Intel Broadwell Thread

[AtenRa]

Personally, i see no reason for 4.5W TDP Core-M existence. You have ATOM, use that and make a good 14nm SoC for Mobile, leave Big-Core designs for Desktop/Laptop/Server.

The Intel obsession to put Big-Core designs in to mobile at that low 4.5 TDP segment is going to backfire on them.

Reality check,

You can get a Quad Core ATOM Windows 10" Tablet for less than $200 today, who is going to spend $800+ for almost the same performance just to have Core-M ??

Not only that, 14nm ATOMs are coming in the next few Quarters, unless they completely castrated them, im sure 14nm ATOMs could be much better than Core-M at 4.5W TDP.
The Tablet segment/market doesnt need big-core SKUs, they need fast, cheap little cores and the fastest iGPU you can have.

A 10W TDP Core-M SKU for ultra thin 13-14" Laptops would be the sweat spot. Thats how low Big-Core SKUs should get, not lower.

 

[IntelUser2000]

A 10W TDP Core-M SKU for ultra thin 13-14" Laptops would be the sweat spot. Thats how low Big-Core SKUs should get, not lower.

I think what's really happening is that process guys are very close to limits of what they can pull off what's a low hanging fruit. They can't get benefits on higher voltages anymore. So they advertise gains that are irrelevant for bleeding edge performance CPUs like Core.

Look what happened:
-Multi-core focus
-Graphics focus
-Low power focus
-Heterogenous focus
-IoT

It's like building a really tall skyscraper. Going to 400m has been fine. Maybe it has been DOUBLING every new building. 25m, 50m, 100m, 200m, 400m. What about going beyond that? I bet even adding 25% of 400m is a daunting task. But people still want 50-100% gains. So you add things like solar panels, earthquake proofing, "green" energy designs to keep the customers and marketing guys happy.

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

Problem is again software. Whatever you are saying may be true in the traditional sense but it seems its perfectly suitable for Smartphone/Tablet apps. Look how big PC software houses like Blizzard is struggling to find direction. They are big, bloated, and can't make clear decisions anymore. The app developers are however simple, elegant, and nimble.

That makes it very ripe for previously irrelevant developers to make their mark. Since in the CPU world ARM-based code is taking over and still making 50%+ gains in CPU per generation w/o process changes and Intel struggles to pull out 10% Intel should be in real trouble technologically. It's not that Intel does not have an advantage in CPU but its not worth the costs and disdvantages. Before Intel CPUs needed a fan, delivered less battery life and form factor was bigger and expensive but who cares it was 5-10x the performance. Now its slightly bigger, it runs a bit hotter and less battery life, and its still expensive but only 20-50% faster.

 

[Qwertilot]

It isn't totally obvious if Intel haven't already gone with 'mostly Atom' as the primary strategy.

Yes, at one point it looked like they were just into trying to shove Core down into everything to retain margins etc, but they did take start taking Atom rather more seriously relatively recently. I guess the time scales are such that they'd have started working on CoreM before they started doing that?

No point not releasing then of course. I'd wait to see if Apple use it - and if so what they do with it - before reaching conclusions.

 

[North01]

Some very pessimistic comments here. I was never expecting Intel to pull a rabbit out of a hat with Core M, but I doubt a proper implementation is going to be just 'slightly better' than Bay Trail Atom. If it does, yes, it would be a disappointment, but we have yet to see.

 

[frozentundra123456]

Some very pessimistic comments here. I was never expecting Intel to pull a rabbit out of a hat with Core M, but I doubt a proper implementation is going to be just 'slightly better' than Bay Trail Atom. If it does, yes, it would be a disappointment, but we have yet to see.

It seems to me some are expecting way too much performance from a 4.5 watt chip. What I am disappointed with so far though are the battery life and temperature readings. But I agree, it is too soon to say how it is going to turn out, although I was hoping for more clear-cut improvements. As for the comments, take them with a grain of salt (actually a huge pile of salt).

Overall though, I see atom as a better fit in these formfactors, simply due to price and performance that is approaching "good enough" and hopefully will improve with 14nm.

