Intel Broadwell Thread

[witeken]

Presentation and Roadmap currentness are two different things. Theoretically ITC could have used an outdated Intel Roadmap for their presentation. This is always a risk on a non Intel presentation. To me it looks like it is based on an outdated Roadmap (5Q Roadmap goes to Q3 2015, so the Roadmap was made in Q2-Q3 2014)

The roadmap, which is how such a slide is called, starts at Q3 and goes on to Q3'15. Well spotted. Another roadmap from 2014 even starts in 2013, so it must have been made in 2012!

Also, how is an Intel Technology Conference presentation a non Intel presentation?

 

[kimmel]

<--- my uneducated guess.

That's far from an uneducated guess. That seems like the most likely explanation.

 

[escrow4]

Upgrading to a tick (or tick plus) is never worth it. Upgrading to the next tock isn't really worth it either as things may be broken (like TSX in Haswell) unless its a substantial speed boost (highly unlikely). So the shrink of the next major tock would be ideal. Haswell --> Cannonlake. Sandy --> Broadwell by that logic too.

 

[jpiniero]

Weird positioning of Skylake. It seems to be taking the "midrange" slot, with both top-end desktop ultra-mobile being taken by Broadwell. I guess the big emphasis with Skylake is on the GPU?

There hasn't been much when Intel is planning on releasing the Skylake desktop dual cores. I had speculated that Intel is releasing it this way to use the locked quads as a guinea pig for DDR4.

 

[bullzz]

from OCN

http://www.ultrabookreview.com/5648-asus-transformer-book-t300fa-review/

$599 broadwell tablet. looks like a better implementation than yoga 3 but 6 hrs battery life

 

[Nothingness]

from OCN

http://www.ultrabookreview.com/5648-asus-transformer-book-t300fa-review/

$599 broadwell tablet. looks like a better implementation than yoga 3 but 6 hrs battery life

Poor screen resolution (come on 1366x768 on a 12.5" screen?!?), poor battery life, poor performance with Chrome. It seems better than Yoga 3 Pro especially given the price, but still not what I expected from Core M

 

[Arachnotronic]

Poor screen resolution (come on 1366x768 on a 12.5" screen?!?), poor battery life, poor performance with Chrome. It seems better than Yoga 3 Pro especially given the price, but still not what I expected from Core M

A lot of users prefer Chrome, especially on Windows. The fact that ASUS decided to ship the device in this state is not encouraging.

 

[Revolution 11]

Upgrading to a tick (or tick plus) is never worth it. Upgrading to the next tock isn't really worth it either as things may be broken (like TSX in Haswell) unless its a substantial speed boost (highly unlikely). So the shrink of the next major tock would be ideal. Haswell --> Cannonlake. Sandy --> Broadwell by that logic too.

Does a tick even matter for desktops? If it does not, you want the IPC and chipset features. If it does, you want the power savings.

For the former case, I'd argue that you want the second tock after your product. So if you are on Sandy, you want Skylake.

If you are the person who only wants ticks for desktops for some reason, get the second tick after your product. So if you are on Ivy Bridge, get Cannonlake.

 

[III-V]

Does a tick even matter for desktops? If it does not, you want the IPC and chipset features. If it does, you want the power savings.

For the former case, I'd argue that you want the second tock after your product. So if you are on Sandy, you want Skylake.

If you are the person who only wants ticks for desktops for some reason, get the second tick after your product. So if you are on Ivy Bridge, get Cannonlake.

Ticks are fully capable of bringing clock speed bumps.

 

[frozentundra123456]

from OCN

http://www.ultrabookreview.com/5648-asus-transformer-book-t300fa-review/

$599 broadwell tablet. looks like a better implementation than yoga 3 but 6 hrs battery life

I dont really get these detachables in anything over 10 or 11 inches. The tablet part is just too unwieldy. I like the idea of a detachable, but just not in a package this large. With the plethora of 100 to 200 dollar windows tablets, I probably would just go for a separate ultrabook and a cheap windows tablet for use on the go.

In fact, I would guess that a part of the battery life problem is due to the large (relatively) screen.

 

[Revolution 11]

Ticks are fully capable of bringing clock speed bumps.

Clock speed bumps are extremely minor these days. You might gain 5% clocks but more like 3% and higher clocks use more power, which is the exact opposite reason you settled for a tick. More important is IPC increases for specific software (Ivy brought 10% gains), but even IPC increases can use more power, again the opposite reason for picking a tick. It is pretty obvious that for desktops, ticks make no sense in terms of power use.

The reason you should buy a tick cycle is because you need new hardware immediately. But I would never choose a tick for voluntary discretionary upgrade cycles.

The only other reason I see is if the new process fixes severe heat or power problems like with AMD's Bulldozer CPUs. A tick would make sense then. But Intel CPUs are already pretty efficient per clock, low-temperatures, and power-sipping.

