asus chromebox which retails for $160 plus $10 mir is 71% the cost of the device. (I am simplifying the numbers to $150 for simplicity)

1) This is before the intel chipset and motherboard
2) This is before the ram
3) This is before the ssd
4) This is before any other electronic part
5) This is before any cheap chinese labor
6) This is before ASUS cut which does include the costs to design the thing originally and the software (the os is free but they still have to support it
7) This is before any packing and misc cost
8) This is before shipping from China to the US
9) This is before the big box store, amazon, or newegg cut
10) This is before any warranty costs

And you really think all those people are going to settle for 29% of the revenue whenthey still have their own expenses and want to make their own dollar to support themselves?

You do know that just a big box store or amazon usually take 10% to 20% right?

Why would asus go with a 2955u when a tegra 4 or k1 a15, qualcomm 800 or 801, or a samsung exynos 5 a15 is a whole lot closer to a a 2955u, the OS supports it, and those chips are significantly cheaper and they are lot closer in performance (the worse intel vs the best other).

And that same cpu with the same transistors is also the $393 tray price 4510u (i7)

Tray price is nowhere near what the OEM pays, Tray price is a marketing/bargaining tool via intel

 

[Tsavo]

No way. So Intel thinks they are going to get away with charging $200-$400 for a 1.3 billion transistor part, when Apple is currently paying TSMC and Samsung probably 1/10th as much per transistor. And all the other ARM competitors will be selling 1+ billion transistor chips for even less. No way is that going to continue forever.

When someone can compete with Intel on a performance per transistor (ugh) basis...then...ah, your argument might have some merit. Until then...wat?

 

[Idontcare]

No way. So Intel thinks they are going to get away with charging $200-$400 for a 1.3 billion transistor part, when Apple is currently paying TSMC and Samsung probably 1/10th as much per transistor. And all the other ARM competitors will be selling 1+ billion transistor chips for even less. No way is that going to continue forever.

If all you had to do was count transistors and charge the customer a uniform price for them then the memory manufacturers and GPU companies would be trillionaires by now.

Fortunately that is not how the world works, and customers pay for performance, not for transistor count.

Transistor count, and associated metrics such as transistor density, are truly useless metrics for analyses or comparisons between chips or between companies because they tell you nothing about how much time, effort and money went into the design itself.

 

[Arachnotronic]

If all you had to do was count transistors and charge the customer a uniform price for them then the memory manufacturers and GPU companies would be trillionaires by now.

Fortunately that is not how the world works, and customers pay for performance, not for transistor count.

Transistor count, and associated metrics such as transistor density, are truly useless metrics for analyses or comparisons between chips or between companies because they tell you nothing about how much time, effort and money went into the design itself.

This.

Also, from a pure technical standpoint, the Core M also includes a PCH on package which has a nonzero number of transistors

 

[Phynaz]

What math?

Divide the cost of the chip by the number of transistors.

 

[witeken]

Divide the cost of the chip by the number of transistors.

Sure (and I obviously didn't invite you to post your calculations with my question :whiste, but did you also take the price of fabs, research, etc. into account?

 

[dahorns]

Sure (and I obviously didn't invite you to post your calculations with my question :whiste, but did you also take the price of fabs, research, etc. into account?

I think he is talking about from the perspective of the buyer. Not the seller. The things you mentioned would be irrelevant to calculating cost to the buyer.

 

[Enigmoid]

I think he is talking about from the perspective of the buyer. Not the seller. The things you mentioned would be irrelevant to calculating cost to the buyer.

Because how you measure value as a buyer is in terms of transistors/$.


No on cares how big a chip is, just how well it performs. If company A can produce 100 mm^2 chips that perform as well (in all metrics) as company B's 250 mm^2 chips then at the same price they are of equal value.

 

[dahorns]

Because how you measure value as a buyer is in terms of transistors/$.


No on cares how big a chip is, just how well it performs. If company A can produce 100 mm^2 chips that perform as well (in all metrics) as company B's 250 mm^2 chips then at the same price they are of equal value.

Woah there! I wasn't saying buyers measured value in terms of transistors/$. I agree with you, it would be stupid.

I was emphasizing what Phynaz had pointed out: the cost per transistor has approached the point where one could argue that transistors are free (limits and all that). For what Phynaz was calculating, actual cost per transistor to a buyer, R&D costs are irrelevant. It is simply cost divided by # of transistors.

It is a funny mathematical reality. That is all.

 

[witeken]

Woah there! I wasn't saying buyers measured value in terms of transistors/$. I agree with you, it would be stupid.

I was emphasizing what Phynaz had pointed out: the cost per transistor has approached the point where one could argue that transistors are free (limits and all that). For what Phynaz was calculating, actual cost per transistor to a buyer, R&D costs are irrelevant. It is simply cost divided by # of transistors.

It is a funny mathematical reality. That is all.

There's another funny thing you have to consider:

If you buy a CPU for larger price than just the price of the transistor * the amount of transistors * some margin for the company, then those companies could develop new, better technology and transistors which would have a lower price as well = deflation. Deflation would lead to those low prices anyway, but with better technology.

Sure, the price per transistor has dropped so that one transistor is basically free, but the amount of transistors in an IC has gone up accordingly (because the world wants more transistors), effectively cancelling the falling price of one transistor.

 

[Enigmoid]

Woah there! I wasn't saying buyers measured value in terms of transistors/$. I agree with you, it would be stupid.

I was emphasizing what Phynaz had pointed out: the cost per transistor has approached the point where one could argue that transistors are free (limits and all that). For what Phynaz was calculating, actual cost per transistor to a buyer, R&D costs are irrelevant. It is simply cost divided by # of transistors.

It is a funny mathematical reality. That is all.

Okay, I see. Sorry that I misread you. I thought you were one of those 'nvidia is selling a mid end die for $500 people'.

Your argument is true but as witiken said you are buying more and more transistors.