Intel Broadwell Thread

[Abwx]

That only applies to the 8 series chipset, not the 9 series, which supports Broadwell.
.

We re talking of Mboards, sure that the chipset is compatible, it s mentioned in the link i posted, but what about the current LGA 1150 MBs.?.

Can one buy a 1150 MB today, slap a i5 4670 in the waiting and then do a straight upgrade using a BDW LGA.?.

Crashtech said 9 series.

He said :

There's been some FUD spread around that Broadwell-K won't actually be compatible with 9-series LGA 1150 boards.

"9 series"

That s a hell of a deflection to not answer the question, going as far as denying Crashtech s exact sayings, so we can conclude that current LGA 1150 are not compatible with the future BDW LGA, that the chipset is compatible doesnt make current plateforms compatible.

 

[NostaSeronx]

My research points to Altera using whatever marketing numbers they want. Regardless, of accuracy or validity of that statement.

For the 70% lower power, Stratix 10 can only have the same performance.
For the 2x high performance, Stratix 10 has to consume 1.3x more power.

These actual performance increases are not from FinFETs but the new architecture.

Altera, pretty much is stating it is just the new architecture. Not FinFETs that are allowing them to hit 2x performance and 70% lower power.

Ground breaking HyperFlex architecture
2x the core performance of prior generation high-performance FPGAs

That actual reason of the performance increase is Hard IP Blocks in the LE units.

Stratix V and other V = Soft FPGA Fixed Point.
Stratix 10 and other 10 = Hard ASIC Floating Point.

That is how they got more performance and it is low-ended to say they are still FPGAs.

====
To get back on-topic;
Is it known that Broadwell is fully out of order from front-end to retire?

 

[witeken]

Okay, so your conspiracy theory is that 14nm is a hoax; it doesn't have any advantage compared to older nodes. That's a crazy suggestion. Then how was Intel able to drop TDP from 11.5 to 4.5? Those numbers (including many more slides) are all made up?

 

[NostaSeronx]

Then how was Intel able to drop TDP from 11.5 to 4.5?

Better architecture and better IVR.

It is liking ask how ARM's power consumption is in milliwatts. It is better design of the architecture.

 

[Roland00Address]

Okay, so your conspiracy theory is that 14nm is a hoax; it doesn't have any advantage compared to older nodes. That's a crazy suggestion. Then how was Intel able to drop TDP from 11.5 to 4.5? Those numbers (including many more slides) are all made up?

Better node allowing better voltage, better ivr, and understand we did not have a 3x decrease in idle power consumption, instead we had this in max power consumption. If you can decrease the voltage it is far easier to decrease max power consumption for voltage has a quadratic effect on power consumption.

That and combine with the fact Haswell in the later parts were closer to 11.5w instead of 15 (thus not a 3 fold drop but roughly 2ish), and you are only going to allow this part to turbo for so long at these higher speeds keeping it bursty and it is very possible for Intel to achieve this.

-----

Now it can all be a hoax, or the turbo can be so short that it is useless for sustained use (great for internet, sucky for real work) but we will find out in less than a month when people actually can purchase these laptops/tablets.

 

[Abwx]

Okay, so your conspiracy theory is that 14nm is a hoax; it doesn't have any advantage compared to older nodes. That's a crazy suggestion. Then how was Intel able to drop TDP from 11.5 to 4.5? Those numbers (including many more slides) are all made up?

I see 37.5% improvement in this slide, are you assuming that perfs improvements at low voltages also hold at higher voltage and regular frequencies..??.

How do you go to 4.5W from 11.5W with 37.5% improvements.??.

 

[TuxDave]

Better architecture and better IVR.

I'm glad you think that microarchitecture and architecture is so amazing that they found a way to cut power by >50% without any help from process ...

... but that would be a big lie.

 

[Roland00Address]

I see 37.5% improvement in this slide, are you assuming that perfs improvements at low voltages also hold at higher voltage and regular frequencies..??.

