Intel Skylake / Kaby Lake

[ShintaiDK]

My 88w tdp 4790k cpu consumes without turbo enabled, 125w on prime95 under fma3. And no, voltage doesnt spike +0.1v under such load, 4.0/1.0v all they way during the load is applied.

Sorry to burst your bubble, but intel tdp is only comparable to actual power consumption on anything that isnt avx2/fma3.

I have to disappoint you again.

91W usage, 91W TDP in the most demanding FFT load.

Using the maximum power consumption FFTs give way below TDP.

75W used with 91W TDP.

All that is done with a 115W max limit set in BIOS. So there is nothing as such preventing the CPU from going all the way to 115W if it wanted to.

 

[Abwx]

I have to disappoint you again.

91W usage, 91W TDP in the most demanding FFT load.

Using the maximum power consumption FFTs give way below TDP.

75W used with 91W TDP.

All that is done with a 115W max limit set in BIOS. So there is nothing as such preventing the CPU from going all the way to 115W if it wanted to.

Lol, that s not the most demanding FFTs because there s several workers enabled that will fill the cache, it wouldnt had been the same if only one worker using the full cache was the only loading app.

Nice try anyway, for the 4790K Hardware.fr measured 80W with Fritzbench and 96W under prime 95, they specified that they didnt use the most demanding task of Prime , otherwise TDP increase 25% over the value they measured.

http://www.hardware.fr/focus/98/core-i7-4790k-test-devil-s-canyon-debarque.html

 

[ShintaiDK]

Lol, that s not the most demanding FFTs because there s several workers enabled that will fill the cache, it wouldnt had been the same if only one worker using the full cache was the only loading app.

Are you saying Prime95 isn't worth using for this test? So what do you wish to test with?

Nice try anyway, for the 4790K Hardware.fr measured 80W with Fritzbench and 96W under prime 95, they specified that they didnt use the most demanding task of Prime , otherwise TDP increase 25% over the value they measured.

http://www.hardware.fr/focus/98/core-i7-4790k-test-devil-s-canyon-debarque.html

You only confirm the usage is at TDP or below. hardware.fr measures the 12V. And that has to pass the VRM before hitting the CPU.

 

[JoeRambo]

Are you saying Prime95 isn't worth using for this test? So what do you wish to test with?
You only confirm the usage is at TDP or below. hardware.fr measures the 12V. And that has to pass the VRM before hitting the CPU.

Imo Linx with latest Intel Linpack binaries, running 4 threads with manually assigned CPU affinity for 4 real cores ( 0, 2, 4, 6 ) is among largests loads possible and in NO WAY representative of what most real world loads would use.

Sad fact is that certain guys here keep on spinning this wattage nonsense when their beloved company products use boatload of watts and have fraction of GFlops at same time...

 

[Abwx]

Are you saying Prime95 isn't worth using for this test? So what do you wish to test with?

You only confirm the usage is at TDP or below. hardware.fr measures the 12V. And that has to pass the VRM before hitting the CPU.

I didnt say that it was prime 95 but your methodology wich is flawed..

As for Hfr they measure the 12V rail that feed the CPU, losses are 10% through the CPU VRMs and were accounted in the numbers i posted.

To get back on Prime they said that 256K FFTs are the more power consuming, likely because the app can be entirely filled in the L2 cache so the only bottleneck will be the L2 bandwith and pipeline execution speed.

With several workers the L3 cache or even the RAM will be needed and this will slow down the core execution speed and hence the TDP.

Imo Linx with latest Intel Linpack binaries, running 4 threads with manually assigned CPU affinity for 4 real cores ( 0, 2, 4, 6 ) is among largests loads possible and in NO WAY representative of what most real world loads would use.

Sad fact is that certain guys here keep on spinning this wattage nonsense when their beloved company products use boatload of watts and have fraction of GFlops at same time...

It s you who is spinning the fact, a FX can sustain its full throughput under Prime 95 at 125W wich is its stated TDP while Intel s have to exceed their rated TDP, if they are throttled then their average throughput can in now way be the one advertised, so much for the Gflps throughput fraction...

Other than this Linpack can reach the peak power of Prime 95 but not the average power, overall its average power drain is 10-20% lower than Prime 95, wich is still above about any regular app.

 

[ShintaiDK]

I dont see the VRM loss was accounted for.

You better come up with some facts....

Other than this Linpack can reach the peak power of Prime 95 but not the average power, overall its average power drain is 10-20% lower than Prime 95, wich is still above about any regular app.

So IDCs measurements and MSI and Asrock calling them 140W CPUs a lie?

Funny considering how baseless you attack other products even when presented with facts.

 

[ShintaiDK]

Sad fact is that certain guys here keep on spinning this wattage nonsense when their beloved company products use boatload of watts and have fraction of GFlops at same time...

Yep. Specially considering those people that will never get cheated by any nonsense (Server/cloud). Dropped their favourite companies products like a turd because they lost badly in all metrics.

