Turbo Boost...Isn't what I thought?

[Panini]

I thought Turbo boost automatically kicks up clock speed when an application uses two or less cores, but this article claims that it does it whenever an application requires more power.

Does this mean that a macbook pro can essentially run with four active cores running at 3.7Ghz? That's really high...

 

[Nozuka]

Its only possible to reach the max speed with a single core.

You'll find all the details here:
http://en.wikipedia.org/wiki/Turbo_Boost

 

[rans0m00]

Should be able to. It gets a little more complicated than just that but nothing too crazy. If you load all 4 to their max it should hit 3.7 on each core. If you only load 2 cores then only those cores raise clock speed.

 

[leman]

AFAIK, it kicks in when the CPU detects particular load increases on the respective cores and the current thermal conditions allow it. So it probably can overclock with 4 cores being active, but surely not the whole way (it must still operate within its TDP, this is the crucial constraint).

P.S. This is an interesting article on the topic: http://news.cnet.com/8301-13512_3-10362882-23.html

 

[lannisters4life]

I thought Turbo boost automatically kicks up clock speed when an application uses two or less cores, but this article claims that it does it whenever an application requires more power.

Does this mean that a macbook pro can essentially run with four active cores running at 3.7Ghz? That's really high...

Nah, it doesn't work like that. Turbo Boost turns cores off when they're not being used, and the leftover power (actual electricity) is diverted into one core instead of spread out over four, for example, and that one core gets its clock speed kicked up. So if your application isn't using multiple cores, but it is still very demanding, the one core that application is using gets boosted at the cost of the rest, who were already at idle.

Edit: http://www.anandtech.com/show/2832/4 Anandtech is quite good for these technical things for me anyway

 

[Panini]

AFAIK, it kicks in when the CPU detects particular load increases on the respective cores and the current thermal conditions allow it. So it probably can overclock with 4 cores being active, but surely not the whole way (it must still operate within its TDP, this is the crucial constraint).

P.S. This is an interesting article on the topic: http://news.cnet.com/8301-13512_3-10362882-23.html

That article was very helpful, thanks.

So if I understand correctly, if you are connected to a power source (charger) and are keeping your computer cool via external methods (external fans, cool room) then with all cores active, you could theoretically reach higher clock speeds than if you weren't in, say, a cool room since the chip would decrease its clock speed to meet ideal temperatures?

----------

Hm. Looks like I got "ninja'd" by lannisters4life.

So here's my question: How many applications actually utlilise all four cores? I remember hearing somewhere that 80% of applications only use a single core while the majority of the remaining 20% use two cores.

Is it only the high-end software like Final Cut, Gaming, and Photoshop that actually use all the cores? Does this mean that on 80% of tasks I will be getting performance superior to my 3.4Ghz gaming desktop?

 

[PS65]

Its only possible to reach the max speed with a single core.

You'll find all the details here:
http://en.wikipedia.org/wiki/Turbo_Boost

Noozuka is spot on, and Evil Spoonman produced a helpful article regarding single threaded applications:

An application that can only take advantage of one virtual core (or thread) is known to be single threaded. There are several reasons for this:

• There is no reason to add support for additional threads. It would just add needless complexity.
• The program is poorly written.
• The problem it is trying to solve is linear and will not benefit from additional threads.

Most of your applications will consume less than one core's worth of performance at any time. Having multiple cores allows you to run multiple single-threaded programs at once without suffering any slowdown. Many of the more basic programs you use everyday will be single threaded, and it should almost never become an issue.

In order to increase the performance, one must change the microarchitecture, or increase the clock speed. Intel has decided to use a technology called Turbo Boost to increase the clock speed of the processor based on the thermal overhead and/or number of cores active. This existed with mobile Westmere (Arrandale), but it was unreliable and would not step up bins as often as one would hope. With Sandy Bridge, Intel made Turbo Boost much more consistent and reliable. Most of the single-threaded performance difference between Ivy Bridge and Sandy Bridge is that Ivy Bridge boosts higher more often.

 

[leman]

That article was very helpful, thanks.