How do you go to 4.5W from 11.5W with 37.5% improvements.??.

from wikipedia even though it is a very well known electrical formula

http://en.wikipedia.org/wiki/Dynamic_voltage_scaling#Power
The switching power dissipated by a chip using static CMOS gates is C·V*V·f, where C is the capacitance being switched per clock cycle, V is the supply voltage, and f is the switching frequency

C=.65 according to that slide
V=.90 according to that slide (10% lower)
F= we get this from intel ark 1.6 to 2.3 for the intel 4300y, 5y70 is 1.1 to 2.6 ghz

.65*.9*.9*Frequency=.5625

.5625*11.5=6.46875 so with significant figures that will be 6.5watts.

But understand this intel with the 4300y was 1.6 ghz to 2.3 with max turbo.
Now the 5y70 is 1.1 ghz to 2.6 ghz

6.5 watts*1.1/1.6=4.5 watts

So the numbers check out, we see very soon how long can a 5y70 can keep its frequency at 2.6 ghz vs 1.1 ghz. This will also depend on the oem choices on the governor and the temperature option and the cooling solution.





On another note do we know what the cTDP up and cTDP down options are for broadwell core m yet? For example with haswell you were talking a 11.5 watt chip which was cTDP down to whatever number made it passive. Some oems were able to keep that chip at high turbos for a good amount of time while other oems (I am thinking lenovo with their yoga 11s) the chip did not keep its turbo for very long if at all.

 

[witeken]

Better architecture and better IVR.

It is liking ask how ARM's power consumption is in milliwatts. It is better design of the architecture.

Just no. I advice you to educated yourself. Some links:

http://www.intc.com/eventdetail.cfm?eventid=149021
http://www.anandtech.com/show/8367/intels-14nm-technology-in-detail
http://www.anandtech.com/show/8355/intel-broadwell-architecture-preview

 

[witeken]

On another note do we know what the cTDP up and cTDP down options are for broadwell core m yet?

4W, 3W and 6W IIRC.

 

[Roland00Address]

4W, 3W and 6W IIRC.

Thank you for the info



I am disappointed by the 6w number. Yeah 6w may be the highest you want to go with a passive cooling solution and you want a passively cooled solution in a true tablet.

But some oems (not many but some) would gladly do a hybrid solution where you can do a higher tdp when the cpu is docked, or you can have a fan that turns on when the cpu is set to high performance mode. Not all the core m parts will be in tablets, some will be in bigger convertibles.

Personally I prefer fanless tablets, but I am not the only tuna fish in the big sea. It would be great to have a 10w option. Then again intel probably does not want this for they rather you buy their more expensive skus instead of the cheaper ones and cTDP them up.

 

[Abwx]

from wikipedia even though it is a very well known electrical formula

C=.65 according to that slide

Yes, and according to the slide the effective result is 25% lower power.

V=.90 according to that slide (10% lower)

Wrong, this is 10% lower power, i dont see where you you took the 10% lower voltage, there s no reference to voltage directly but only to its results in matter of power drain.

Power reduction at regular frequencies are 25% due to lower capacitances and 10% due to lower leakage, that s 37.5%.

The low voltage figures are only for low voltage operation, that is , close to the minimum frequency of the CPU, those numbers dont apply to higher voltage/frequencies settings.

 

[witeken]

The way I read it is that voltage is 10% lower across the board; Intel's 14nm process allowed them to get the same performance at a lower voltage at any point on the curve (it isn't that if you go above minimum voltage that there is suddenly a 10% spike to the voltage levels of 22nm and then following the same curve as 22nm. That doesn't make sense.

So you get 100% - 25% - 20% - 10% = 45% of the power of Haswell-Y.

In any case, at least we agree that most of the improvement comes from the new process node, although Intel has made a number of other changes (3DL, FIVR2, DCC, Gen8, BDW, etc.) to further improve efficiency as well.

