turbo boost 3.0

O

oleNBR

Limp Gawd
Joined
Dec 9, 2016
Messages
372
started with broadwell-HEDT now onto skylake-X & kabylake-X mid next year, is anyone using broadwell-HEDT currently that has tested turbo 3.0?

tbh im quite interested to see how it works, i have seen review from toms and anand explain how intel make you install a software which lists all the cores from best quality to worst, and of course the highest core can clock best but we have literally no control over it.

for example core 0 to core 7 in an 8 cores cpu, if core 0 and core 1 are extremely good, you can literally disable core 2-7 and overclock crazy on core 0 and 1. if core 0 and 1 are poor, even if you disable 2-7 and overclock 0-1 won't get you anywhere.

and for the most part, we'd want to clock all of our CPU cores at say 4.4ghz, then what is the point of turbo boost 3.0? does it add additional frequency to the best core? does anyone know?
 
Basically, it runs single threaded apps on a core that's highly clocked, helping to decrease the penalty for having more cores and it maintains this high clock speed for that core even if you load the rest of the cores with additional tasks (unlike 2.0 which throttles ALL cores down to the max all core frequency). The technology is of limited use for many overclockers (who overclock all their cores anyway), but I see it being of way more value if/when it rolls out to high core count chips where overclocking all the cores isn't an option. If an HCC chip had Turbo Boost 3.0 for four cores (with each running at 4.0GHz), one could run single and lightly threaded apps on the high speed cores and have the heavily multi-threaded apps allocated to the slower cores.
 
lutjens said:
Basically, it runs single threaded apps on a core that's highly clocked, helping to decrease the penalty for having more cores and it maintains this high clock speed for that core even if you load the rest of the cores with additional tasks (unlike 2.0 which throttles ALL cores down to the max all core frequency). The technology is of limited use for many overclockers (who overclock all their cores anyway), but I see it being of way more value if/when it rolls out to high core count chips where overclocking all the cores isn't an option. If an HCC chip had Turbo Boost 3.0 for four cores (with each running at 4.0GHz), one could run single and lightly threaded apps on the high speed cores and have the heavily multi-threaded apps allocated to the slower cores.
Click to expand...

wait so what you're saying is i can say choose my core 5 and clock it to say 5ghz while rest of all my other core stay at 4.2ghz? what would happen when a software uses all the cores, would everything still only run at 4.2ghz? i dont really get how it will work out say if a software running single threaded at 5ghz while i run another software that uses multi thread but rest of the cores at 4.2ghz i dont know how it'll turn out. however from what you said sounds really amazing, a good quality core can literally go up to 5.5ghz without having too much problem voltage/heat wise.
 
SomeGuy133 said:
maybe just read intels website first and other sites? It is pretty basic

http://www.intel.com/content/www/us...y/turbo-boost/turbo-boost-max-technology.html
http://www.intel.com/content/www/us/en/support/processors/000021587.html
http://www.anandtech.com/show/10337...6900k-6850k-and-6800k-tested-up-to-10-cores/2
https://en.wikipedia.org/wiki/Intel_Turbo_Boost
Click to expand...

the links you have mentioned doesn't seem to include the one im looking for. all of these review sites do these two things and only these.
1. overclock the cpu on all cores see how high it can go
2. run turbo boost max 3.0 on default stock clock 3 ghz and 1 core turbo 3.9ghz

what I wanted to find out was, if its capable of overclocking 10 cores at say 4ghz, and then 1 core turbo to 5+ ghz using the bios and TBM3.0 software in windows.. and after long search it seems it IS doable but still lacking confirmation from multiple sites or people talking about it. for those who wanted to know this is only one i could find.

Point 5 and 6 from PCworld - http://www.pcworld.com/article/3076...out-intels-crazy-10-core-broadwell-e-cpu.html
Conclusion page from bit-tech - https://www.bit-tech.net/hardware/2016/05/31/intel-core-i7-6950x-broadwell-e-review/8

this proves what i was looking for and hopefully it really runs 1 core at the frequency we set it to, assuming the bios is good..
 
oleNBR said:
the links you have mentioned doesn't seem to include the one im looking for. all of these review sites do these two things and only these.
1. overclock the cpu on all cores see how high it can go
2. run turbo boost max 3.0 on default stock clock 3 ghz and 1 core turbo 3.9ghz

what I wanted to find out was, if its capable of overclocking 10 cores at say 4ghz, and then 1 core turbo to 5+ ghz using the bios and TBM3.0 software in windows.. and after long search it seems it IS doable but still lacking confirmation from multiple sites or people talking about it. for those who wanted to know this is only one i could find.

