Would you pick 2.3GHz or 2.6GHz if both had same price for rMBP? (Speed vs Batt,Heat)

[johnnylarue]

Yes this is correct. You need to understand how chips are fabricated and categorised by Intel.

They make a wafer which is a large disc that contains many many processor dies. The die is the part of the processor that houses all the transistors for the chip, all the cores and so forth.

Not all dies that come from these wafers are born equal. Subtle variations during the manufacturing process can result in some dies which simply outperform other dies when benchmarked by Intel in PPW (Performance Per Watt).

What this means is they test each die they make and some will run at higher speeds than others when presented with the same voltage level. This also means faster chips can run at slower speeds with less voltage than others.

Intel then categorises these processors as different model numbers, it is called speed binning. Not a lot of people know this but the Core i5 and i7 processor in the MacBook Air and 13" MacBook Pro which are Dual Core processors are not actually Dual Core they are Quad Core dies which simply don't perform adequately enough to be sold as Quad Core parts.

This could be for a number of reasons such as 2 cores not performing under low voltage tolerances or two cores not working at all or it could be a single core that is physically damaged. Whatever the case they "bin" those dies down and sell them as Dual Core parts.

The same thing happens even when all the cores do work fine thus we have the same processor with the same TDP of 47 Watts in three different speed offerings, 2.0GHz, 2.3GHz and 2.6GHz

You need to look at their power consumption as a gradient from their idle speedstep state of 1GHz to their highest performance state (2.0, 2.3 and 2.6) they all have the same total TDP of 47 Watts, the TDP is calculated by the processor being at maximum load under a normal working scenario, this means application load indicative of user behavior not a program that loads every transistor in the processor like an Intel CPU Burn test.

This means that the 2.6GHz Processor is the most efficient in PPW (Performance Per Watt) because in the same power envelope as the other two processors it is able to deliver 300MHz more performance.

I hope this was helpful.

Amazing! Thanks. I guess that's one of the (several) reasons that processor benchmarks can vary so much from one machine to another.

I guess the concept I'm not totally grasping though, is the notion that the faster CPUs would generate more heat/use more power despite sharing the same TDP rating. Is the simple fact of power traveling through them at a greater speed enough to cause an increase in the amount of heat they generate?

 

[Starfyre]

Yes this is correct. You need to understand how chips are fabricated and categorised by Intel.

They make a wafer which is a large disc that contains many many processor dies. The die is the part of the processor that houses all the transistors for the chip, all the cores and so forth.

Not all dies that come from these wafers are born equal. Subtle variations during the manufacturing process can result in some dies which simply outperform other dies when benchmarked by Intel in PPW (Performance Per Watt).

What this means is they test each die they make and some will run at higher speeds than others when presented with the same voltage level. This also means faster chips can run at slower speeds with less voltage than others.

Intel then categorises these processors as different model numbers, it is called speed binning. Not a lot of people know this but the Core i5 and i7 processor in the MacBook Air and 13" MacBook Pro which are Dual Core processors are not actually Dual Core they are Quad Core dies which simply don't perform adequately enough to be sold as Quad Core parts.

This could be for a number of reasons such as 2 cores not performing under low voltage tolerances or two cores not working at all or it could be a single core that is physically damaged. Whatever the case they "bin" those dies down and sell them as Dual Core parts.

The same thing happens even when all the cores do work fine thus we have the same processor with the same TDP of 47 Watts in three different speed offerings, 2.0GHz, 2.3GHz and 2.6GHz

You need to look at their power consumption as a gradient from their idle speedstep state of 1GHz to their highest performance state (2.0, 2.3 and 2.6) they all have the same total TDP of 47 Watts, the TDP is calculated by the processor being at maximum load under a normal working scenario, this means application load indicative of user behavior not a program that loads every transistor in the processor like an Intel CPU Burn test.

This means that the 2.6GHz Processor is the most efficient in PPW (Performance Per Watt) because in the same power envelope as the other two processors it is able to deliver 300MHz more performance.

I hope this was helpful.

Okay, there is something wrong here... back when the Macbook Air had reviews on thermals and battery life for its i5 to i7 upgrade, there was a comparison of heat and power consumption under low, medium and heavy loads.