 

[ShintaiDK]

Thank you for the info



I am disappointed by the 6w number. Yeah 6w may be the highest you want to go with a passive cooling solution and you want a passively cooled solution in a true tablet.

But some oems (not many but some) would gladly do a hybrid solution where you can do a higher tdp when the cpu is docked, or you can have a fan that turns on when the cpu is set to high performance mode. Not all the core m parts will be in tablets, some will be in bigger convertibles.

Personally I prefer fanless tablets, but I am not the only tuna fish in the big sea. It would be great to have a 10w option. Then again intel probably does not want this for they rather you buy their more expensive skus instead of the cheaper ones and cTDP them up.

5W seems to be the regular one for ARM performance smartphones and tablets. The key issue is throttle.

Latest example:

 

[Abwx]

The way I read it is that voltage is 10% lower across the board; Intel's 14nm process allowed them to get the same performance at a lower voltage at any point on the curve (it isn't that if you go above minimum voltage that there is suddenly a 10% spike to the voltage levels of 22nm and then following the same curve as 22nm. That doesn't make sense.

The datas that are misinterpreted on this slide :

"Low voltage transistor performance", means the performance of the transistor at low voltages...

"Lower minimum operating voltage", that is the lowest voltage at wich the CPU is still working correctly.

 

[Khato]

C=.65 according to that slide
V=.90 according to that slide (10% lower)
F= we get this from intel ark 1.6 to 2.3 for the intel 4300y, 5y70 is 1.1 to 2.6 ghz

.65*.9*.9*Frequency=.5625

.5625*11.5=6.46875 so with significant figures that will be 6.5watts.

But understand this intel with the 4300y was 1.6 ghz to 2.3 with max turbo.
Now the 5y70 is 1.1 ghz to 2.6 ghz

6.5 watts*1.1/1.6=4.5 watts

So the numbers check out, we see very soon how long can a 5y70 can keep its frequency at 2.6 ghz vs 1.1 ghz. This will also depend on the oem choices on the governor and the temperature option and the cooling solution.

Mostly correct, but you're only accounting for the switching power. Which is why the Intel slide has .65x capacitance 'only' resulting in a 25% reduction in power.

Regardless, taking the Intel specified reductions you get 25% for capacitance, 20% for voltage, and 10% for leakage - 1/(1.25*1.2*1.1) = 0.606x. Which is slightly above the figure you arrived at, but still works well enough for explaining the reduction in TDP.

 

[Hans de Vries]

Okay, so your conspiracy theory is that 14nm is a hoax; it doesn't have any advantage compared to older nodes. That's a crazy suggestion. Then how was Intel able to drop TDP from 11.5 to 4.5? Those numbers (including many more slides) are all made up?

You are making wonderful amazing claims which were never made by Intel itself....
Please read the Intel sheets carefully.

Comparing a 22nm processor and a 14nm processor both at 4.5W !!

For CPU applications the performance improvement is 19%, 12% and 11%

Intel told to journalists that the 22nm Core i5 had been down clocked to
lower frequencies to make it operating at 4.5 Watt instead of it's normal
TDP of 11.5 Watt in order to compare at equal TDP (4.5W)

The sheet says: previous generation Intel Core i5-4302Y @4.5W 1
The sheet says: 1 Configuration information in backup
The backup says: prior generation Intel Core i5-4302Y. 4.5W Thermal design power

Hans

 

[Khato]

You are making wonderful amazing claims which were never made by Intel itself....
Please read the Intel sheets carefully.

Comparing a 22nm processor and a 14nm processor both at 4.5W !!

For CPU applications the performance improvement is 19%, 12% and 11%

Query - are any of those three applications actually limited by TDP? SYSmark 2014, WEBXPRT, and TouchXPRT don't exactly strike me as workloads which are likely to heavily stress the CPU, much less for long periods of time. If they're able to run the bottle-necked portions of those benchmarks at full turbo, well, 2.6/2.3 = a 13% increase in frequency and hence performance.