So if I understand correctly, if you are connected to a power source (charger) and are keeping your computer cool via external methods (external fans, cool room) then with all cores active, you could theoretically reach higher clock speeds than if you weren't in, say, a cool room since the chip would decrease its clock speed to meet ideal temperatures?

Theoretically, yes, but for other reasons than you suggest (has to do with throttling) If I understand it correctly, the CPU will adjust its clocks to meet the TDP — and TDP is independent of ambient temperature. Of course, if consuming TDP would produce more heat than the cooling system is able to dissipate, the CPU will reduce its power consumption again.

So here's my question: How many applications actually utlilise all four cores? I remember hearing somewhere that 80% of applications only use a single core while the majority of the remaining 20% use two cores.

Is it only the high-end software like Final Cut, Gaming, and Photoshop that actually use all the cores? Does this mean that on 80% of tasks I will be getting performance superior to my 3.4Ghz gaming desktop?

There are plenty of tasks that use multiple cores, but very few which use them to their full capacity. Professional image and video editing programs should. Gaming, not so much. What CPU is in your gaming desktop? The 2.3 Ghz i7 in the 15" MBP is faster than many but the newest desktops out there.

 

[Panini]

Theoretically, yes, but for other reasons than you suggest (has to do with throttling) If I understand it correctly, the CPU will adjust its clocks to meet the TDP — and TDP is independent of ambient temperature. Of course, if consuming TDP would produce more heat than the cooling system is able to dissipate, the CPU will reduce its power consumption again.




There are plenty of tasks that use multiple cores, but very few which use them to their full capacity. Professional image and video editing programs should. Gaming, not so much. What CPU is in your gaming desktop? The 2.3 Ghz i7 in the 15" MBP is faster than many but the newest desktops out there.

So will cooling the laptop externally provide no performance boost? It seems reasonable enough, but it makes me wonder why they sell the laptop cooling pads.

The way the Turbo boost works seems a little tedious to me, though. You are saying the clock speed will fluctuate to meet the TDP - doesn't this mean inconsistent performance when the processor clocks down? This could also mean inaccurate benchmark readings, correct?

Also, if lots of apps use multiple cores with only one core doing the majority of the work (if I understand correctly) then the other 3 cores would be using less power so the one core could be clocked higher even though all four cores are active, correct? My desktop has a i5 3570k which is pretty new since it's ivy, but the fact a mobile cpu is getting so close is mind boggling.

 

[leman]

So will cooling the laptop externally provide no performance boost? It seems reasonable enough, but it makes me wonder why they sell the laptop cooling pads.

It certainly won't provide any boost, but it will prevent it from overheating and thus throttling. Also, cooler hardware tends to live longer. Most cooling pads are just useless IMHO and are just there to make money.

The way the Turbo boost works seems a little tedious to me, though. You are saying the clock speed will fluctuate to meet the TDP - doesn't this mean inconsistent performance when the processor clocks down? This could also mean inaccurate benchmark readings, correct?

Well, I don't think so. Of course the CPU will go under the TDP if it has nothing to do. But if you have a situation where one core gets lots of processing to do and other three are mostly idling, the CPU will give more power to the active core (i.e. increase the clock multiplier) while downclocking or even shutting down the other three. It can happen very fast though - I believe the CPU can actually adjust the clocks every other instruction or close to it. So in reality the clocks might even jump between 3.3/0/0/0 and 2.3/2.3/2.3/2.3 every few instructions, depending on how the OS utilises the cores. If you are interested in the technical side of the things, you can also look up the HLT and MONITOR/MWAIT assembly instructions, which, I believe, are largely responsible for how modern CPUs handle their power usage.

The benchmark results should be consistent as a benchmark usually produces very similar CPU usage scenarios over consecutive runs.

Also, if lots of apps use multiple cores with only one core doing the majority of the work (if I understand correctly) then the other 3 cores would be using less power so the one core could be clocked higher even though all four cores are active, correct? My desktop has a i5 3570k which is pretty new since it's ivy, but the fact a mobile cpu is getting so close is mind boggling.

Your desktop will still be faster, but not that much Yes, the speed of mobile CPUs had a dramatic increase with Sandy Bridge and even more so with Ivy Bridge.