Could Intel Turbo Boost add a new dimension to overclocking?

[b82]

Supposing I'm looking at getting a Core i5-650 early in 2010. I've read about the Turbo Boost feature, but I'm not overly impressed by the 2 x 133mHz clock step changes that it seems capable of making (the i5-650 is rated at 3200mHz, rising to 3467mHz with maximum turbo).

A 266mHz increase on 3200mHz is a 1/12 increase in clock frequency and I might decide that's a gimick that's not really worth paying 25-40% extra for over say a Core i3 without Turbo Boost. Yes I know Core i3 is slower, but I expect they could be clocked up to 3200mHz or slightly more with very little trouble.

Will it be possible to take an i5-650, underclock it by 3 steps to 2800mHz to save power and tweak the Turbo Boost for a maximum of 3600mHz giving it a full 800mHz (6 step) change in performance that might be useful and noticeable?

Maybe you'd have to 'hack' the Turbo Boost feature and maybe that wouldn't be a good idea, but I still have one other question. What is the difference between buying a Core i3-540 operating at 3067mHz and buying a Core i3-530 (2933mHz) for $20 less and overclocking it by 133mHz? Is that all Intel would do to produce the i3-540, or would they make some other changes to the chip?

 

[b82]

Although this was about CPUs, when I joined up just now I seem to have inexplicably missed the overclocking section of the forum where this would probably be most appropriately posted. If any admin wants to move it there I'm certainly not going to complain.

 

[cyberkuberiah]

no you cannot change the number of steps the cpu makes , it is like a locked multiplier situation . but you can disable it , and overclock the CPU . the best way would be to buy cheapest i3 and oc it as much as possible ! turbo is not needed , and besides that its actually a hindrance at high overclocks .

even more "value" would be to take the rumored pentium G6950 for 87 dollars at 21x133=2.8 ghz, and overclock it to death !

an oc'ed g6950 to 3.46(21x166) ghz would shine against core 2 , and many apps/games that are single threaded .

take this with a grain of salt , but there was also a rumor that someone on hardwarecanucks managed a 4.7 ghz overclock on i5 650 (default 24x133) , on air .

it would be fun to see the shootout between amd's "value" dual and tri core processors without going head to head against possibly cheap i3 processors .

 

[cyberkuberiah]

but as always , dont count on anything like speeds and pricing until the whole lineup is actually launched ! but one thing is for sure : people who were getting core 2 before would get this setup when it comes , surely .

 

[b82]

no you cannot change the number of steps the cpu makes , it is like a locked multiplier situation . but you can disable it , and overclock the CPU . the best way would be to buy cheapest i3 and oc it as much as possible ! turbo is not needed , and besides that its actually a hindrance at high overclocks .

Thanks for the info. I think this is a shame though. A turbo with more 'range' could offer energy savings when you don't need the power, and lots of speed when you do need it. As currently proposed the 'two-step' turbo doesn't seem worth the money, but does seem like a wasted opportunity. To me, dynamically managed o/c ing seems like a fair idea in principle.

even more "value" would be to take the rumored pentium G6950 for 87 dollars at 21x133=2.8 ghz, and overclock it to death !

an oc'ed g6950 to 3.46(21x166) ghz would shine against core 2 , and many apps/games that are single threaded .

take this with a grain of salt , but there was also a rumor that someone on hardwarecanucks managed a 4.7 ghz overclock on i5 650 (default 24x133) , on air .

Yes, I read about the i5 overclock. I also read that a german site AwardFabrik got their hands on an i3-540 and it showed similar potential. Google will tell you all about it.

I do currently see myself getting an i3. Tell me, what's the difference between a lightly o/c ed 530 and a stock 540? Nothing?

I have to admit that I'm not a really a gamer, and for me a massive overclock (while slightly fascinating) would probably not be worth it.

 

[Mousemonkey]

This topic has been moved from the section CPU & Components to section Overclocking by Mousemonkey

 

[cyberkuberiah]

This topic has been moved from the section CPU & Components to section Overclocking by Mousemonkey

thanks

 

[cyberkuberiah]

the difference at stock between any two chips is guranteed minimum stable speed , this is called "product binning" (ref. wikipedia) .

of you get a successfull oc of a 530 to 540 speeds than there is NO difference .

it is actually good that you dont require a massive overclock , because higher binned chips are likely to result in higher MAX overclocks , but such kind of max overclocks would require exotic cooling methods anyway like liquid N2 etc .

that's why we take a good old (yeah , sometime has passed) i7 920 and oc it with decent air cooler , and the resulting overclock is enough for many people . why bother with 975 , its so costly !