Luckily we shouldn't have to wait too much longer to have information from reviews rather than PR slides.

 

[jpiniero]

That s a hell of a deflection to not answer the question, going as far as denying Crashtech s exact sayings, so we can conclude that current LGA 1150 are not compatible with the future BDW LGA, that the chipset is compatible doesnt make current plateforms compatible.

I'm still not seeing what the problem is if Broadwell-K's requirement goes from 9 series only to 9 series plus Broadwell-K specific changes. The amount of people with a 9 series board who would want to upgrade to it esp when the CPU performance increase is going to be 0% at best over a 4790K is tiny.

 

[crashtech]

I'm still not seeing what the problem is if Broadwell-K's requirement goes from 9 series only to 9 series plus Broadwell-K specific changes. The amount of people with a 9 series board who would want to upgrade to it esp when the CPU performance increase is going to be 0% at best over a 4790K is tiny.

I see a problem if we have, for instance, a G3258 user that bought an Z97 board, thinking Broadwell-K might be a nice upgrade option, only to find that it will not work. Can we confirm what the original English of the VR-Zone article says, or not?

This presents a number of backwards compatibility issues, and means that only motherboards that support refreshed Haswell will be compatible with the 9-series Broadwell chips.

Read more: http://vr-zone.com/articles/intels-9-series-will-support-broadwell/53203.html#ixzz3G4S2YgdF

 

[IntelUser2000]

Yes, and according to the slide the effective result is 25% lower power.

Abwx,

its comparing to traditional scaling. So compared to regular 14nm design its 38% lower because Core M gets specific optimizations. So you still need to take into account what 14nm itself without the optimizations brings.

If you consider that average per process gen reduction in power is usually 30% then the total you get 0.4375x. If you reverse calculate that from 4.5W you get 10.3, not too far from 11.5W on the Y parts.

If they're able to run the bottle-necked portions of those benchmarks at full turbo, well, 2.6/2.3 = a 13% increase in frequency and hence performance.

2.3GHz on the 4302Y is the top Turbo frequency, the dual core frequency is likely lower. 2.6GHz on the 5Y70 is both top Turbo AND top dual core Turbo frequency.

The other Y chips like 4202Y can only Turbo 100MHz higher than the Base 1.6GHz when 2 cores are active, even though the top Turbo is 1.9GHz.

The backup says: prior generation Intel Core i5-4302Y. 4.5W Thermal design power

If you assume that the Haswell Y was running at 4.5W TDP, then you must assume also that Intel made incredible improvements just with the Y flavor of Haswell. Because the power differences are 3.3x while performance differences with the U are only 30%.

Also, note that even the lower 4202Y can get 1.8 points in a fanless form factor for Cinebench R11.5. The 4302Y can probably reach 2 points meaning Core M can do 2.4, pretty much in line with the 4200U.

What makes more sense to you? That they were able to increase performance/watt 2.5x at the same process and architecture or they did that at 14nm? More likely that Haswell Y had to reach SDP to do so while for Broadwell Y the same performance is TDP. That's probably why it has greater gain in an Office test than the Web one. The former has greater demand. As you go into more demanding applications it can probably widen the gap even more.

On another note do we know what the cTDP up and cTDP down options are for broadwell core m yet? For example with haswell you were talking a 11.5 watt chip which was cTDP down to whatever number made it passive. Some oems were able to keep that chip at high turbos for a good amount of time while other oems (I am thinking lenovo with their yoga 11s) the chip did not keep its turbo for very long if at all.

The Nominal TDP of all available Core M chips are at 4.5W. I believe only the 5Y70 has cTDPup of 6W, while cTDPdown for 5Y70 and 5Y10a is at 3.5W while 5Y10(without the a) is at 4W.

The Haswell Y's weren't using TDP. They were using SDP. So if you able to keep the chip cool enough by running mostly "bursty" applications than it can be verified for SDP. I think though some manufacturers decided to lock the thermal limits to SDP itself, meaning for those devices SDP essentially became TDP.