Point 5 and 6 from PCworld - http://www.pcworld.com/article/3076...out-intels-crazy-10-core-broadwell-e-cpu.html
Conclusion page from bit-tech - https://www.bit-tech.net/hardware/2016/05/31/intel-core-i7-6950x-broadwell-e-review/8

this proves what i was looking for and hopefully it really runs 1 core at the frequency we set it to, assuming the bios is good..
Click to expand...
There are basic settings you can do with OCing.

There is base freq and turbo freq.

Base freq you can only lower. There is an internal algorithm for voltages that works different for base and turbo. It isn't very intelligent.

If you run 26x base freq and 30x (26+4) turbo you get 1.3v(making this number up) If you do 20x base and 30x turbo you get 1.32v. Setting lower base freq is only useful for battery/energy saving but kills you when you turbo. I tested this on IB. its useful for laptops if you want to allow software base voltage of 900mhz to conserve battery. You have to restart PC to change base voltage (at least back in IB). so i was able to set like 1.2ghz base feq on IB and set turbo to 4.0Ghz which allowed me on battery to force 1.2 or 1.6GHz when only surfing and typing in class where super slow CPU was okay and gae me like 50% more juice back in IB but this is generally pointless now for current gen due to idle wattage is so damn low as is. IB idle wattage was like 7-14 w TDP and skylake is like 2-7w TDP or something stupid low so not really worth while but still works.

You can set turbo multipliers for each core 1, 2, 3 ect.
http://i.imgur.com/TxrEsOw.jpg

This can be done in BIOS or software.

You can also force fixed freq which uses a lot more power but makes your PC snappier. This is much more apparent on older gens (broadwell and older). Skylake really made speedshift a lot faster but still snappiness gains are possible with forcing max feq.

This is most noticeable with opening windows or small files or programs. The snappiness is quite noticeable depending on the task. Tasks that take 50ms to load are the most significantly affected like explorer or opening chrome for first time.
http://www.anandtech.com/show/9751/examining-intel-skylake-speed-shift-more-responsive-processors


You can also do fixed voltage or do adaptive. I usually do fixed voltage to ensure stability. Generally, you dont want to do more than .75v adaptive voltage or instability can happen with shifting voltages.

There are also ways to disable weak core to set a better overclock but thats a shit ton of time testing to find weakest core. OCing enthusiasts do this for setting records. There are practical applications for this but generally not worth the effort and time.

I do not know if you can program independent freqs that are static to an exact core. That might be possible but i have never heard of anyone doing that.

Does this answer everything?
 
Last edited:
SomeGuy133 said:
There are basic settings you can do with OCing.

There is base freq and turbo freq.

Base freq you can only lower. There is an internal algorithm for voltages that works different for base and turbo. It isn't very intelligent.

If you run 26x base freq and 30x (26+4) turbo you get 1.3v(making this number up) If you do 20x base and 30x turbo you get 1.32v. Setting lower base freq is only useful for battery/energy saving but kills you when you turbo. I tested this on IB. its useful for laptops if you want to allow software base voltage of 900mhz to conserve battery. You have to restart PC to change base voltage (at least back in IB). so i was able to set like 1.2ghz base feq on IB and set turbo to 4.0Ghz which allowed me on battery to force 1.2 or 1.6GHz when only surfing and typing in class where super slow CPU was okay and gae me like 50% more juice back in IB but this is generally pointless now for current gen due to idle wattage is so damn low as is. IB idle wattage was like 7-14 w TDP and skylake is like 2-7w TDP or something stupid low so not really worth while but still works.

You can set turbo multipliers for each core 1, 2, 3 ect.
http://i.imgur.com/TxrEsOw.jpg

This can be done in BIOS or software.

You can also force fixed freq which uses a lot more power but makes your PC snappier. This is much more apparent on older gens (broadwell and older). Skylake really made speedshift a lot faster but still snappiness gains are possible with forcing max feq.

This is most noticeable with opening windows or small files or programs. The snappiness is quite noticeable depending on the task. Tasks that take 50ms to load are the most significantly affected like explorer or opening chrome for first time.
http://www.anandtech.com/show/9751/examining-intel-skylake-speed-shift-more-responsive-processors


You can also do fixed voltage or do adaptive. I usually do fixed voltage to ensure stability. Generally, you dont want to do more than .75v adaptive voltage or instability can happen with shifting voltages.

There are also ways to disable weak core to set a better overclock but thats a shit ton of time testing to find weakest core. OCing enthusiasts do this for setting records. There are practical applications for this but generally not worth the effort and time.