 

[cyberkuberiah]

a utopian turbo feature would be this : the user just specifies a power draw limit and places a cooling solution on the cpu , and the cpu should know its own voltage limits , and dynamic temperature from sensors , and overclocks itself to the max speed bins available . processors already can change the voltage dynamically . the problem would be for marketing folks , the tech people in amd or intel would love this ! we wont require the hours of prime95 testing , multiple reboots with this sort of "turbo unleashed" ! . just change the cooler and get different results automatically ! most cpus are good on the other end of the spectrum , when it comes to saving power when idle .

 

[Ancient_1]

Thanks for the info. I think this is a shame though. A turbo with more 'range' could offer energy savings when you don't need the power, and lots of speed when you do need it. As currently proposed the 'two-step' turbo doesn't seem worth the money, but does seem like a wasted opportunity. To me, dynamically managed o/c ing seems like a fair idea in principle.


Yes, I read about the i5 overclock. I also read that a german site AwardFabrik got their hands on an i3-540 and it showed similar potential. Google will tell you all about it.

I do currently see myself getting an i3. Tell me, what's the difference between a lightly o/c ed 530 and a stock 540? Nothing?

I have to admit that I'm not a really a gamer, and for me a massive overclock (while slightly fascinating) would probably not be worth it.

The one thing a higher turbo would do (considering it works the same as a i920) is give a higher multiplier when you overclock, which if done like I do would allow a lower idle speed when all power saving features are enabled to run the same OC. I run my 920 with a bclk of 191 which gives me a load speed of 4000 and an idle speed of 2290 and load vcore of 1.36 and idle of 1.136 with HT enabled and 1.296 load and idle of 1.12 with HT disabled.

It would be nice to have one of the extreme chips just to get the high multi so that you could use 30 as a multi to get 4000 cpu speed at load and only 1600 at idle or have a lower multi as a minimum on all the i series chips which could accomplish the same thing.

I don't see why so many people say to disable all power saving features to OC, which really only adds power consumption and heat with no real advantages. I use my PC mostly as a media center recording, editing and trans-coding HD TV shows to xvid for archiving which puts my PC at load for about 8hrs a day and basically idle rest of the day. I see no reason for my CPU to run at 4ghz and high vcore while I surf or watch video.

I feel the standard power saving features do just what you are getting at even tho either a higher max multi or lower idle multi would make it more efficient.

Here are a couple of screen shots I took of my i7 at idle and load with HT enabled (I ran my e8400 the same way but the difference between idle and load vcore was much smaller).
I wish the D0 stepping were out when I got mine which would of allowed a reduction in vcore but I don't think the payback would justify the expense of a new cpu.

Load

Idle

 

[b82]

a utopian turbo feature would be this : the user just specifies a power draw limit and places a cooling solution on the cpu , and the cpu should know its own voltage limits , and dynamic temperature from sensors , and overclocks itself to the max speed bins available . processors already can change the voltage dynamically . the problem would be for marketing folks , the tech people in amd or intel would love this ! we wont require the hours of prime95 testing , multiple reboots with this sort of "turbo unleashed" ! . just change the cooler and get different results automatically ! most cpus are good on the other end of the spectrum , when it comes to saving power when idle .

Yes, this is great! Full Auto Turbo. I like this idea. I wonder what kind of technal knowledge you'd need to develop such a system (way more than mine for sure!) Mind you, it would kind of take the 'fun' out of overclocking for some people.

Interested in what you say about 'most cpus are good on the other end of the spectrum, when it comes to saving power'. If the turbo feature isn't a power saving feature, then I don't know what it is. Consider that the top rated Core i5 (Westmere) BOTTOM turbo setting is the same as the base Core i5's TOP turbo setting. So the turbo as configured CAN'T be about discovering the limits of the cooling available (as you've specified in Utopian Turbo), because the turbo range is so small that it wouldn't begin to probe those limits (as I see it).

My (probably poorly informed) opinion is that Turbo (as currently specified) is an overpriced gimick, and so must exist primarily for marketing purposes. If anyone can tell me why I'm wrong, please do.