 

[witeken]

The effect of the processor on battery life is negligible. It has literally nothing to do with it.

 

[pw257008]

The effect of the processor on battery life is negligible. It has literally nothing to do with it.

please explain further

 

[Qwertilot]

Well I guess thinking that the screen, storage stuff etc etc somewhat dominate the power draw in normal use. Wouldn't be a surprise if so.

This is definitely a big question for the 3.5/4.5w versions of these vs the 10/15w Broadwell's that will be due sometime soon. Definitely for notebook style stuff.

 

[pw257008]

If processor makes no difference, why does Intel invest so much in the efficiency of power gating and low power states? Why did Haswell-U offer serious battery life improvements? Not that other components don't matter, but stick the same battery and cooling and chassis and display with a Core-M and a i7-QM part, and which will have longer battery life?

 

[NTMBK]

The effect of the processor on battery life is negligible. It has literally nothing to do with it.

Depends what you do with your tablet. If it is something taxing like gaming, then processor makes a massive difference.

 

[witeken]

please explain further

If processor makes no difference, why does Intel invest so much in the efficiency of power gating and low power states? Why did Haswell-U offer serious battery life improvements? Not that other components don't matter, but stick the same battery and cooling and chassis and display with a Core-M and a i7-QM part, and which will have longer battery life?

I was obviously talking about (~4.5W) SoCs for fanless systems, comparing it with a 45W CPU is a straw man. All technologies introduced before and with Broadwell (and Snapdragon, and Ax,...) have made the CPU's power consumption an irrelevant factor for battery life since the rest of the platform consumes so much more, especially the screen. Unless those's power consumption gets reduced, the impact of even lower CPU power consumption won't be substantial.

So if a Broadwell-Y device has poor battery life, the CPU almost certainly won't be the cause. Unless, of course, your battery life test consists of a heavy workload for the SoC, but that isn't typical usage, is it? (But then energy consumption depends on efficiency, which is where Core M excels, being the most efficient SoC available.)

 

[frozentundra123456]

I think it is exaggerating to say the cpu power usage is not a factor. Point is, as cpu power usage gets lower and lower, further improvements come at a greater cost and give relatively less benefits in overall power usage, unless you improve other parts of the system as well. As shown by witeken's graphs.

BTW, what kind of work is being done on improving efficiency and power/weight ratios of batterys? Is there any room for improvement there?

 

[AtenRa]

The effect of the processor on battery life is negligible. It has literally nothing to do with it.

By having a lower power, less leaking CPU, you can design a lower power board using less parts cutting total Package(CPU + Board) power.
CPU package power does have a high affect in the overall device power usage as you can see from the slide you posted.
Also, missing from that slide is gaming usage. I believe the CPU package power usage will be substantially higher playing games than watching Video.

 

[witeken]

BTW, what kind of work is being done on improving efficiency and power/weight ratios of batterys? Is there any room for improvement there?

There is a lot of research being done, but so far it seems none of the reported significant breakthroughs have been commercialized or maybe aren't even possible to produce in high volumes, or maybe they're simply still in development. There have been some evolutionary improvements in the past year, though. (http://en.wikipedia.org/wiki/Research_in_lithium-ion_batteries)

Sure, SoC power consumption is some factor, but it isn't a major factor that makes battery life worse than e.g. ARM. That's my point.

 

[pw257008]

I was obviously talking about (~4.5W) SoCs for fanless systems, comparing it with a 45W CPU is a straw man. All technologies introduced before and with Broadwell (and Snapdragon, and Ax,...) have made the CPU's power consumption an irrelevant factor for battery life since the rest of the platform consumes so much more, especially the screen. Unless those's power consumption gets reduced, the impact of even lower CPU power consumption won't be substantial.

So if a Broadwell-Y device has poor battery life, the CPU almost certainly won't be the cause. Unless, of course, your battery life test consists of a heavy workload for the SoC, but that isn't typical usage, is it? (But then energy consumption depends on efficiency, which is where Core M excels, being the most efficient SoC available.)