In ALL cases, the higher processor consumed more battery life. In the "Low Workload Case" there was barely a different at around 12 mins, but in high workload, the higher specced processor lost nearly a full hour of battery life and was significantly higher in wattage/thermals that made a difference.

Anandtech Source
http://www.anandtech.com/show/7113/2013-macbook-air-core-i5-4250u-vs-core-i7-4650u/4

If we look at reviews of the Macbook Pro Retina (late 2013 haswell) that are out now... unfortunately no 2.6... only 2.3 or 2.4. It is disturbing to see that..

With the 2.0 processor, battery life lasted for over 10 hours...
WIth the 2.3 processor, battery life lasted for about 8 hours 47.
With the 2.6 processor..... there is no review for this but based on the reviews we have available, maybe 8 hours??? With more heat?

Perhaps the battery estimates are based on the average of the model with the least battery life, which in this case would be the 2.6. The 47 TDP across all 3 might also be primarily for the 2.6, whereas other processors may have lower TDP as they are not clocked as high, which means they generate less heat and saves more battery.

And if they do throttle the performance and eat battery while plugged in... that would just stink. I hope a review or something on the line of thermals and battery for the 2.6 processor comes out soon.......

Please prove me wrong, I would not want fans to kick up more often because of the .3 boost.

 

[Quu]

Amazing! Thanks. I guess that's one of the (several) reasons that processor benchmarks can vary so much from one machine to another.

I guess the concept I'm not totally grasping though, is the notion that the faster CPUs would generate more heat/use more power despite sharing the same TDP rating. Is the simple fact of power traveling through them at a greater speed enough to cause an increase in the amount of heat they generate?

They won't generate more heat / use more power if the TDP is the same as-long as Intel uses the same testing when the processor is created to determine the TDP level.

Generally you can only compare TDP for processors within the same architecture because the tests to determine a chips TDP change over time so comparing for example a 47 Watt Ivy Bridge CPU to a 47 Watt Haswell CPU may not be an accurate test.

Another thing to consider is that the TDP is for the entire CPU package which includes the memory controller, PCIe controller, CPU Cores and now the onboard GPU where as before the CPU didn't even have a GPU on board (a few years ago).

Okay, there is something wrong here... back when the Macbook Air had reviews on thermals and battery life for its i5 to i7 upgrade, there was a comparison of heat and power consumption under low, medium and heavy loads.

In ALL cases, the higher processor consumed more battery life. In the "Low Workload Case" there was barely a different at around 12 mins, but in high workload, the higher specced processor lost nearly a full hour of battery life and was significantly higher in wattage/thermals that made a difference.

Anandtech Source
http://www.anandtech.com/show/7113/2013-macbook-air-core-i5-4250u-vs-core-i7-4650u/4

If we look at reviews of the Macbook Pro Retina (late 2013 haswell) that are out now... unfortunately no 2.6... only 2.3 or 2.4. It is disturbing to see that..

With the 2.0 processor, battery life lasted for over 10 hours...
WIth the 2.3 processor, battery life lasted for about 8 hours 47.
With the 2.6 processor..... there is no review for this but based on the reviews we have available, maybe 8 hours??? With more heat?

Perhaps the battery estimates are based on the average of the model with the least battery life, which in this case would be the 2.6. The 47 TDP across all 3 might also be primarily for the 2.6, whereas other processors may have lower TDP as they are not clocked as high, which means they generate less heat and saves more battery.

And if they do throttle the performance and eat battery while plugged in... that would just stink. I hope a review or something on the line of thermals and battery for the 2.6 processor comes out soon.......

Please prove me wrong, I would not want fans to kick up more often because of the .3 boost.

Honestly there is no good explanation for the results. But what I can tell you is that processors are not all the same and Intel do not sell processors with hard set voltages. They do not for example ship 100,000 Core i7 4960HQ processors all at 0.900v they will ship them with a voltage gradient of maybe between 0.7v and 1.0v.

Intel does this for all processors that they sell and such small voltage changes may not seem like they would cause such a drastic change in power consumption but it does. Sometimes simply comparing two identical processors can lead to wildly different power consumption numbers.

I have two 3930K chips. These are identical processors, they are fabricated in the same place and use the same stepping (meaning they were built using identical silicon design without any alterations). One of them is stable at 3.8GHz at 1.1v while the other is only stable at 1.3v - The power consumption difference here at maximum load is almost 20 Watts but the processors are identical. Intels own website says they ship this processor with a voltage gradient between 0.600V and 1.350V (Source: http://ark.intel.com/products/63697).

