Does Turbo Boost really matter?

All the new Intel CPUs flaunt it. But does it really do much? How does it work? How does it relate to real world operating results? How and when does it get applied? Who sets the parameters for its operation?

I know Adobe products can take advantage of multi-core/multi-threading, but do these turbo boost "afterburners" really add anything?

 

I can't say i tried disabling it for real world usage to see how much it matters. However, it does indeed work as advertised when i run single threaded benchmarks, it rarely turbos to full because of either tdp or windows scheduling the other tasks to another core.

It does work if you send heavy work to one or two threads though, the clock speeds will indeed increase to process the task faster. How much does that translate in real world time, i'm not really certain. I can see having benefits for single and dual threaded workloads, but on 3 threads or more, the difference won't probably be noticeable unless you run a multithreaded intensive task that can take a while.

As for how TB works, i can't really say in details for SNB, but for clarksfield, it will "put the unused cores to sleep" (power gating iirc) and then increase the clock speeds on the active ones up to the max specified turbo or until it hits the CPU's TDP. The increase in clock speed will depend on how many cores are used and how much they are used. I know this is a somewhat shallow explanation, but it is the best i could gather so far from testing with my 920xm.

I can post screenshots of throttlestop reporting the clock speeds while running wprime on 1, 2, 4 and 8 threads if you'd like.

 

In my experience, most people won't notice the 1 or 2 core TurboBoost at all, no matter what apps they're using, unless they're using a pre-Windows XP operating system.

The reason is that the OS itself is very good at spreading the load of any application (single-, double-, triple- or whatever threaded - doesn't matter) over all of the existing cores.

TurboBoost just increases the frequency for one or two cores so that the OS doesn't have to spread the load, saving a couple of milliseconds at most, which is unnoticeable in everyday use (though good for benchmarks if every tenth of a second counts).

However, if you're fully loading all 4 cores, there will be a slight (but noticeable) difference in performance even if only one or two cores are using TB, because when you need all the performance your CPU can provide, every extra MHz counts.

TurboBoost on all cores is obviously useful for the same reason, but you'll only get it if your laptop has a good cooling system...

 

throttlestop and be always turbo boosted?

 

On mobile CPUs with large turbo boost ranges? A large difference.

 

It can increase the speed at which the CPU runs.

Depends how it is set up but traditionally if some cores are not doing much then more power can be delivered to the other cores enabling them to run faster while keeping the package within it's TDP specification.

Usually the faster a cpu runs the quicker it can finish a task.

See above.

Intel. Although some configurable parameters may be available to the end user.

Usually the faster a cpu runs the quicker it can finish a task.

 

it matters where you notice it most: for the quick speed when needed. imagine decoding a movie or something. there might be a short moment, where you'd have a cpu spike resulting in too low of a performance to decode it in realtime. if the cpu is capable of overclocking at that short moment, you won't notice a stutter, or audio dropout, or similar.

so when ever there would be that short moment where your system feels non-smooth, turbo boost can help to smoothen it out by quickly boosting it for a short moment.

i have a 1.6ghz core i5 samsung tablet. it's a very slow cpu, theoretically. but you'll never notice, as it always, on peaks (like scrolling in a browser, starting an app, loading and analyzing tracks while playing others (i'm a dj)) boosts to 2.3ghz or so, as needed.

it, essentially, fixes all the moments where the system can't correctly multithread, as a single thing running gets stuck and cries for more performance.

for me, the impact is very feelable, similar to what an ssd does. it fixes the typical snappiness-bottlenecks.

 

Very true, mobile CPUs have MUCH more aggressive turbo ratio than the desktop CPUs.

My 2630qm sits tightly at 2.6ghz when all 4 cores are stressed, that's 30% faster than the stock 2.0ghz. Whenever I am doing video encoding and 3d rendering, it makes a BIG difference, you won't notice the speed boost during normal use though.

Of course, the turbo boost depends largely on the cooling system, I have seen plenty of people can't maintain their SB CPUs at intel's advertised speed when they are under load simply because of inadequate cooling.

 

Not so sure about that.

My desktop SNB CPU turbo's from 33x to 50x, others have been up to 59x. An older i7-860 was 21x to 26x and I even had a i5-655k that could turbo from 24x to 99x (75 turbo bins) although that was just done to see how many bins could actually be used.

Even the locked SNB desktop chips can support an additional 4 bins with the right setup, for instance i5-2400 31x to 38x.

 

I think he's talking about stock frequencies. E.g. an i5-2300 desktop chip turbos from 2.8 to 3.1 while a mobile i7-2630qm turbos from 2.0 to 2.9ghz.

 

Yes, at stock, Sandy Bridge has more aggressive turbo on laptops, also don't compare K CPUs to non K models, you'd have to compare those to the 2920xm and 2960xm which can definitely be pushed beyond a stock 2600K given adequate cooling.