 

[cyberkuberiah]

the turbo feature is getting better with each line : core i7 had one bin for one core , then i5 for all cores etc .

now , arrandale , the mobile westmere , can reportedly go 2.26 from 1.20 base clock (double the speed bins minus one) and thats great , all within its tdp .

oh wait , i think i've caught on with intel on this one ! consider this : when they came out with nehalem , probably the heat produced was on the higher side with a constant high speed , what they did was instead of releasing these egg cookers , they released lower speed/heat processors that could go up to higher speeds for some time and come back as soon as danger temp zone was encountered . this really is an innovative solution .

for the utopian turbo , you need to (i've refined the above version):-

1. change voltage dynamically , which is already done (the phenom under-volts and under-multiples itself ). but add to this the processor knowing its max safe voltage limit like the cpu-z information is also stored on it . both reduce and increase voltage based on limits , keeping in mind the power draw which depends on voltage as well as frequency . the processor should also know its own viable frequencies for a voltage .

2.read the temperature dynamically , which is already done for individual cores . add to this thermal load balancing on chip itself , so that one hot core does not bring the other 3 down in speed , or migrate the process (i wonder if this is being done already , and i dont know how much one core effects thermals of the neighbouring one today ) .

3. idle power and shutting of cores completely : voltage , and more so the multiplers should have deep low limit , like radeon 5000 low idle clocks vs 4000 high idle clocks .

4.voltage regulation would depend on the motherboard , so get this info from the motherboard (bios , or something else) like voltage steps and upper/lower limits as well as upper power draw limit .

5.the key to saving power is not overusing it . together with low idle , the application should convey to the processor via the os or directly , the maximum needed amount or unlimited power . example is that in a gpu bottlenecked game , you dont get any increase from better processor performance , so keep it at only the required amount plus some headroom . but this should be changeable rapidly , otherwise situations like physics would suddenly impact the performace , and in the end it may or may not be viable . the key here is how rapidly the processor shoot up performance , and it would require radical engineering steps to increase that transition speed to a fraction of clock speed itself . i read somewhere that phenom can do it 10 times per second , but that may not be enough .

6.the cooling is simple , all the user needs to do in the whole process is decide on the cooling . but the processor should report at least a rough idea on how much heat is being generated and cooled w.r.t time , like a heat-time graph in overdrive software of amd . this would benchmark the coolers , as there are a variety of designs , and it should not take a thermodynamics expert to figure out the heat sink capacity of a cooler , be it water or air , or have 2 or 4 copper tubes etc .

the software eg. overdrive should keep the fan speed minimum so as to make minimum noise in accordance with the dynamic data from the processor , and as mentioned above should benchmark the fan , keep the thermodynamic graphs on the system and use them as reference for automatic fan control .

7. different processors have different overclocking capacities , from the start when they are manufactured . they do bin the chips based on testing , but it would be impossible to make this data available in the chip itself after the chip has left manufacturing . but i doubt whether they test for maximum overclock . on multi core processors it may be possibly easy for one core to monitor the stability of another in stress testing . in this way all the cores can gather this data one by one about the other cores .


8.this type of heat/power management system will be proofed against fan failures : the chip will alert the user as the cooler is showing an abnormal thermodynamical response keeping in mind the ambient temperature in the room , of course .

9.whoa , my head feels lighter now . FTW and waiting for your feedback !

 

[cyberkuberiah]

my whole idea is to make the process as simple on the users behalf , and it may be a personal choice , but i would rather spend my time saved shooting away alientech of crysis or exploring african jungle in far cry 2 !

the above process is the best of power saving and shooting for top notch performance in one go !

 

[b82]

7. different processors have different overclocking capacities , from the start when they are manufactured . they do bin the chips based on testing , but it would be impossible to make this data available in the chip itself after the chip has left manufacturing . but i doubt whether they test for maximum overclock . on multi core processors it may be possibly easy for one core to monitor the stability of another in stress testing . in this way all the cores can gather this data one by one about the other cores.

Does this mean that if you plan to O/C a Core i3 you'd be better to buy a 540 on the basis that the 530s are more likely to come from the 'rather flaky' category? Or do the 'rather flakies' all end up with the Pentium brand?

You're obviously very excited by the potential that Turbo offers. That's great! But have you considered the field day the Intel marketing department would have with your ideas? Based on all that they'd come up with enough logos, stickers and new feature brand names to justify bumping the price right up to $400!

