Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Core M (rMB): Maximum Time in Turbo Mode

iRun26.2

iRun26.2

macrumors 68020
Original poster
Aug 15, 2010
2,123
344
We all know that the Core M processor (say the 1.3GHz version) can enter turbo mode (2.9MHz) until it gets too hot and needs to throttle down. But does anyone have any idea how long this period might typically be?

1) 10ms
2) 100ms
3) 1 second
4) 10 seconds
5) 1 minute
6) 5 minutes

Obviously it's not a fixed time but can anyone give a 'ballpark' estimate?

(For example, if I ran a complex MATLAB simulation with my rMB, how long could it run before it would get bogged down based on the Core M processor's thermal restriction... assuming the core was just at 1.3MHz for a long time before the test?)
 
iRun26.2 said:
We all know that the Core M processor (say the 1.3GHz version) can enter turbo mode (2.9MHz) until it gets too hot and needs to throttle down. But does anyone have any idea how long this period might typically be?

1) 10ms
2) 100ms
3) 1 second
4) 10 seconds
5) 1 minute
6) 5 minutes

Obviously it's not a fixed time but can anyone give a 'ballpark' estimate?

(For example, if I ran a complex MATLAB simulation with my rMB, how long could it run before it would get bogged down based on the Core M processor's thermal restriction... assuming the core was just at 1.3MHz for a long time before the test?)
Click to expand...

Depends on core load essentially. Running 100% load on all cores would certainly drop the clocks to around 2.0 GHz (based on 5 W TDP) after 10 ms and then throttle to 1.3 GHz from there when temperatures exceeds 95 degrees Celsius. On single core benchmark, expect speeds from 2.6 to 2.9 GHz until temps get too high.
 
iRun26.2 said:
We all know that the Core M processor (say the 1.3GHz version) can enter turbo mode (2.9MHz) until it gets too hot and needs to throttle down. But does anyone have any idea how long this period might typically be?

1) 10ms
2) 100ms
3) 1 second
4) 10 seconds
5) 1 minute
6) 5 minutes

Obviously it's not a fixed time but can anyone give a 'ballpark' estimate?

(For example, if I ran a complex MATLAB simulation with my rMB, how long could it run before it would get bogged down based on the Core M processor's thermal restriction... assuming the core was just at 1.3MHz for a long time before the test?)
Click to expand...

i doubt more than 10 seconds. it will be throttled by the fanless design but also by the battery.
 
Any answer will be an assumption we will find out soon enough. The embargo will be lifted the day before and we will see then. Remember this machine is not designed to be a heavy weight.
 
If you need processing power, get a computer who's minimum speed is equal to or greater than the speed you need to have consistently available.
 
iRun26.2 said:
We all know that the Core M processor (say the 1.3GHz version) can enter turbo mode (2.9MHz) until it gets too hot and needs to throttle down. But does anyone have any idea how long this period might typically be?

1) 10ms
2) 100ms
3) 1 second
4) 10 seconds
5) 1 minute
6) 5 minutes

Obviously it's not a fixed time but can anyone give a 'ballpark' estimate?

(For example, if I ran a complex MATLAB simulation with my rMB, how long could it run before it would get bogged down based on the Core M processor's thermal restriction... assuming the core was just at 1.3MHz for a long time before the test?)
Click to expand...

Considering that TDP still keeps the maximum power usage at 5w, and the iPad air 2 a8x chip is 4.5w and runs constantly without overheating and is fanless, the retina macbook should be able to run at turbo mode all day long.

Heat is directly related to energy usage. If the energy is capped at 5 watts, you couldn't heat it up if you wanted to. Have you ever used an LED lightbulb? Those 11 watt LED lightbulbs use 11 watts, and produce significantly less heat. What causes regular incandescent lightbulbs to get so hot is that they are using 60 watts and at the same time producing 60 watts of heat. Energy usage = heat produced. Size of chip and other things don't matter.

When over-clockers worry about their chips overheating, it is because in addition to over-clocking the chip, they have to increase the amount of energy to the chip.