 

So are you saying the bins on mobile K have been pushed much more than desktop K? I'd say the difference would not be that much myself.

As for comparing non K, I already did that with Nehalem i7-860 and SNB i5-2400

 

Changing a K's ratio is not turbo, that's overclocking.

Also I doubt that 99x multi was stable enough to even boot.

The review on the first page looked at a stock clocked desktop part, I said that a mobile part will show a larger gap.

My 2720QM will go from 2.2ghz to 3ghz, thats a 36% increase.

 

Any multiplier over the HFM is a turbo bin.

Believe what you like, it was 100% stable. It was not intended for normal operation, being run at a lower bclk but used to check how far the multi could be pushed.

Here is a post where it was used to show an error in RealTemps reporting of bclk.



My response showed that high ratio's can exist on desktop. Seems okay for you to be selective but not me, why not compare the mobile i3-2310M then which has no turbo at all.

 

A first gen I7 at 133mhz base clock and 99x multi is 13ghz, please don't insult my intelligence any further by saying this was any way usable.

While technically a "turbo" bin you are playing with semantics here. In reality you are running it above spec and therefore its overclocking an unlocked CPU.

That's like saying an overclocked Q9200 is turboed to 3ghz rather than overclocked to it when you change its maximum multiplier.

 

Of course. Faster clock speed is always better.

 

lol, No need to insult your intelligence when you have done that already yourself I already explained it was run at a lower bclk.

 

I was saying that comparing bin for non K laptop models to K desktop models was kinda useless. Comparing xm laptop CPUs to K desktop models is of course totally relevant. What we're saying is that if you compare the default bin settings for say an i5-2400 and an i7-2670qm, the turbo is more aggressive on the laptop model (more bins at default turbo). It is clear that in the end the desktop chips will win once you start pushing them to the max that you can though.

For example, take the 2630qm, it had 6x default turbo on 4 cores and x9 on 1 core, that is rather aggressive compared to desktop parts, especially the max default turbo on 4 cores. A desktop SNB CPU will of course win due to the higher overall clock speeds. I'm talking about the relative increase or in % if you prefer, that laptop cpus have a higher turbo relative to their non turbo clock speeds.

I hope i was more clear in what i meant, i was only speaking of comparing laptop and desktop parts, not desktop parts and desktop parts.

Also, i'm not saying laptop parts have been pushed more than desktop parts, that would be silly. Yes, you can push a SNB xm CPU past the stock clocks of a 2600K, of course you can overclock the 2600K too so the 2600K will overclock higher than the 2960xm obviously. I only meant that statement as a comparison of laptop and desktop parts with unlocked CPUs to show that both can be pushed quite high.

 

Yes tijo, I agree with the thread, well most of it and what you are saying. I only tried to point out that in some cases desktops can support a high span of turbo but it all seems to have gotten out of hand even to the point of someone insinuating that I'm lying. Very sad really.

 

You edited it after I posted, and why you circled incorrect information in that picture I don't know.

Either way you were overclocking at it was not the standard intel turbo so why you even bring that up is just nonsense.

Turbo for 655k is two bins above default to 3.46ghz (otherwise known as a pathetic boost) you then overclocked it beyond that with an unlocked multiplier.

Also who cares what the timing circuit can do when it has no effect on performance? No advantage or disadvantage to running the base clock higher or lower when you change all the multipliers (except maybe a less stable clock at lower frequencies).

So back to the original point, mobile turbo has a larger impact than desktop turbo. Of course K series CPUs can be overclocked, but that's obvious, though in the first page chart they were just looking at standard clocks.

 

It was already stated that a lower bclk was used before you replied but you went ahead and used the default bclk of 133MHz instead. The editing was done to supplement the reply with a post link and screen shot that was taken ~2 years ago and needed some digging to find it.

That was also explained but seems you did not read that either. To repeat, that particular screen shot was used to show Unclewebb author of RealTemp and Throttlestop that there was an error in bclk calculation with RealTemp.

It was done to get a better understanding of the hardware. For instance similar testing with multipliers on the C2D lead me to the discovery of Dual-IDA. IDA being Intels first implementation of using an opportunistic frequency multiplier that was further developed into turbo boost 1.0 & 2.0.

Who cares about these things, not many it seems. At the time in 3 months only 2 people showed an interest in Dual-IDA whilst everybody else dismissed it or quoted Intel specification that only one core could used with IDA. One of those persons who took an interest was Unclewebb, nice guy.

I apologize if my content about turbo (k or non-k, any multi above the HFM is a turbo) is deemed unsuitable. I will not bother posting again.

 

Yes, it does make a difference. I render video on my computer (1080p quality). Sometimes it takes up to 40 minutes on my i5 CPU to do so, depending on the length of the video.