 

[witeken]

Clock speed bumps are extremely minor these days. You might gain 5% clocks but more like 3% and higher clocks use more power, which is the exact opposite reason you settled for a tick.

To see a trend, you need at least 2 3 data points. The reality is that you have to wait 2 years for every data point. Moreover, it is a well known fact that FinFET stagnated (altered) the general trend, so you can't include 22nm:

So you have to go back to 32nm and 45nm if you want to predict anything. So it isn't a surprise to see that clock speed bumps are minor.

 

[witeken]

I was just looking at the Broadwell Wiki page which lead me to Ark, where I noticed that those chips do have an SDP, which I presumed was retired since Intel hadn't mentioned it. Broadwell-Y's is 3.5W.

http://ark.intel.com/products/84672

 

[Nothingness]

I was just looking at the Broadwell Wiki page which lead me to Ark, where I noticed that those chips do have an SDP, which I presumed was retired since Intel hadn't mentioned it. Broadwell-Y's is 3.5W.

http://ark.intel.com/products/84672

Only the Q4 models have that "information" and all of the chips have the same one despite vastly different frequency rqnges. I guess it's due to binning or perhaps just that max turbo is used for shorter amounts of time on higher end chips.

 

[witeken]

Off topic, but I like it, so here's an interview with Mark Bohr. I like how he basically confirms that 10nm will introduce a new technology, which would be Ge and III-V semiconductors for the fin channel.

https://www.youtube.com/watch?v=r28Z4mjLUDk

 

[jpiniero]

Off topic, but I like it, so here's an interview with Mark Bohr. I like how he basically confirms that 10nm will introduce a new technology, which would be Ge and III-V semiconductors for the fin channel.

https://www.youtube.com/watch?v=r28Z4mjLUDk

I only looked at the end but didn't really get that impression that he was confirming anything. You have to remember that it was from 3 years ago, they probably had not finalized things yet.

 

[mikk]

Off topic, but I like it, so here's an interview with Mark Bohr. I like how he basically confirms that 10nm will introduce a new technology, which would be Ge and III-V semiconductors for the fin channel.

https://www.youtube.com/watch?v=r28Z4mjLUDk

From 2011 lol.

 

[pw257008]

from OCN

http://www.ultrabookreview.com/5648-asus-transformer-book-t300fa-review/

$599 broadwell tablet. looks like a better implementation than yoga 3 but 6 hrs battery life

That CB11.5 score seems to indicate it's not running at full boost (2.0 ghz) speed for the entire run. They get 1.62, while I got 1.64 locking my SB i5 at 1.8 via Windows power settings.

 

[Abwx]

from OCN

http://www.ultrabookreview.com/5648-asus-transformer-book-t300fa-review/

$599 broadwell tablet. looks like a better implementation than yoga 3 but 6 hrs battery life

Results are about the ones of a Mullins A10 u6700T but noticeably below a Beema 6410 set apart for ST perf, overall that s quite a failure if we are to remember all the hype and extraordinary claims and subsequent hopes about the much glorified 14-16nm process, there s a threshold at wich point increasing IPC is counter productive in respect of perf/watt , it s likely that this point was already reached during the SB/IB transition but it was shadowed by the node shrink that occured on the same row, and the change of process for new core M iterations will surely not change the picture globally.

 

[IntelUser2000]

from OCN

http://www.ultrabookreview.com/5648-asus-transformer-book-t300fa-review/

$599 broadwell tablet. looks like a better implementation than yoga 3 but 6 hrs battery life

Ewww!

1.62 in Cinebench. That's actually worse than the Yoga 3 Pro. It should at least have scored better if thermals were the reason.

The fault is increasingly due to Core M, not the company or the design. Atom performance at significantly higher price and costs more. What's the point?

overall that s quite a failure if we are to remember all the hype and extraordinary claims and subsequent hopes about the much glorified 14-16nm process,

I am not sure if its just a failure to increase IPC efficiently but a failure of the 14nm and the CPU architecture itself.

 

[dahorns]

Ewww!

1.62 in Cinebench. That's actually worse than the Yoga 3 Pro. It should at least have scored better if thermals were the reason.

The fault is increasingly due to Core M, not the company or the design. Atom performance at significantly higher price and costs more. What's the point?

I am not sure if its just a failure to increase IPC efficiently but a failure of the 14nm and the CPU architecture itself.

its the lower end processor. The yoga 3 is so bad because it performs the same with the higher end model.

 

[Arachnotronic]

@ IntelUser2000

Eh, I don't think Core M will be all that impressive in real life. It doesn't integrate many of the SoC functions needed to be a premium tablet chip, and has that darn 32nm PCH sitting there. Not only does Intel lose efficiency by the mere fact that it's built on 32nm, but I'd imagine there is a power overhead that comes from the communication between PCH and CPU/GPU complex.

