Intel Broadwell Thread

[rushmore]

Dupe post. Sorry

 

[liahos1]

http://www.engadget.com/2014/12/30/samsung-ativ-book-9-ativ-one-7/?ncid=rss_truncated

another core m device.

 

[witeken]

These (expensive) laptops need Broadwell-U, not Broadwell-Y.

 

[Dufus]

Running furmark 1.15.1.0 at 640x480 Windowed on the HD5500 of the i7-5500U

Seems to be about 10W unaccounted for, cache? Not much drawn by either the IA or GT cores as can be verified by the low temps. 795points vs 820 points on HD4600 at 950MHz. Unfortunate the HD5500 is throttling, might have to limit the clocks a few hundred MHz less to try and get a better idea.

HD5500 Pkg Pwr 18-15W, GT cores 6-4W, IA Cores 4-1W, GT Temp 67C
HD4600 Pkg Pwr 28W, GT cores 16W, IA Cores 8W, GT Temp 87C

Yeah, the thermals for the HD4600 suck. The IA cores take about 4 times as much power to generate similar core temps.

EDIT:
Okay, clock for clock for the HD5500 and HD4600 a 14% improvement with the HD5500U at 33% less power. IA Clock 2.9GHz, Uncore clock 2.7GHz, GT Clock 0.75GHz.

          MHz    Pkg J    Score  IA J     GT J     GT Temp
HD5500    750    14.5W    741    1.4W     3.6W     61C    
HD4600    750    21.5W    652    0.8W    10.6W     78C

 

[witeken]

What do you mean with other 10W? It's a 15W SKU.

 

[Dufus]

What do you mean with other 10W? It's a 15W SKU.

Look at the pic. The IA cores are using 1W and the GT cores are using 4W yet the package power is using 15W. 15W -4W -1W = 10W. What in the package is using the 10W?

 

[witeken]

Look at the pic. The IA cores are using 1W and the GT cores are using 4W yet the package power is using 15W. 15W -4W -1W = 10W. What in the package is using the 10W?

Maybe the IA core number is wrong by a factor of 10?

 

[Dufus]

Could be something weird with Intels power estimations I suppose, don't know. Going by some Linpack tests the IA cores wouldn't be out by a factor of 10 otherwise I'd have over a hundred Watts of power for them unless they get messed up when the GT cores are loaded. You don't think it could be the caching?

Updated that post, I could manage 750MHz to keep it under 15W.

 

[mikk]

Package power seems accurate. Not sure how accurate the other numbers are. You could try other tools like Aida.

 

[Dufus]

GeekBench3 32-bit 2896 - 5942 Maximum package power 14.5W
http://browser.primatelabs.com/geekbench3/1611372

 

[mikk]

Can you do a 3dmark11 (P) run?

 

[Enigmoid]

Running furmark 1.15.1.0 at 640x480 Windowed on the HD5500 of the i7-5500U

Seems to be about 10W unaccounted for, cache? Not much drawn by either the IA or GT cores as can be verified by the low temps. 795points vs 820 points on HD4600 at 950MHz. Unfortunate the HD5500 is throttling, might have to limit the clocks a few hundred MHz less to try and get a better idea.

HD5500 Pkg Pwr 18-15W, GT cores 6-4W, IA Cores 4-1W, GT Temp 67C
HD4600 Pkg Pwr 28W, GT cores 16W, IA Cores 8W, GT Temp 87C

Yeah, the thermals for the HD4600 suck. The IA cores take about 4 times as much power to generate similar core temps.

EDIT:
Okay, clock for clock for the HD5500 and HD4600 a 14% improvement with the HD5500U at 33% less power. IA Clock 2.9GHz, Uncore clock 2.7GHz, GT Clock 0.75GHz.

          MHz    Pkg J    Score  IA J     GT J     GT Temp
HD5500    750    14.5W    741    1.4W     3.6W     61C    
HD4600    750    21.5W    652    0.8W    10.6W     78C

I think its having trouble reading the chip. Is this version of HWinfo validated for broadwell?