 

[JoeRambo]

It s you who is spinning the fact, a FX can sustain its full throughput under Prime 95 at 125W wich is its stated TDP while Intel s have to exceed their rated TDP, if they are throttled then their average throughput can in now way be the one advertised, so much for the Gflps throughput fraction...

I am 100% certain that you can find THAT condition where bad chip in bad thermals running near 100C, with higher voltage on MB that is feeding higher voltage to CPU than requested etc. Same applies to AMD btw.

How relevant is that special condition to real world TDPs when original claims were straight from conspiracy theory books about Skylake being 666W TDP chip and busting 91w TDP due to some evil of Intel?

 

[Abwx]

I dont see the VRM loss was accounted for.

You better come up with some facts....

Take the CPU 12V rail power and do the multiplication by 0.9...

So IDCs measurements and MSI and Asrock calling them 140W CPUs a lie?

Asrock said nothing of the sort, as for IDC go check his "review", he did set the voltage at a fixed 1.377V rather than the stock value of 1.24V, this way he artificialy increased power by a ratio (1.377/1.24)^2 = 1.233.

Hence the 125W TDP became "magically" 125 x 1.233 = 154W, it was easy to get by removing the automated LLC regulation and forcing the max VID value wich is effectively 1.377V, doing the same with say a Haswell would push a 84W official TDP to more than 130W according to Intel s datasheet.

 

[Nothingness]

I'm not sure it's been mentioned in this thread, but it looks like Intel is putting pressure on OEM to not have OC on platforms not designed for it:
http://www.tomshardware.com/news/asrock-non-z170-chipset-overclock,30915.html

Update, 1/21/16, 12:05pm PT: During CES, ASRock showed us some non-Z170 motherboards designed for overclocking, and we produced this story below. We spoke with many other companies during the show, and there were varying opinions on the matter. We followed up with this article. Behind the scenes, Intel hasn't been pleased. In fact, one source at Intel said outright that the company did not support these efforts in the slightest. Now, word comes from ASRock that the company has decided it will not be moving forward with these boards. Here is the statement, in full:

ASRock has decided to remove SKY OC technology from these motherboards’ feature lists since it is not compliant with Intel Skylake CPU specifications.
​

Super Micro also got in touch with us to inform us that its non-Z170 overclocking motherboards will have C-State and Turbo Boost enabled when the BCLK is set to 100 MHz, but the system automatically disables these features to ensure stability when overclocking beyond 100 MHz.

I certainly wouldn't blame Intel for doing that, but I'm not sure OC fans will agree with me.

 

[Sweepr]

I'm not sure it's been mentioned in this thread, but it looks like Intel is putting pressure on OEM to not have OC on platforms not designed for it:
http://www.tomshardware.com/news/asrock-non-z170-chipset-overclock,30915.html
I certainly wouldn't blame Intel for doing that, but I'm not sure OC fans will agree with me.

Regarding non-Z170 motherboards, probably because of this:

Asus, Asrock and MSI told us that non-Z170 chipsets modified to enable overclocking will lose Hyper-Threading support and will be unable to use C-states to reduce voltage and power consumption when the system is idling.

And of course Intel wouldn't like to see BCLK OC marketed as a motherboard feature, they would rather sell us K chips. Doesn't mean they will disable it in any way.

 

[Nothingness]

And of course Intel wouldn't like to see BCLK OC marketed as a motherboard feature, they would rather sell us K chips.

Even though it's a matter of economics, IMHO it's also a matter of reputation: you wouldn't want to have a wave of furious people complaining because OC doesn't work despite OEM claiming their boards support OC even though these are not real OC platforms.

Doesn't mean they will disable it in any way.

Does Intel have a way to disable it?

 

[ShintaiDK]

Take the CPU 12V rail power and do the multiplication by 0.9...

So its just made up by you?

Asrock said nothing of the sort, as for IDC go check his "review", he did set the voltage at a fixed 1.377V rather than the stock value of 1.24V, this way he artificialy increased power by a ratio (1.377/1.24)^2 = 1.233.

Hence the 125W TDP became "magically" 125 x 1.233 = 154W, it was easy to get by removing the automated LLC regulation and forcing the max VID value wich is effectively 1.377V, doing the same with say a Haswell would push a 84W official TDP to more than 130W according to Intel s datasheet.

Could you link me to the datasheet from AMD with all the specifications?

 

[Abwx]

So its just made up by you?

It s not made up, it s you who dont understand how electronics circuits works, switch mode power supplies have 10% losses on average, if 100W enter a CPU regulation 90W will be actualy comsumed by the CPU.

Now please stop your deffamation because Hardware.fr numbers are explicit, here the link again, show us what is made up :

http://www.hardware.fr/focus/98/core-i7-4790k-test-devil-s-canyon-debarque.html

So..?..

Could you link me to the datasheet from AMD with all the specifications?

It is stated in their official specs, if you dont trust them then buy a chip and do some measurements, and preferably without using fraudulous parameters.