Intel won't even be remedying this with Skylake, which will also have a separate on-package PCH (Source: http://www.cpu-world.com/news_2014/2...rocessors.html). I hope it's at least 22-nanometer.

Also, it's pretty easy to dismiss a single result as non-representative, but I think as more Core M systems come out, they'll simply reaffirm the Yoga 3 Pro results.

http://forums.anandtech.com/showpost.php?p=36847126&postcount=110

I wouldn't blame the 14nm process -- I'm sure it's fantastic. I'd blame Intel for trying to shoehorn a fundamentally PC-oriented design into tablets.

These guys really need to take the Core CPU architecture and build a fundamentally mobile-oriented SoC around it...or make Atom a lot better.

Honestly at this point, Intel would be better off allowing its Atom team to run free. Core is encumbered by a lot of functionality that sits there and wastes die space such as AVX2. Atom affords Intel's architects to focus on including what makes sense for phone/tablet oriented processors without any of the baggage that comes over from the higher-performance chips. The company really has a golden opportunity to take the best and most efficient parts of core and put them into Atom, but I fear that Intel might be afraid of making Atom "too good."

This is the wrong mentality. Let Atom be "very good" and make Core even better. As Intel keeps trying to push Core into lower power, sacrifices will eventually be made that hurt its competitiveness in the higher end portions of the market like desktops, high wattage laptops, and servers.

 

[IntelUser2000]

Intel17: It sounds like you want them to do the same as when they opted for Pentium M-derivative, rather than Netburst.

Pentium M was awesome for what it was. Atom isn't, Silvermont or not. If Intel was the ONLY CPU maker they would be fine. However Nvidia and Apple already does better, in fact much better without the "fantastic" 22nm process. Really I think it would take "even more awesomer" 14nm Cherry Trail to barely go against Denver and A8X. I've always been a proponent of "product speaks for itself". So despite what people were saying about how Intel's 14nm seemed too good, I thought a good product would show how it really fares. Well, Core M sucks.

I think they would be in real trouble if ARM guys were given free reign on Windows and all apps were compatible. This is a forced monopoly itself. Like I said it makes me suspicious why they stopped Nvidia at all costs from getting an x86 license. Perhaps Intel isn't a real powerhouse CPU maker after all and that's why they try so hard from others getting the license.

its the lower end processor. The yoga 3 is so bad because it performs the same with the higher end model.

Hardly. Core M 5Y10 in the T300FA has a max 2 core Turbo of 2.0GHz. That's good enough for 2.3 points in Cinebench. The Y3P throttles enough that it only gets 1.9 points. That's well within the range of 5Y10's maximum performance.

Also notice how in some of the 3D tests the Y3P actually performs better. Core M doesn't go anywhere near Intel's hype, that's what's really going on.

 

[Arachnotronic]

Intel17: It sounds like you want them to do the same as when they opted for Pentium M-derivative, rather than Netburst.

Pentium M was awesome for what it was. Atom isn't, Silvermont or not. If Intel was the ONLY CPU maker they would be fine. However Nvidia and Apple already does better, in fact much better without the "fantastic" 22nm process. Really I think it would take "even more awesomer" 14nm Cherry Trail to barely go against Denver and A8X.

I think they would be in real trouble if ARM guys were given free reign on Windows and all apps were compatible. This is a forced monopoly itself. Like I said it makes me suspicious why they stopped Nvidia at all costs from getting an x86 license. Perhaps Intel isn't a real powerhouse CPU maker after all and that's why they try so hard from others getting the license.

I'd be careful in assuming that Geekbench results for a Denver-like architecture mean much...look at AnandTech's benchmarks where Denver's performance is all over the map. It seems too fragile as a general purpose consumer processor, but it looks like it'd be great for what it was probably originally designed for...feeding GPUs in HPC workloads

A8X is a fantastic processor, and from personal experience, the real-world performance is very good. However, Cyclone and Enhanced Cyclone have their own problems...see 3DMark Physics test. Notice how triple core Enhanced Cyclone doesn't scale much above dual core Enhanced Cyclone in that test?

I think Core is a more robust general purpose architecture than any of these ARM CPUs, but at the SoC level at the very least, Core M is not an optimal implementation of a "big core" mobile-oriented SoC.

 

[witeken]

I only looked at the end but didn't really get that impression that he was confirming anything. You have to remember that it was from 3 years ago, they probably had not finalized things yet.

He confirmed FF for 14nm and then said she'd have to wait until another time to hear about 10nm. Not that we didn't know that 10nm would get a materials upgrade, but it's nice to hear it from Mark Bohr.