Intel Broadwell Thread

[witeken]

Leakage is too high with the current revision of intel's 14nm FF, but they will fix it before 2H 2015. Clearly Intel has put out NDAs to prevent people from reviewing BW in it's current state. I don't think Apple will even accept 14nm BW in it's current state. Lenovo is OK with screwing it's customers, and make no bones about it anybody who bought a Yoga 3 Pro got screwed. 14nm FF in its current state is a worse performing but denser process than 22nm, and categorically worse than TSMC 20nm planar, and samaung has 20nm FF running great!

Where do you base all these things on? Apple is fine using 20nm, and 14nm blows that leaky planar process away.

 

[bronxzv]

Clearly Intel has put out NDAs to prevent people from reviewing BW in it's current state.

mmm "clearly"

https://www.google.ch/search?q=core+m+review

 

[Nothingness]

Where do you base all these things on? Apple is fine using 20nm, and 14nm blows that leaky planar process away.

And where do you get that 14nm blows that leaky planar process? And I mean something not taken from an Intel slide

 

[witeken]

And where do you get that 14nm blows that leaky planar process? And I mean something not taken from an Intel slide

You could ask Idontcare; he'll confirm that leakage becomes worse at such small geometries and that FinFETs play a key role in overcoming all those issues. Because why else would Intel make the investment? If planar was good enough for 22nm, they wouldn't have changed it. But it was necessary and you also see it in the improvement TSMC claims from its 20nm process: a measly 20-25% versus Intel's 50% on top of the 50% from 22nm.

 

[NTMBK]

Broadwell NUCs incoming. http://techreport.com/news/27499/are-these-intel-first-broadwell-nucs

 

[witeken]

Broadwell NUCs incoming. http://techreport.com/news/27499/are-these-intel-first-broadwell-nucs

4 pages ago : http://forums.anandtech.com/showpost.php?p=36976648&postcount=1188

 

[NTMBK]

4 pages ago : http://forums.anandtech.com/showpost.php?p=36976648&postcount=1188

Damn, this thread moves fast

 

[Kallogan]

i PUKE on NUCs

 

[frozentundra123456]

I share your sentiments. However, I recently toured a clinical diagnostic laboratory, and they had hundreds of automated analyzers crowded together, with small form factory PCs still running XP at each station. I could see a place for NUC like devices in a such a situation, although I am surprised that they did not just have dumb terminals at each machine and all of them networked into a central data handling system.

 

[Revolution 11]

Having worked in a large-scale clinical lab, I can tell you that the industrial software that comes with even the latest greatest expensive sequencers is notoriously temperamental and borderline unstable. Many times, you can not upgrade the OS without junking the instrument.

And most of these instruments are fully networked and linked to the local intranet.

 

[Fjodor2001]

Seems Broadwell has been a slow starter. Not the typical Intel release where most CPUs are released at the same time with a big bang. Do you think CES 2015 will be the "real" release of Broadwell, where most models (except Broadwell-K) are released?

 

[North01]

Apparently an 11" version of the Yoga 3 Pro will be coming soon with Core M 5Y10C or 5Y71:

Source

 

[Fjodor2001]

Lenovo Yoga 3 11 mit Intel Core M 5Y10c, 4GB RAM, 128 GB SSD – 599 Euro
Lenovo Yoga 3 11 mit Intel Core M 5Y71, 8 GB RAM, 256 GB SSD – 999 Euro

So €400 more for a minor CPU and GPU clock bump, 4 GB extra RAM, and 128 GB SSD more. They sure know how to overcharge for the stuff they know customers want. D:

Let's at least hope they get the cooling right this time, and I might be a customer after all.

 

[witeken]

Seems Broadwell has been a slow starter. Not the typical Intel release where most CPUs are released at the same time with a big bang. Do you think CES 2015 will be the "real" release of Broadwell, where most models (except Broadwell-K) are released?

CES is way too early.

After Q1:

I did do the math for the back half saying that -- that as well as the Q4 cost of 14-nanometer out of multiple factories, is the thing that keeps us in the low 60s as opposed to higher. So, I think your math probably is pretty close.

After Q2:

When we get to Q4, we’re ramping multiple 14-nanometer factories simultaneously. We think we’ll see a little bit of an increase in cost in Q4. And in addition in the fourth quarter, because our 14-nanometer products Broadwell is a family of products, we’ll still see some pre-PRQ quals that will cause reserves to go up a little bit in Q4 is our best prediction at this time.

IDF/August 11:
14nm has been in volume production since ~mid-14.

14 nm manufacturing fabs are located in Oregon (2014), Arizona (2014) and Ireland (2015)

Production yield and wafer volume are projected to meet the needs of multiple 14 nm product ramps in 1H ‘15

As I was looking at the August 11 presentation were Intel initially unveiled Broadwell, I came across this tibdit:

Stephan Jourdan – Intel Fellow, Platform Engineering Group
Formerly chief architect of Broadwell, Stephan currently directs the definition and architectural development of Intel's SoCs for tablets and phones.