These are all good points. I took issue mostly with the way your statement was phrased (universal, no caveats or hints of caveats), and thus my straw man to bring out what you actually meant. It still remains to be seen, for me at least, whether Core M is really as capable in battery life as competing SoCs (including Atom). The 5y10 review on the previous page shows the difference in power consumption that comes from the SoC and software that utilizes it (power consumption at 1080p in IE vs. VLC vs. Chrome, for example). Whether another SoC thrown into the same device would provide different results is an open question.

 

[Fjodor2001]

You can get a Quad Core ATOM Windows 10" Tablet for less than $200 today, who is going to spend $800+ for almost the same performance just to have Core-M ??

But is the performance of Quad Core Atom really almost the same as Broadwell Y Core M? I found this (Google translated to English):

                              Core M 5Y70    Core i5-4200U    Atom Z3770
Cinebench 11.5 (Multi):       2,77 / 2,48    2,51             1,49
Sunspider Javascript 1.0.2:   114,7 / 142,8  112              327
3DMark – Ice Storm Unlimited: 45997 / 50985  –                16430
3DMark – Cloud Gate:          4576 / –       4565             1267

 

[IntelUser2000]

But is the performance of Quad Core Atom really almost the same as Broadwell Y Core M? I found this (Google translated to English):

                              Core M 5Y70    Core i5-4200U    Atom Z3770
Cinebench 11.5 (Multi):       2,77 / 2,48    2,51             1,49
Sunspider Javascript 1.0.2:   114,7 / 142,8  112              327
3DMark – Ice Storm Unlimited: 45997 / 50985  –                16430
3DMark – Cloud Gate:          4576 / –       4565             1267

So far, the scores shown there aren't translated into real products. You have to cut off about 30% off that score, and it really gets about 1.9-2.2. The slightly lower end 5Y10 gets 1.6 points, which is really no better than Atom. The latest Atom Z3795 gets 1.64 points.

For 3D, Ice Storm Unlimited the Core M devices get 26-28k points while Cloud gate gets 2700-2800. Cherry Trail should match that if they can get the doubling of performance.

Really, the only advantage is in Javascript, which is so bursty that its basically measuring response rates. Would you really pay 3x the price for "feeling" faster, while needing a fan and running hotter with less battery and bulkier too?

The real competition are the ARM guys as well, which pummels the Atom chips. If you ignore the shoddy IE11 sunspider results(because IE11 is good at sunspider and nothing else) but Chrome, which is more comparable to what ARM devices are running at, you are talking about 200ms vs 290ms.... for the 5Y70. For the 5Y10 its 260ms vs 290ms. So 5Y10 is hardly better than the ARM competition.

The effect of the processor on battery life is negligible. It has literally nothing to do with it.

You aren't wrong except Atom devices and ARM devices get much better battery life per WHr. Because of that Intel worked on the *PLATFORM* to lower power use on Haswell. They still aren't low enough to be competitive with ARM devices while performance regressed and competition leaped forward while keeping same power use. Users don't care whether the CPU power management doesn't matter much to battery life or that they focused on the platform - they just want better products.

 

[witeken]

A dual-issue architecture of course can't compete against Core, which has higher boost clock speeds too (most products have the mid-end SKUs of 1.86GHz or so).

 

[Fjodor2001]

So far, the scores shown there aren't translated into real products. You have to cut off about 30% off that score, and it really gets about 1.9-2.2. The slightly lower end 5Y10 gets 1.6 points, which is really no better than Atom. The latest Atom Z3795 gets 1.64 points.

For 3D, Ice Storm Unlimited the Core M devices get 26-28k points while Cloud gate gets 2700-2800. Cherry Trail should match that if they can get the doubling of performance.

Really, the only advantage is in Javascript, which is so bursty that its basically measuring response rates. Would you really pay 3x the price for "feeling" faster, while needing a fan and running hotter with less battery and bulkier too?

The real competition are the ARM guys as well, which pummels the Atom chips. If you ignore the shoddy IE11 sunspider results(because IE11 is good at sunspider and nothing else) but Chrome, which is more comparable to what ARM devices are running at, you are talking about 200ms vs 290ms.... for the 5Y70. For the 5Y10 its 260ms vs 290ms. So 5Y10 is hardly better than the ARM competition.