But bringing things back down to reality, and especially in the light of your comments about Arrandale, don't you see the Clarkdale Turbos as rather limited?

 

[cyberkuberiah]

http://74.125.153.132/search?q=cache:bDSgz3z1zj4J:www.ocforums.com/archive/index.php/t-375965.html+processor+binning&cd=1&hl=en&ct=clnk

explains processor binning in detail . is essence a processor which can hit more frequency at low voltage is higher binned , BUT sometimes , even though they are capable of higher performance , the marketing factor kicks in . yes , this is what seems to be the case in clarkdale , especially when arrandale can make this much of a jump . two things come to my mind (thanks for pointing it out , i kinda missed this ) :-

1. the clock speeds of both clarkdale and arrandale are not correct as rumored in the market as intel has not released any official statements .

2.the clock speeds are indeed correct , as that 4.7 ghz on air mentions , then clarkdale should debut at 4.0 ghz . let us consider why intel marketing would kill this :- ill keep posting what comes to mind , and you too :-

 

[cyberkuberiah]

and yes , coming back to reality , very true and needed

let us consider three processors :

processor A : single core 4ghz (hypothetical)

processor B : dual core 4ghz (clarkdale at stock , better than cooled core 2 duo e84xx and more performance per clock )

processor C : intel core i7 920 at 4.0 (again , cooling required ) .

in single threaded apps , A , B , C all are good but A is enough , B and C are overkill .

in dual threaded (yes , many games dont use more than two ) , B and C perform good , but B is enough and C is overkill.

in 3/4 threaded apps , C alone wins . but what if in this case it was core 2 quad series ? 2.66 ghz nehalem takes out 3.2ghz c2q . this means 20 % more performance per clock .

now , 4.0 ghz nehalem dual == 2.0 ghz nehalem quad (yes you can do this , if the app is multithreaded , and even better with hyper threading) == 2.4 (20 % more ) Ghz core 2 quad (even the ones with 12m cache ) .

overclocked clarkdale at 4.4 will be comparable to discontinued q9450 at 2.66 .

overclocked clarkdale at 4.8(i know , lets hope with possible revisions and steppings etc ) is comparable to 2.4 nehalem quad *(1.20 more PPC) = q 9550 , thats right .

thats the whole story as per my imagination

 

[cyberkuberiah]

this is where product binning comes in again . it would be interesting to see the speed limits of the whole westmere dual core lineup ! what do you say ! cpu steppings and revisions will improve surely , just like the 200 mhz improvement the amd 965 BE got today .

 

[cyberkuberiah]

if 2.26 is the "turbo" , and arrandale is default 133x9 . but will it go lower than 9x ? now that would make a nice laptop , or , as a joke , an excellent netbook .

 

[b82]

1. the clock speeds of both clarkdale and arrandale are not correct as rumored in the market as intel has not released any official statements .

2.the clock speeds are indeed correct , as that 4.7 ghz on air mentions , then clarkdale should debut at 4.0 ghz . let us consider why intel marketing would kill this :- ill keep posting what comes to mind , and you too :-

a) Thanks for information on processor binning

b) The clock speeds are incorrect? I will see in 2 months. You say the 'Clarkdale should debut at 4.0 ghz'. You seem to be basing this purely on overclocking headroom.
1) That headroom only goes up to 4.7 apparently. So given that the there's got to be a range of products within Core i3 and Core i5 Clarkdale, you are implying (if you mean by debut, 'start at') that the top of the range models will have to sit on the limit of what early overclockers have managed. That is not plausible.
2) Overclocking headroom is not Intel's only consideration. Many manufacturers now feel the need to bring products to market that are more energy efficient than the previous products - to show an improvement. I should have thought Intel was one such manufacturer. Of course some of the TDPs are slightly higher (than some Wolfdales), but they will point out that this now includes the IGP.

 

[cyberkuberiah]

yes , of course one has to watch out for tdp , and the stock voltage will matter as tdp increases with the square of voltage .

http://www.tomshardware.com/reviews/core-i5-750-overclock,2438-11.html

in the above chart , going from 2.66 to 4.0 (1.5x) cause a 90 watt increase in peak system power consumption . so it would be interesting to see the actual power -voltage-frequency of the released chips .

i hope there is a bios option to completely disable the igp if needed , like turbo on/off . lets wait for westmere .

and , the rumored 130w for corei9 at 3.09 ghz for 6 cores is a nice indication of the benefits of 32nm .