That can only mean good things for Atom, right?

And for those still in doubt about 14nm leakage:

2nd generation Tri-gate transistors with improved low voltage performance and lower leakage

Leakage Power Reduction:
 Design Process co-optimization to reduce minimum operating voltage
 Lowered Tjmax to reduce voltage

So that's a 1 year delay. Reason?

“So the effect was that our slow learning on yield in fact delayed the entire product development cycle.” --William Holt

 

[Fjodor2001]

CES is way too early.

Ok, so are you suggesting 2015Q2 or so? In that case it'll be a 2 year span from the "big" Haswell release to the "big" Broadwell release. That's a 2 year cadence. Now it remains to be seem if Skylake will be released simultaneously with Broadwell too though. Intel release plans sure are a bit confusing and hard to interpret these days.

 

[witeken]

Maybe they'll launch more Core M in Q1, but Broadwell-U in "(very) early spring": March. Broadwell-M maybe a bit later, Broadwell-K Computex, Skylake IDF. 14nm doesn't have to delay 10 though; Intel financials are good.

 

[frozentundra123456]

Having worked in a large-scale clinical lab, I can tell you that the industrial software that comes with even the latest greatest expensive sequencers is notoriously temperamental and borderline unstable. Many times, you can not upgrade the OS without junking the instrument.

And most of these instruments are fully networked and linked to the local intranet.

Yea, the mass spec instruments I work with are still running XP on P4 Dell optiplex computers. They finally have a version of the software that runs on Win 7, but haven't updated yet. Our university wont allow internet access to XP machines anymore, but since the analytical instruments are not connected to the net, they can get by using XP.

 

[Sweepr]

To give some perspective if I put the best SHA numbers from Cortex-A57 into the best Moorefield score, the Moorefield integer score will go from 1026 up to 1223, and the total score will go from 944 up to 1001. Certainly not insignificant, but far from allowing to reach A57 (or even A17 which doesn't even have AES and SHA) scores.

I know you were kidding, but I think that might be nonetheless interesting

That's an OCed 2.0/1.5GHz Exynos 5433, probably using Ultimate custom kernel.
You're looking at ~1300/4300 for stock clocks (1.9/1.3GHz), which is still impressive IMO.

Ps: Expect a ~20% boost for A57 when running in 64-bit mode (Exynos scores are in 32-bit mode).

 

[Dufus]

Seems Broadwell has been a slow starter

Some laptops kicking around with BDW-U but not on the shelves yet such as

i7-5500U 15W HFM 24 Turbo 30 (LFM 5)!
Acer Geekbench 3.2.2 32-bit 2743/5741
ASUS X751LB
Dell Vostro 14-5480
HP Envy 15 3DMark Vantage P5124 from here

i5-5200U 15W HFM 22 Turbo 27 CPU Mark 3866
Lenovo X1
Dell Inspiron 7548

 

[dahorns]

Some laptops kicking around with BDW-U but not on the shelves yet such as

i7-5500U 15W HFM 24 Turbo 30
Acer
ASUS X751LB
Dell Vostro 14-5480

I[5]-5200U 15W HFM 22 Turbo 27 CPU Mark 3866
Lenovo X1
Dell Inspiron 7548

That puts it at a 17% increase to the 4200U. Matching the higher wattage M processors from Ivybridge isn't too shabby.

 

[Nothingness]

From a Geekbench point of view that i5 5200 doesn't look that good: http://browser.primatelabs.com/geekbench3/compare/258568?baseline=1302050

Poor machine again?

 

[dahorns]

From a Geekbench point of view that i5 5200 doesn't look that good: http://browser.primatelabs.com/geekbench3/compare/258568?baseline=1302050

Poor machine again?

Who knows? The 5200u is ahead in most categories. It gets killed in memory performance.

Honestly, Geekbench scores are so inconsistent. The exact same machine in the exact same setup often differs by more than 10%.

For instance, you picked the highest score for the TOSHIBA P55-A with a 4200u. The median score looks to be ~4200. That would put the difference between the 5200u and 4200u at 19%.

Also, I think these scores highlight some of the problems with using Geekbench to compare across architectures. I mean, do we really think Samsung's latest Exynos in a phone is faster than a Haswell Ultrabook? That just doesn't pass the smell test.

 

[Arachnotronic]

I just hope the OEMs put out good systems based on BDW-U.

 

[Nothingness]

Also, I think these scores highlight some of the problems with using Geekbench to compare across architectures. I mean, do we really think Samsung's latest Exynos in a phone is faster than a Haswell Ultrabook? That just doesn't pass the smell test.

The Geekbench score of the 5200U is twice that of the Exynos in Note 4.

 

[witeken]

Apparently Broadwell-Y has a base frequency of 1.4GHz with the 6W TDP.

http://ark.intel.com/products/84672/Intel-Core-M-5Y71-Processor-4M-Cache-up-to-2_90-GHz