The CPU is clocking down. I wouldn't say really say its throttling because you are not applying any meaningful load to it (1.7%). It likely went into idle mode (Is the notebook on high performance?). Even on high performance mode on my 3630qm it will not peg all four cores at 3.2 ghz unless I load them.

Also isn't the igp of the 5500U from 300-950 mhz? The gpu isn't really throttling unless it falls below base.

TDP includes CPU, memory controller, VRMs, igp, and chipset.

 

[Dufus]

Can you do a 3dmark11 (P) run?

The problem with 3DMark and the 5500U with a 15W limit is that it throttles due to power limiting so the result is IMO inconclusive, please bear that in mind.

With discrete card disabled.
http://www.3dmark.com/3dm11/9214844


@Enigmoid, the CPU is configured to run maximum turbo unless it is power limited, clock modulated or thermally throttled irrespective of Windows power plan. No where near thermal throttling but the power limit is a problem. Yes, the integrated graphics max is 950MHz with option of unlocked bins to ~2.8GHz . Not that it would get even close to that. The edit I added to that post was with the integrated graphics maximum performance set to 0.75GHz so throttling was not a problem.

Linpack 11.2.1. 2 threads at 2.6GHz to keep it at 15W. Max package power 15.1W, IA Core power 13.8W temp 76C.

 

[mikk]

The problem with 3DMark and the 5500U with a 15W limit is that it throttles due to power limiting so the result is IMO inconclusive, please bear that in mind.

With discrete card disabled.
http://www.3dmark.com/3dm11/9214844

It's not different for a Haswell ULT. This score is better than Haswell ULT with 40 EUs: http://www.anandtech.com/show/7072/intel-hd-5000-vs-hd-4000-vs-hd-4400

I would say this is roughly 50% faster than a HD4400.

 

[Enigmoid]

@Enigmoid, the CPU is configured to run maximum turbo unless it is power limited, clock modulated or thermally throttled irrespective of Windows power plan. No where near thermal throttling but the power limit is a problem. Yes, the integrated graphics max is 950MHz with option of unlocked bins to ~2.8GHz . Not that it would get even close to that. The edit I added to that post was with the integrated graphics maximum performance set to 0.75GHz so throttling was not a problem.

Linpack 11.2.1. 2 threads at 2.6GHz to keep it at 15W. Max package power 15.1W, IA Core power 13.8W temp 76C.

I guess its changed then for newer CPUs. I generally thought that power saver mode reduces the clock to idle speeds (none or little turbo), balanced kept it at idle or stock speeds and performance kept the CPU idle at max turbo (not the entire chip but the cores that need it). That's been my experience with a 3630qm. Obviously that has changed.

 

[IntelUser2000]

Impressive performance compared to the 11.5W TDP Haswell-Y predecessor, more than 50% faster @ 3DMark Cloud Gate (and 3x faster than Bay Trail). CB 11.5 score roughly matches the 5Y70-based Yoga 3 Pro. Let's see how the new revision (Core M-5Y10c, 5Y31, 5Y51, 5Y71) performs in a proper design like this.

Hold your horses.

http://www.laptopspreview.com/2014/12/hp-elitebook-folio-1020-g1.html
http://arstechnica.com/gadgets/2014...usiness-ultrabook-youll-want-to-bring-home/2/

Elitebook's Geekbench 3 score of 3,814 fell behind such Core i5-controlled contenders as the XPS 13 (5,153), the Thinkpad Yoga (5,057) and the Macbook Air (5,393), and our 4,165 ultraportable normal.

That's no better than 5Y10. That's also verified by reading Helix 2 user reviews in TabletPCReview forums.

http://forum.tabletpcreview.com/threads/thinkpad-helix-2.63682/page-20#post-419224
http://forum.tabletpcreview.com/threads/thinkpad-helix-2.63682/page-20#post-419262

So the results are that 5Y70/5Y71 is faster than 5Y10 only in Single Thread, while 5Y10 might end up even faster in multi-thread. Considering the latter is significantly cheaper, this is a no-brainer IMO.