So why is there a discrepancy between the numbers in the table I referenced and reality? Is it due to Broadwell Y Core M throttling too much under normal circumstances in products we've seen so far?

So under better cooling conditions, it can perform as in the table I listed?

 

[IntelUser2000]

So under better cooling conditions, it can perform as in the table I listed?

I used to think so, but I think that kind of performance is only sustainable under short-term circumstances, like in benchmarks.

Intel's own words indicate there can be a 30% drop in performance after running 8+ instances or so because the initial performance gain is due to extra thermal headroom its using up. It's not like Sandy Bridge's gain because most of the gains on Sandy Bridge were sustainable under LOT of circumstances.

It simply does not live up to the hype, not at all: http://www.pcper.com/files/imagecache/article_max_width/review/2014-08-11/broadwell-12.jpg

You can get one of the bullet points there, maybe...

 

[dahorns]

I used to think so, but I think that kind of performance is only sustainable under short-term circumstances, like in benchmarks.

Intel's own words indicate there can be a 30% drop in performance after running 8+ instances or so because the initial performance gain is due to extra thermal headroom its using up. It's not like Sandy Bridge's gain because most of the gains on Sandy Bridge were sustainable under LOT of circumstances.

It simply does not live up to the hype, not at all: http://www.pcper.com/files/imagecache/article_max_width/review/2014-08-11/broadwell-12.jpg

You can get one of the bullet points there, maybe...

You realize if the issue were that it is a benchmarking chip, it'd probably be, I don't know, performing better on the benchmarks? We've only seen one product with high end Core M, and it performs worse than the product with low-end Core M. Doesn't that strongly suggest its lack of performance is at least partly due to poor implementation?

 

[IntelUser2000]

dahorns:

I get your point. That wasn't really what I was trying to say. Sorry for that.

It performs decent when you are testing behind closed doors, with what seems like a special metal backing, and set at 6W. Even then there's rather large fluctuation(see the two results given by Fjdor2001). Thats a benchmarking chip.

See reference platforms do not always reflect real world products. One older example is here: http://www.anandtech.com/show/4084/intels-sandy-bridge-upheaval-in-the-mobile-landscape

The performance is expected but you won't get that sort of battery life on an actual system you buy. Still Sandy Bridge was an excellent product. That not being my point though.

When you use words like "reinvent" you expect the gains to be real good, and a slide with that word should meet most, if not all they claim: http://www.pcper.com/files/imagecache/article_max_width/review/2014-08-11/broadwell-12.jpg

I don't have to tell you that 3 of the tested systems already can only meet 2 out of the 5 "bullet points" there. Now if you compare against an outdated 2010 system the Core M looks ok. But now you are telling me a device with once-in-few years next gen process specifically tuned for the chip can't do better than the vanilla Haswell Y? And people are saying that its fine?

Charlie D(now at SA) first said about the amazing performance on the Core uArch chips. Now he's spot on about this too: http://semiaccurate.com/2014/07/11/intel-castrates-broadwell-gutting-performance/

 

[NTMBK]

I can't wait to see how 14nm Atom does. Hopefully it will get a good clock speed boost in tablets.

 

[Nothingness]

I can't wait to see how 14nm Atom does. Hopefully it will get a good clock speed boost in tablets.

If I remember correctly, Cherry Trail will run at 2.7GHz max offering about a 10% increase in clock frequency. The micro-arch should be the same with perhaps some tweaks here and there. The focus is on improving the GPU.

 

[Arachnotronic]

If I remember correctly, Cherry Trail will run at 2.7GHz max offering about a 10% increase in clock frequency. The micro-arch should be the same with perhaps some tweaks here and there. The focus is on improving the GPU.

Cherry Trail-T will run at 2.7GHz, and Morganfield SoC (Goldmont/Gen. 9 SoC for phones) will run at 2.7 GHz. No data available for the tablet Goldmont SoC.

Source: http://s1.directupload.net/images/140528/bvfcnsqp.png