I have the Turbo Boost meter running - it shows the CPU boosting from its normal 2.66GHz to 2.80-2.93GHz almost the entire time, on both cores. That's about 10% increase in speed so about 10% less time to render.

 

Some of you might argue that that's about 4 mins 30 secs saved in Charles case, but multiply that by a lot of videos or heavier workloads that would run for hours and the benefits are definitely there.

 

It's just I can see the posts already:

"Why can't my CPU turbo to 5Ghz"

We need to keep the terms straight.

 

I do say, I rather enjoy my 76% overclock



Turbo is what Intel calls the overclock, right?
With no Turbo I would just be at a boring 2.5Ghz

I can adjust my turbo duration and power thanks to the XM chip.

 

Turbo boost has restrictions of TDP/time on it, a straight up OC does not.

Intel guarantees turbo and it is part of the specifications.

 

That is probably as far as you can push it though, nice overclock by the way. Also, no Turbo boost is not what Intel calls overclock, they are two very similar things though.

Turbo does increase the clock speeds, but there is alimit to the turbo multipliers while overclocking is associated either with bclk clock speed modifications or unlocked multipliers. Unlocked multipliers is abit like turbo boost, but like i said, the turbo function is present on non unlocked Intel CPUs as well so it is better to differentiate them.

tl;dr

Turbo boost: small multiplier increase set by Intel that also remains within the CPU's TDP.
Multiplier overclocking: increasing the multipliers on an unlocked CPU to push it as high as possible and past the TDP listed by Intel. There is no way that your 2960xm stays within it's TDPof 45W att hose clock speeds. At stock turbo, it will stay at 45W of heat dissipated most of the time.

EDIT: Ninja'ed by Meaker, although i should add that the time restriction is there only on Sandy Bridge CPUs if you start going over the TDP, as long as you're under or at the TDP you will stay at max turbo, the turbo will decrease shortly after you go over the TDP. On first gen core i CPUs (arrandale and clarksfield), the turbo will go down immediately after going over TDP.

 

I'd say turbo is just a fancy name for an overclock. The QM chips' overclock ratios/tdp/durations are limited and locked by Intel.

I could unlock my 2860QM TDP thanks to Intel XTU and an unlocked BIOS letting it turbo forever.

Overclocking the BCLK is just another form of overclock, and that affects other parts of your system like the RAM.

 

Yes, in a sense turbo is a fancy name for an overclock, but i still feel that the line has to be drawn between a full fledged overclock be it by BCLK or unlocked multis and TB since TB is advertised as a feature on non unlocked CPUs as well. Since SNB CPUs can't be overclocked via BCLK, that makes the distinction even more important.

 

It's not an overclock since it's a factory setting If turbo is an overclock then overclock as a word has no meaning anymore.

 

What i mean is that technically, TB is a slight overclock, however i wouldn't call it an overclock for the reasons you mentioned.

 

I see where you are going, and I fully agree.

However, don't forget: anything that operates higher than its base operating frequency would technically be defined as an overclock. Turbo boost is simply a dynamic overclock

Well, my stock ratios are 2.5Ghz-3.5Ghz, with the base clock at 2.5Ghz. I'm simply overclocking Intel's dynamic overclock to 2.5Ghz-4.4ghz thanks to the XM.

This is a play with words more than anything.
I'll stop now.

 

Then what is a "stock frequency" everything becomes an underclock apart from the maximum frequency any one chip can run.

 

That would be the 'Max Non Turbo Boost Ratio' - check Intel XTU.

Or basically what Windows will report your CPU as under My Computer, Properties.

 

I'd define the stock frequency at the frequency that your CPU is guaranteed to run on all 2/4 cores (i5/i7). Your CPU is guaranteed to run at stock frequency unless it voerheats, the turbo frequencies are the max allowed by factory settings and may or may not exceed TDP depending on the situation and the CPU will downclock back to stock if needed. Overclock: anything over turbo frequencies. I'll take my 920xm as an example to make my thoughts more clear.

Stock frequency: x15 -> 2.0GHz
Turbo frequencies:
x17 -> 3/4 cores
x23 -> 2 cores
x24 -> 1 core

I'm currently running a scan with MSE and a few other programs opened. CPU is at x27 which is what i'd call an overclock and a conservative one at that.

 

It's just a massive play with words. You're saying an overclock is higher than w/e Intel says your CPU goes to at max, e.g. BCLK, no arguing with that.

I'm saying that Intel has their own overclock in the form of TB, which is more marketing than anything. Technically once the boost wears off, you will return to the base clock.