I find it weird on the Ars Technica review that in multi-core, the 5Y71 is worse than 5Y70, perhaps that's because 5Y7x chips are more handicapped by TDP than 5Y1x chips and is throttling. Perhaps if we put 5Y10 in the Y3P it would perform similar too.

I think the 5Y1x chips are a pretty good deal, especially if it becomes cheap as in the rumored Yoga 3 version where its at 599 pounds for the base, and the markup for devices in britain would mean it should end up at $599. But the 5Y7x is a rip-off. Intel and the manufacturers are lying to us by hyping Core M performance in closed, controlled tests when the system is cool and saying its faster than Haswell U.

(Honestly I would go further about how it implies their "14nm" lead is really vapor but that's another topic)

 

[Dufus]

I guess its changed then for newer CPUs. I generally thought that power saver mode reduces the clock to idle speeds (none or little turbo), balanced kept it at idle or stock speeds and performance kept the CPU idle at max turbo (not the entire chip but the cores that need it). That's been my experience with a 3630qm. Obviously that has changed.

There are so many ways to change the performance of a CPU these days it makes things not so simple. There was a change from 3rd Gen to 4th with the way that the CPU handles C-State wake up and the CPU can be made through HW registers to treat any P-State as a request for full turbo thereby negating Windows power plan in respect to CPU performance states.

For instance.
http://forums.anandtech.com/showpost.php?p=37033609&postcount=201

 

[mikk]

I have seen a slide from a Broadwell 2+3 ULT SKU from Intel. Official specs: 1.9 Billion transistors and 133 mm² die size.

Based on this I measured 82 mm² die size for the GPU.

 

[witeken]

I have seen a slide from a Broadwell 2+3 ULT SKU from Intel. Official specs: 1.9 Billion transistors and 133 mm² die size.

Based on this I measured 82 mm² die size for the GPU.

Are it 1.9 billion real transistors ?

Broadwell-U is 10% less dense.

 

[mikk]

Are it 1.9 billion real transistors ?

Broadwell-U is 10% less dense.

Transistor Count 1.9 Billion is written on this slide for 2+3 BDW, they compared it to 1.3B for Haswell U @181 mm².

Regarding your link maybe Intel made an error at IDF, they possibly mixed up the number for Core M and 2+3 ULT. Core M at launch was specified at 1.3 Billion. http://www.extremetech.com/wp-content/uploads/2014/09/intel-core-m-broadwell-y-die-diagram-map.jpg

 

[Arachnotronic]

Are it 1.9 billion real transistors ?

Broadwell-U is 10% less dense.

The IDF slide mistakenly paired the BDW-U 2 + 3 transistor count with the BDW-Y/BDW-U 2+2 die size.

 

[witeken]

Transistor Count 1.9 Billion is written on this slide for 2+3 BDW, they compared it to 1.3B for Haswell U @181 mm².

Interestingly, scaling is 2x instead of BDW-Y's 2.2X. Any guess on the reason for that?

Regarding your link maybe Intel made an error at IDF, they possibly mixed up the number for Core M and 2+3 ULT. Core M at launch was specified at 1.3 Billion. http://www.extremetech.com/wp-content/uploads/2014/09/intel-core-m-broadwell-y-die-diagram-map.jpg

I know that BDW-Y has 1.3B transistors; it was sarcastic, since you didn't give a source.

The IDF slide mistakenly paired the BDW-U 2 + 3 transistor count with the BDW-Y/BDW-U 2+2 die size.

Makes sense.

 

[seitur]

What about normal voltage non-Y / non-U laptop Broadwell series?

Are they coming? If so when? Or they like desktops are omitted and there is Haswell->Skylake transition planned?

 

[witeken]

Spring, probably.

 

[mikk]

Normal voltage models weren't listed in all Roadmaps if you refer to 35/37W Dualcore models, I doubt they are coming. Even for Skylake I haven't seen such SKUs.