What I will say is that when Intel says a chip will hit a specific TDP (which remember is the maximum) that is their own set goal and guideline and sometimes chips deviate from this, both high and low end processors, but you know there are a lot of variations that can result in lowered battery life even among identical systems.

 

[Atomic Walrus]

You can't use TDP to determine the power consumption of a chip, that's simply not what it means. The TDP defines the average thermal output limit (the Thermal Design Point). A great example would be looking at single core performance, where TDP is totally irrelevant as neither chip will be anywhere near it.

(( Little more on TDP: Chips with X TDP will all work acceptably in machines designed for X TDP. That doesn't mean some won't consume more power or produce more heat performing the same tasks, nor does it guarantee that some chips won't need to throttle to stay within their rated TDPs. ))

While the difference between the three chips used in the 15" rMBP is indeed their ability to hit target frequencies at sufficiently low voltages, this does not mean they consume the same amount of power. In the best case scenario, the 2.6 might hit 3.8Ghz at the same voltage the 2.3 uses to reach 3.5Ghz. This avoids an exponential increase in power consumption, however chips operating at different clocks at the same voltage will still consume different amounts of power (how else could this even make sense?).

The only way for the faster CPUs to consume the same amount of power as the slower versions would be for them to hit their boost clocks at even lower voltages, which is highly unrealistic.

So what I'm getting at is that while it's true that the higher end chips are more efficient in terms of speed/watt, they still end up producing more heat and using more power because these better chips are used to hit significantly higher performance targets. There probably is a middle point where they could perform slightly better and use the same amount of power (maybe closer to 3-4% faster than the next model down), but Intel knows that many users would rather see the larger performance bump. Benchmarks will support this (and already have in the past).

What the efficiency of a higher level CPU gives you in reality is a nearly 1.0 speed/power consumption increase ratio, as seen in Anandtech's Air i5/i7 comparison; the faster chip used about 15-18% more power while delivering about the same percentage performance boost. The only place this could break down (and they mention this in the article) is at 100% load, where the faster chip would provide a ~18% speed boost but may use even more power as it stays at slightly higher voltages constantly.

There's something interesting to get out of all this: Since the 2.6 chip is indeed more efficient as far as voltages required to hit frequencies, it would almost certainly get better battery life than the actual 2.3 chip if you under-clocked it to the same frequencies (Anyone want to try it?). That's assuming that Intel's boost system is smart enough to realize that and lower the voltages accordingly, which I believe it based on those air results. Imagine what the 2.6 might be able to do in terms of power efficiency clocked to match the 2.0. This is essentially how you build a MBA-like machine: Grab a chip that can hit decent frequencies and then under-clock it and reap the power efficiency benefits.

TL;DR: The 3 levels of rMBP CPU will definitely consume different amounts of power and produce different amounts of heat. Due to binning the % increases in performance vs power consumption should be similar, but you do still get that extra heat. Under-clocking the 2.6 to 2.3 or 2.0 frequencies might provide those levels of performance at lower power consumption (theoretically, if it could be done).

 

[Quu]

[...]

Great post Atomic and I don't disagree with any of what you've said.

 

[johnnylarue]

Walrus + Quu: thanks for taking the time to break this down for us. Extremely informative and easy to read. You guys should start a podcast.

 

[oseres]

Hi folks,
Thanks for this analysis.
I found an interesting battery life benchmark in MacG that gives a longer battery life (20mn both in Safari+ mail and HD*video display) to the 2.6 Ghz than to the 2.0Ghz.
Does it sound realistic ?
Regards

 

[Quu]

Hi folks,
Thanks for this analysis.
I found an interesting battery life benchmark in MacG that gives a longer battery life (20mn both in Safari+ mail and HD*video display) to the 2.6 Ghz than to the 2.0Ghz.
Does it sound realistic ?
Regards

Their analysis does not surprise me and I expected this result as the 2.6GHz CPU is the most efficient out of the three processors offered.

On my own system I easily got battery life of 6 hours in normal web browsing while connected to WiFi which is close to the test results on that page.

 

[oseres]

Their analysis does not surprise me and I expected this result as the 2.6GHz CPU is the most efficient out of the three processors offered.