I do not know if you can program independent freqs that are static to an exact core. That might be possible but i have never heard of anyone doing that.

Does this answer everything?
Click to expand...

yea i know the voltage thing you're taking about but i have tried with ivy-E. 4960x is 3.7ghz turbo all cores i think but then i tried using XTU and throttlestop to set 1 core at say 4.8ghz while rest are 4.5ghz with a working voltage that run stables at 4.7.. if its 1 core at 4.8 it will never use the voltage for 4.7 as that 4.7 was for all cores. then i run my software single threaded either cinebench R15 ST or any software i set windows affinity to 1 core, they dont run at 4.8ghz at all, only at max 4.5ghz..

this prompted me to look into TBM 3.0 which looks like its finally possible to have it like that. what do you think?
 
oleNBR said:
yea i know the voltage thing you're taking about but i have tried with ivy-E. 4960x is 3.7ghz turbo all cores i think but then i tried using XTU and throttlestop to set 1 core at say 4.8ghz while rest are 4.5ghz with a working voltage that run stables at 4.7.. if its 1 core at 4.8 it will never use the voltage for 4.7 as that 4.7 was for all cores. then i run my software single threaded either cinebench R15 ST or any software i set windows affinity to 1 core, they dont run at 4.8ghz at all, only at max 4.5ghz..

this prompted me to look into TBM 3.0 which looks like its finally possible to have it like that. what do you think?
Click to expand...
I have talked to unclewebbs via email in the past about some of how intel CPUs work (this was IB/HW era) but i am unsure on how well turbo boost really works or how well programs like TS actually reads the CPU. The C0% or whatever is the best sign of a full load IIRC. If it shows anything above 0 C0% that means it is 100% XX.X% of the time of a second. So 25% C0 is 25% of a single second it is maxed out. Also how windows works it shifts loads around the CPU so a single thread program bounces between all 4 cores so you might not notice it at 4.8GHz but shows 4.5GHz but the single thread maybe running at 4.8GHz. I am not sure how well turbo boost really works. I have tried messing around in the past but this was like 3-5 years ago and i don't recall my findings. I tried 1 core 4GHz and the others at 2.6GHz but thats not really scientific but will show real world practicality. Because maybe turbo boost does technically work but in reality there is always background stuff that fuck it up???

I can't give you an exact answer because i either don;t remember or couldn't find one several years ago.

I do feel your best off just putting the max you can get on all cores and leaving it at that.

Intel voltages are not super smart so understand its hyper aggressive with voltages, which is good in terms of stability. Also HW has FIVR which helps. I never got around to finishing my testing. I was going to test the practicality of FIVR and how well it worked comparing IB to HW. I was interested in its regards to ultra-portable quad systems and the future of using quads that could adapt to various cooling situations. As in a quad running at fast 1/2 core(4/2.5GHz) or slow 4 core (1.6GHz) when it is limited to 15w TDP and when it has 45w TDP it runs at full.

FIVR is supposedly really good at boosting overall efficiency when it comes to unused cores.

So my point is i don't think TS even can monitor each core voltage (on FIVR systems) So there are clear limitations what these programs can see that is going on in the CPU plus the CPU changes these things 10s-100s-1000s of times a second...maybe more? Speedshift in SKL is a prime example. You can't really observe Speed shift because it is working at the ms speed and TS doesn't poll that fast even in fast/more data mode. It might be possible but that requires cpu cycles to poll that fast.

Does this help any further?

Again sorry last time i did this testing was years ago when i was talking to unclewebb about this stuff. Really want to finish my testing but doubt i'll ever get around to it.
 
oleNBR said:
yea i know the voltage thing you're taking about but i have tried with ivy-E. 4960x is 3.7ghz turbo all cores i think but then i tried using XTU and throttlestop to set 1 core at say 4.8ghz while rest are 4.5ghz with a working voltage that run stables at 4.7.. if its 1 core at 4.8 it will never use the voltage for 4.7 as that 4.7 was for all cores. then i run my software single threaded either cinebench R15 ST or any software i set windows affinity to 1 core, they dont run at 4.8ghz at all, only at max 4.5ghz..

this prompted me to look into TBM 3.0 which looks like its finally possible to have it like that. what do you think?
Click to expand...
oh also something i learned from IB was 1.8GHz was the sweet spot in regards to performance per watt. 1.8GHz was more efficient than any other GHz IIRC. (1.6 might have been more efficiency but i have written 1.8 in my old notes.)


just some random info is your interested.
 