 

[Enigmoid]

You are making wonderful amazing claims which were never made by Intel itself....
Please read the Intel sheets carefully.

Comparing a 22nm processor and a 14nm processor both at 4.5W !!

For CPU applications the performance improvement is 19%, 12% and 11%

Intel told to journalists that the 22nm Core i5 had been down clocked to
lower frequencies to make it operating at 4.5 Watt instead of it's normal
TDP of 11.5 Watt in order to compare at equal TDP (4.5W)

The sheet says: previous generation Intel Core i5-4302Y @4.5W 1
The sheet says: 1 Configuration information in backup
The backup says: prior generation Intel Core i5-4302Y. 4.5W Thermal design power

Hans

Highly doubt this. Likely what is happening is that the Haswell @ 4.5W is running right at the edge of the power envelope; any more any it would throttle.

The broadwell 14 nm chips likely consuming significantly less than the 4.5W TDP (remember that is with graphics and under heavier applications).

I would be extremely shocked if going from 22 nm to 14 nm SOC only resulted in a <20% efficiency gain, ESPECIALLY when considering the new features in BW.

 

[IntelUser2000]

I would be extremely shocked if going from 22 nm to 14 nm SOC only resulted in a <20% efficiency gain, ESPECIALLY when considering the new features in BW.

And instead it is Haswell Y that brought majority of the gains over Haswell U.

Yes, and according to the slide the effective result is 25% lower power.

To further clear it up, the slide says:

Broadwell Y 14nm Design/Process Optimizations Delivered 2x Lower Power than Traditional Scaling

So whatever traditional scaling would have brought process/design optimizations bring further 2x lower power. So Core M with straight 14nm shrink might have ended up at 9W.

 

[VengenceIsMine]

Core M (Broadwell-Y) now available - Lenovo Yoga 3 Pro (launched today).

Lenovo introduces the Yoga 3 Pro convertible notebook



http://liliputing.com/2014/10/lenovo-introduces-yoga-3-pro-convertible-notebook.html
http://shop.lenovo.com/us/en/laptops...a-3-pro-laptop

Seen a prototype of this one, it's not as nice as the spec sheet would have you believe. Unfortunately the hinge has a little too much flex in it so using the touch screen is mushy and it's not well balanced, too much weight in the touch screen., if you touch the screen the whole laptop tips easily. Could be an anomaly of being a prototype but I don't think so, the 2 pro had the same balance issue as do a lot of Win 8 ultrabooks and I think Core M being fanless and a much smaller PCB just magnifies the issue. About the only one existing haswell ultrabook that really gets the touchscreen thin enough with a good hinge is the dell xps 11 but the touch keyboard on that design is wretched or the fiercely expensive HP 1040.
I have hopes for the ASUS UX305 but I might be stuck waiting for a macbook air refresh to run Win 10 on and just live without touchscreen.

 

[VengenceIsMine]

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?

Very reasonable and what a lot of heavy/pro users want and I'm pretty sure you have a better chance of seeing a unicorn.

Anything that has Ethernet won't be fanless for a start, it'll be a 15W sku. If you can live with a dongle usb 3/Ethernet adapter (and once I've adapted to that, it's not completely horrible now that there are stable & good usb 3 ones) there are some candidates out there for a lot of it (Yoga 3 Pro, HP has a 12-12.5" clamshell ultrabook coming, Lenovo will refresh Carbon & T but Carbon won't have Ethernet and T series will likely go for 15W cpu maybe the ASUS UX-305 but no Ethernet and definitely a glossy screen though), and not enough will have anything but 16x9 screens and matte is unfortunately a real rarity these days. VGA is going away for a lot of this gen, get used a HDMI or display port mini to VGA adapter. Don't expect to see USB C in clamshells for a while, it'll come to tablets and convertibles first.
Your best bet for most of it is a Macbook Air refresh sad to say.