When the retina macbook runs at turbo mode, it still doesn't go over the 5w TDP that the chip is designed to run at.
 
macguy360 said:
Considering that TDP still keeps the maximum power usage at 5w, and the iPad air 2 a8x chip is 4.5w and runs constantly without overheating and is fanless, the retina macbook should be able to run at turbo mode all day long.

Heat is directly related to energy usage. If the energy is capped at 5 watts, you couldn't heat it up if you wanted to. Have you ever used an LED lightbulb? Those 11 watt LED lightbulbs use 11 watts, and produce significantly less heat. What causes regular incandescent lightbulbs to get so hot is that they are using 60 watts and at the same time producing 60 watts of heat. Energy usage = heat produced. Size of chip and other things don't matter.

When over-clockers worry about their chips overheating, it is because in addition to over-clocking the chip, they have to increase the amount of energy to the chip.

When the retina macbook runs at turbo mode, it still doesn't go over the 5w TDP that the chip is designed to run at.
Click to expand...

Unlike ARM, Intel certainly allows for higher than 5 watt TDP for a brief time (when tjunction is far from max) then it throttles back to lower than max Turbo to maintain the TDP
 
cbautis2 said:
Unlike ARM, Intel certainly allows for higher than 5 watt TDP for a brief time (when tjunction is far from max) then it throttles back to lower than max Turbo to maintain the TDP
Click to expand...

You might be right. I was reading up on it and apparently TDP doesn't mean the processor will stay at less than 5watts of energy usage. It means the chip will try to stay around that much. So when turbo boost happens, either hitting thermal limits or TDP limits will tell the processor to down clock.

With the fanless design, we may see a lot of quick short bursts of speed and then throttle. Hopefully it won't produce a jittery experience.
 
macguy360 said:
You might be right. I was reading up on it and apparently TDP doesn't mean the processor will stay at less than 5watts of energy usage. It means the chip will try to stay around that much. So when turbo boost happens, either hitting thermal limits or TDP limits will tell the processor to down clock.

With the fanless design, we may see a lot of quick short bursts of speed and then throttle. Hopefully it won't produce a jittery experience.
Click to expand...

the issue is that in any benchmark test, this will NOT show up. however in real life experience, it might be a deal breaker.

first thing I'm going to do is put it in some serious real life test, i got 15 days to try it out and i'm going to take advantage of it
 
iRun26.2 said:
We all know that the Core M processor (say the 1.3GHz version) can enter turbo mode (2.9MHz) until it gets too hot and needs to throttle down. But does anyone have any idea how long this period might typically be?
Click to expand...

kingofwale said:
the issue is that in any benchmark test, this will NOT show up. however in real life experience, it might be a deal breaker.

first thing I'm going to do is put it in some serious real life test, i got 15 days to try it out and i'm going to take advantage of it
Click to expand...

Some benchmarks do run for a while, and presumably would test it. CPU benchmarks and reviews indicate that the CPU part does just fine (for a mid-speed CPU -- not a high-end CPU). Benchmarks, and reviews, are less enthusiastic about the GPU, though. If you are doing office stuff (Powerpoint, etc.) and normal video viewing, should be no problem, but, the reviews indicate the GPU will not satisfy gamers or people who otherwise need GPU horsepower.

The GPU side is certainly a regression though. Core M’s very restrictive TDP of just 4.5 watts means that the GPU is limited a lot quicker than Haswell-U or Broadwell-U GPUs are. It has the same basic architecture as the Broadwell-U GPU, and therefore it should have similar performance if given the headroom for this. Intel still has some work to do on the GPU side to make it more efficient, and they lag some of their competitors there, although less so with HD 5300 than the woeful Atom N2840’s Intel HD Graphics. They have made some headway here, but still have some more room to improve.

As a tablet, the Core M powered Yoga 3 Pro will run circles around other tablets when performing CPU tasks. The GPU is a bit behind, but it is ahead of the iPad Air already, so it is not a slouch. The CPU is miles ahead though, even when compared to the Apple A8X which is consistently the best ARM based tablet CPU.
Click to expand...

http://www.anandtech.com/show/9061/lenovo-yoga-3-pro-review/7
 
How a CPU turbo boosts and throttles itself doesn't seem so simple on a regular personal computer during a numerical simulation. I'm guessing that it's probably because their behavior is optimized for regular use by the average consumer, which is quite different than what you want to do.