SomeGuy133 said:
I have talked to unclewebbs via email in the past about some of how intel CPUs work (this was IB/HW era) but i am unsure on how well turbo boost really works or how well programs like TS actually reads the CPU. The C0% or whatever is the best sign of a full load IIRC. If it shows anything above 0 C0% that means it is 100% XX.X% of the time of a second. So 25% C0 is 25% of a single second it is maxed out. Also how windows works it shifts loads around the CPU so a single thread program bounces between all 4 cores so you might not notice it at 4.8GHz but shows 4.5GHz but the single thread maybe running at 4.8GHz. I am not sure how well turbo boost really works. I have tried messing around in the past but this was like 3-5 years ago and i don't recall my findings. I tried 1 core 4GHz and the others at 2.6GHz but thats not really scientific but will show real world practicality. Because maybe turbo boost does technically work but in reality there is always background stuff that fuck it up???

I can't give you an exact answer because i either don;t remember or couldn't find one several years ago.

I do feel your best off just putting the max you can get on all cores and leaving it at that.

Intel voltages are not super smart so understand its hyper aggressive with voltages, which is good in terms of stability. Also HW has FIVR which helps. I never got around to finishing my testing. I was going to test the practicality of FIVR and how well it worked comparing IB to HW. I was interested in its regards to ultra-portable quad systems and the future of using quads that could adapt to various cooling situations. As in a quad running at fast 1/2 core(4/2.5GHz) or slow 4 core (1.6GHz) when it is limited to 15w TDP and when it has 45w TDP it runs at full.

FIVR is supposedly really good at boosting overall efficiency when it comes to unused cores.

So my point is i don't think TS even can monitor each core voltage (on FIVR systems) So there are clear limitations what these programs can see that is going on in the CPU plus the CPU changes these things 10s-100s-1000s of times a second...maybe more? Speedshift in SKL is a prime example. You can't really observe Speed shift because it is working at the ms speed and TS doesn't poll that fast even in fast/more data mode. It might be possible but that requires cpu cycles to poll that fast.

Does this help any further?

Again sorry last time i did this testing was years ago when i was talking to unclewebb about this stuff. Really want to finish my testing but doubt i'll ever get around to it.
Click to expand...


the thing is, i dont solely rely on throttlestop, i also use hwinfo and hwmonitor etc to monitor the frequency, heres the interesting point. if i leave my CPU and default stock everything stock, no OC in bios nothing, it'll run 1 thread at the intended speed set by intel however once anything is overclocked, it won't do intended no matter what I set in throttlestop or even XTU.

in XTU i set it again 4.8ghz core 0 and every other core at 4.5ghz, samething in throttlestop and doesn't do a thing where as if its left runing stock it'll do as intended. take a look at the two links article i sent you and they quoted these...

5. It has per-core overclocking
Besides Turbo Boost Max 3.0, Broadwell-E will also allow you to overclock particular cores. With Haswell-E, you could overclock cores based on thread load. For example, with two cores in use, you could tell the CPU to overclock to, say, 4.5GHz. The OS and Haswell-E though, would just throw the overclock at any cores. With Broadwell-E, you can specify that it overclock Core 1 and Core 5, which, you’ve determined through testing, overclock the best.


We could get per-core performance as good as a Core i7-6700K “Skylake” chip by overclocking our Broadwell-E to 4.5GHz.

6. We had a decent overclocking experience
We’re loath to pass judgement on an entire series of CPUs based on our one early press sample, but our experience overclocking was mostly satisfactory. We easily pushed our 10-core Core i7-6950X to 4GHz across all cores without even resorting to “scary” things like voltage tweaks. We then pushed the core that Intel identified at the factory as being the best to 4.5GHz on single-threaded loads
Click to expand...

This turned out to be 4.4GHz using a vcore of 1.44V, and it was heat rather than instability which was the final nail in the coffin.

That's excessive and shouldn't be used as a 24/7 overclock, so the ability to reach 4.6GHz using Turbo Boost Max 3.0 with far less voltage is very useful unless you're gunning for maximum performance from all ten cores.
Click to expand...

expand the quotes..
 
oleNBR said:
the thing is, i dont solely rely on throttlestop, i also use hwinfo and hwmonitor etc to monitor the frequency, heres the interesting point. if i leave my CPU and default stock everything stock, no OC in bios nothing, it'll run 1 thread at the intended speed set by intel however once anything is overclocked, it won't do intended no matter what I set in throttlestop or even XTU.

in XTU i set it again 4.8ghz core 0 and every other core at 4.5ghz, samething in throttlestop and doesn't do a thing where as if its left runing stock it'll do as intended. take a look at the two links article i sent you and they quoted these...





expand the quotes..
Click to expand...
interesting news to me.
 
You must log in or register to reply here.
Back
Top