Yes but it seems to be contradictory with Starfyre tests which give a longer battery life to the lowest frequency processor (he tested 2.3 vs 2.6 and found a 90mn difference in favor of the 2.3.)

(in fact i haven' t made my choice between the 2.3 and 2.6 because a 90mn autonomy is probably more important to me than the extra processor speed, but if the 90mn is reduced to zero on browsing tasks , i' ll take the 2.6 without hesitation

 

[red321red321]

I got the 2.6 G but my computer does indeed run very HOT. UPside is that it's incredibly fast and has never ever lagged under any conditions.

 

[Quu]

I got the 2.6 G but my computer does indeed run very HOT. UPside is that it's incredibly fast and has never ever lagged under any conditions.

This is bizarre to me because the 2.6GHz model I received (Which has now gone back to apple due to a faulty display) ran very very cool. With web browsing on the Iris Pro graphics the system's highest temperature readout according to iStat 4 was 43c and that was the CPU.

I've actually found it to be really really cool in all my usage scenarios except one: Moving 100GB of data over 802.11ac to the internal PCIe Storage. That made the notebook go up to 85c-90c.

Honestly I've been really happy with the 2.6GHz CPU and I will admit I had some reservations about it but those were unfounded. I've had three separate 2.6GHz models now (3rd one just went back to Apple as noted above) and all of them performed the same in temperature, noise and battery life.

 

[red321red321]

This is bizarre to me because the 2.6GHz model I received (Which has now gone back to apple due to a faulty display) ran very very cool. With web browsing on the Iris Pro graphics the system's highest temperature readout according to iStat 4 was 43c and that was the CPU.

I've actually found it to be really really cool in all my usage scenarios except one: Moving 100GB of data over 802.11ac to the internal PCIe Storage. That made the notebook go up to 85c-90c.

Honestly I've been really happy with the 2.6GHz CPU and I will admit I had some reservations about it but those were unfounded. I've had three separate 2.6GHz models now (3rd one just went back to Apple as noted above) and all of them performed the same in temperature, noise and battery life.

I initially had a 2.6 G that ran cool, but had to get it replaced due to wifi problems. The new in box replacement machine that I received ran much hotter than the old machine.

 

[Wuiffi]

With web browsing on the Iris Pro graphics the system's highest temperature readout according to iStat 4 was 43c and that was the CPU.

My girlfriend (2.6/16/512/Iris Pro only) is doing some light photoshop and has a lot of stuff opened (office, safari, itunes and all the other usual stuff). CPU 35°C, GP 36°C

My 2010 model idles (Macrumors, a few tabs, mail and itunes) at 51°C ....

EDIT: I think the istat temperatures were way off. Installed another sensor app, and it states up to 95°C while Cinebench and around 50-55°C while doing photoshop

 

[Badrottie]

I choose most powerful, fastest CPU over heat..... Why? Apple offers AppleCare and warranty they knows CPU can fail overtime.

 

[dextr3k]

Actually, I don't see this mentioned anywhere.

With ALL things being equal, everyone is comparing how hot/power hungry 2.6 would run vs 2.3 ___for the same duration___.

This is not what we see in real life. With both TDP being the same, the 2.6 will complete your tasks faster than the 2.3. Hence its pegged for less duration.

So say you are encoding a movie, it takes 30 minutes to encode on the 2.3, it will only take 25 min to encode on the 2.6. While they are both pegged 100% for the duration of the encode, the 2.6 will be resting for the 5 minutes its not encoding. So while I am not claiming anything without any tests to back me up, as the 2.6 could very well be consuming more power in the 25 min to negate the 5 minute rest, I think it should be taken into consideration as well.

 

[oseres]

So while I am not claiming anything without any tests to back me up, as the 2.6 could very well be consuming more power in the 25 min to negate the 5 minute rest, I think it should be taken into consideration as well.

Totally agreeing.
We can also differentiate the tasks we usually perform in mobility and the ones we usually perform in desktop mode.
As for me :
Mobility usage : surf, playing videos (while traveling), writing and keynoting
Desktop usage (aka MBP plugged): mobility usage +encoding+ CPU demanding tasks

Therefore, i' m much concern for long duration performing mobility tasks and would like to know is the 2.3 is far superior to the 2.6