 

One more thing concerning turbo boost, there is more going on behind the scene than a mere overclock. On 4 cores, granted, the turbo is mostly a glorified overclock, however on 2 or 1 core, you need working power gating to be able to cut off most of the power going to the other cores to be able to raise the clock speed of the remaining active cores without exceeding the specified TDP. While i agree that Turbo Boost is in part a play with words, there is more to it than merely increasing the multipliers to boost clock speeds.

Also, if we start calling turbo boost overclocking, there will be no end to confusion.

With this i've pretty much said everything i had to say on the subject and we're going off topic, but it was still relevant to turbo boost in a sense at least.

 

You must be a 3D Designer or video producer to see the difference, especially when talking about laptops, Turbo Boost is essential.

I have an i7-2670QM @ 2.2GHz. Compared to desktop CPUs, this is moderately slow. But it can turbo boost up to 3.1GHz when needed which really fills the gap for the absence of power when you need it.

 

turbo is not overclocking as it's about how to balance your resources, not about how many resources you're allowed to use. overclocking means you allow the system to go hotter, use more power, etc.
turbo allows, within the same temperature and power consumtion limit, to allow single threaded (or lowcount-threaded) situations to change the cpu balance. it's still within the same limits as it would be when all cores would run at 100%.

turbo brings much more, thus, than overclock. and it still works if the overclock is done right, too.

turbo = getting the max within the limits.
overclock = going beyond the limits (and there, again, turbo going max within those new limits).

turbo is the best thing ever for mobile systems. a 1.6ghz dualcore tablet with ssd feels as snappy as a non-turbo-enabled quadcore at 3.33ghz. it does do cpu taxing tasks less fast, but for those random bursts of cpu speed, it's perfect.

i would never buy an i3 system because of turbo. it's all about killing bottlenecks. turbo fixes one of the main perceived bottlenecks: the short moments where the system stutters, or is non-responsive. that + ssd = most responsive and fast performing systems ever.

 

I'd say it. Turbo Boost is highly dependent on your cooling system. When under load, a 2630QM in a laptop with a good cooling system can outperform a 2670QM in a relatively weaker one. The reason being good cooling can keep the processor boost at higher bins for longer periods of time. eg 2.60GHz vs 2.20GHz

 

Calling "turbo" overclocking indicates a fundamental misunderstanding how modern CPUs work. Overclocking is exceeding the guaranteed specifications of a processor while turbo stays in specification while providing more performance.
Turbo either allows a processor to stay close to the TDP or, like in Sandy Bridge (2nd generation Core i#) even exceed it temporarily (as long as the processor temperature doesn't exceed specifications) but Intel/AMD have painstakingly simulated and tested the processors in order to make sure they work as specified in turbo mode.

And to answer the original question, yes it makes a huge difference in some tasks and less in some others. But it is "free" (ignoring the efforts of Intel/AMD).

 

In my opinion, Turbo Boost is like an underclocked CPU overclocking itself back to it's normal maximum recommended speed. I think that's how I get it.

What do you think?

 

Sandy Bridge kinda changed the whole turbo rule with the dynamic Turbo boost 2.0.

Before Sandy Bridge, Nehalem, we had CPUs that could turbo up to the maximum turbo boost speed and stay there. With Sandy Bridge the CPUs can only stay at the maximum turbo for a certain amount of time and fall back again to a certain speed that is decided on how good the cooling system in your laptop is. The good thing about Turbo boost 2.0 is that it can turbo above the maximum TDP (over 45W for Quads) and finish whatever they are doing and then fall back again. So the specs you see @ ark.intel is not the maximum turbo that is within the TDP of that certain CPU, it is maximum above TDP and is controlled by a timer.

I remember a review from Anandtech where the stable turbo frequency of the 2820QM in Quad mode was 2.7ish GHz. There it could stay forever. The specs however clearly says 3.1 GHz with Quad cores turboing. Confusing if you ask me, because I`m not still shure if I understand it myself lol

 

That was the first article on Sandy Bridge by AT IIRC. If in doubt, look for the laptop with the most robust cooling system. Heck, the 2630QM may even beat the 2820QM LOL.

 

If a 2820qm only operates at 2.7ghz in quad mode, the cooling system must be really bad. My 2630qm runs at 2.6ghz all the time when all 4 cores are under load. 2820qm should be able to maintain 3.0ghz+ in quad mode with adequate cooling .

 

Yeah it did at anandtech`s test of it. Have no idea why. They tested a 17" Compal notebook.
AnandTech - Intel?s Sandy Bridge i7-2820QM: Upheaval in the Mobile Landscape

 

Well, it is obvious that given the same cooling, the 2820qm will destroy the 2630qm, but given different cooling systems it could go both ways.

 

Yes, exactly. During my last video editing project, I rendered it 15 times. The minutes saved add up.

 

Hey Dufus, sorry for off topic question: is your bios with unlocked advanced menus for Acer TM4630 still available, desperately need one,



thanks cd020465