I have monitored how my CPU's frequency goes up and down on my maxed out 27" iMac out of curiosity during Monte Carlo simulations with my handcoded software in C. But its behavior is not extremely predictable. Indeed, sometimes it doesn't turbo boost at all or even throttle itself even if there seems to be a bunch of tasks waiting and the temperature is very low. I doubt its behavior would become more predictable if I compiled C programs as mex files by Matlab. It does tend to boost if there are many jobs to be done and the temperature is low enough. But it can just be a blip, a few seconds, or none at all, regardless of the amount of computation. It seems like a very complicated throttling mechanism is going on.

So, if you run simulations with Matlab etc. on a tiny laptop optimized for web browsing, email, and the like and only for those simple tasks, I tend to doubt that your CPU behaves the way you'd expect from what a dumbed-down explanation of turbo boost says.
 
Last edited:
magamo said:
How a CPU turbo boosts and throttles itself doesn't seem so simple on a regular personal computer during a numerical simulation. I'm guessing that it's probably because their behavior is optimized for regular use by the average consumer, which is quite different than what you want to do.

I have monitored how my CPU's frequency goes up and down on my maxed out 27" iMac out of curiosity during Monte Carlo simulations with my handcoded software in C. But its behavior is not extremely predictable. Indeed, sometimes it doesn't turbo boost at all or even throttle itself even if there seems to be a bunch of tasks waiting and the temperature is very low. I doubt its behavior would become more predictable if I compiled C programs as mex files by Matlab. It does tend to boost if there are many jobs to be done and the temperature is low enough. But it can just be a blip, a few seconds, or none at all, regardless of the amount of computation. It seems like a very complicated throttling mechanism is going on.

So, if you run simulations with Matlab etc. on a tiny laptop optimized for web browsing, email, and the like and only for those simple tasks, I tend to doubt that your CPU behaves the way you'd expect from what a dumbed-down explanation of turbo boost says.
Click to expand...

My Matlab simulations only take about 10 seconds on my 2010 11" MBA (Core 2 Duo). I was hoping I might be able to see improvements over this set-up. I was figuring that if typed for minutes at a time (fixing bugs) and then reran my test I might be able to get a vast majority of the intensive calculations done in turbo mode before the temperature got too hot.

I'm still going to buy the computer and test it out. If it doesn't meet my needs I will return it. I've heard so many good things about the Core M processor (being compared to the 2012 version, for instance) that I think it will likely still meet my needs. High portability is still one of my highest priorities and I highly desire a retina resolution screen.
 
It depends largely on how good Apple's cooling system is, and on the CPU/GPU division of workload is for your use case.

MATLAB would be almost-entirely CPU to my understanding (I haven't used it in years, has it been updated to use GPU compute/OpenCL?)

If your workload is nearly 100% CPU, with no GPU load, and Apple's cooling is good, then, based on similar notebooks, you should be able to go at full Turbo for a minute or more.

If your workload is both CPU and GPU (the CPU's max power is for *BOTH* combined, so using heavy GPU leaves less thermal headroom for the CPU,) and the MacBook's cooling is bad, it might be as low as a few seconds.

I have never seen a system whose cooling was so bad that Turbo shunted down in fewer than 5 seconds, though.
 
Hmmm... The review says the Chi throttles both the CPU and GPU clock frequencies when on battery. I hope the rMB doesn't do that, as it seems to be intended to be run on battery most the time.
 
Four by Six said:
Hmmm... The review says the Chi throttles both the CPU and GPU clock frequencies when on battery. I hope the rMB doesn't do that, as it seems to be intended to be run on battery most the time.
Click to expand...

If 5W is the limit I see no reason why batteries should be any different than plugged in. Maybe that's an automatic Windows software setting thing.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back