That said there s tons of reviews and none point more than 125W under Prime 95 or Linpack, only the intel chips exceed their rated TDP as proved by those same tests, that it please you or not numbers dont lie.

 

[ShintaiDK]

So you cant link it and you just make guesses. Super

Let me know when you have some facts.

Considering how easy it is to measure package power. There shouldn't be any doubt.

 

[MrTeal]

It s not made up, it s you who dont understand how electronics circuits works, switch mode power supplies have 10% losses on average, if 100W enter a CPU regulation 90W will be actualy comsumed by the CPU.

Now please stop your deffamation because Hardware.fr numbers are explicit, here the link again, show us what is made up :

http://www.hardware.fr/focus/98/core-i7-4790k-test-devil-s-canyon-debarque.html

So..?..

It is stated in their official specs, if you dont trust them then buy a chip and do some measurements, and preferably without using fraudulous parameters.

That said there s tons of reviews and none point more than 125W under Prime 95 or Linpack, only the intel chips exceed their rated TDP as proved by those same tests, that it please you or not numbers dont lie.

Got a source for your number that SMPS have 10% losses on average?

 

[MarkizSchnitzel]

Shouldn't there be a fixed workload scenarion, and than just measure wall power and time and voila?

Energy per work and than also idle should be the only metrics anyone cares about?

Or am I missing something?

 

[Abwx]

Got a source for your number that SMPS have 10% losses on average?

Because you couldnt google the thing.?..

https://en.wikipedia.org/wiki/Switched-mode_power_supply

You can also check it more practicaly here although a 350W PSU has more losses than a CPU regulation :

http://www.computerbase.de/2015-10/...sea-sonic-super-flower/5/#abschnitt_effizienz

 

[MrTeal]

Because you couldnt google the thing.?..

https://en.wikipedia.org/wiki/Switched-mode_power_supply

You can also check it more practicaly here although a 350W PSU has more losses than a CPU regulation :

http://www.computerbase.de/2015-10/...sea-sonic-super-flower/5/#abschnitt_effizienz

You think an active PFC double forward AC/DC SMPS design like you'd see in a platinum power supply is the same thing you'll see in a synchronous buck on the VRM of a motherboard?

A very good VRM might get 90% at the peak of its efficiency. For example, the IR3550 driver/fets used on some top end MBs have a peak efficiency converting 12V to 1.2V of 93.5%, but that is just the mosfets itself and is of course at peak efficiency, which trails off reasonably quickly outside its peak. Add in inductor losses and resistive losses in the PCB and socket, and a peak 90% efficient VRM over any kind of broad range in converting 12V into the CPU package is a very good design.
Taking a CPU load and multiplying it by 0.9 is a gross over-simplification and probably quite optimistic. Saying "switch mode power supplies have 10% losses on average" is just silly.

 

[Abwx]

You think an active PFC double forward AC/DC SMPS design like you'd see in a platinum power supply is the same thing you'll see in a synchronous buck on the VRM of a motherboard?

A very good VRM might get 90% at the peak of its efficiency. For example, the IR3550 driver/fets used on some top end MBs have a peak efficiency converting 12V to 1.2V of 93.5%, but that is just the mosfets itself and is of course at peak efficiency, which trails off reasonably quickly outside its peak. Add in inductor losses and resistive losses in the PCB and socket, and a peak 90% efficient VRM over any kind of broad range in converting 12V into the CPU package is a very good design.
Taking a CPU load and multiplying it by 0.9 is a gross over-simplification and probably quite optimistic. Saying "switch mode power supplies have 10% losses on average" is just silly.

Efficency is largely 90%, it s not by chance that multiphases PSUs are used in MBs, power losses are actualy low because the current is distributed between as much mosfets.

At 90W in the CPU and a 90% effiency this let 10W in the VRMs and attached components including the MB copper trace, that s considerable power to dissipate, so i would say that anything below 90% efficency at the CPU supply level is bad design.

FTR i measured a few laptops PSUs, best one is a 70W one wich has 94-95% effiency at 65W output power, but there was nothing that was below 90%, and that s not multiphases designs.

Edit : Hfr power and overlocking tests are done on the 12 phase MB Asus Z87 pro....

 

[Dufus]

Intersil ISL95824 4+2 Multiphase PWM Regulator for Intel IMVP8 Desktop CPUs

 

[JDG1980]

I'm not sure it's been mentioned in this thread, but it looks like Intel is putting pressure on OEM to not have OC on platforms not designed for it:
http://www.tomshardware.com/news/asrock-non-z170-chipset-overclock,30915.html
I certainly wouldn't blame Intel for doing that, but I'm not sure OC fans will agree with me.

I'm starting to hope for a big market crash. The major players in the tech industry are becoming far too arrogant: Intel's attempts to crack down on BCLK OC plus upcoming price hikes to $1500 on HEDT, Microsoft's incessant attempts to jam Win10 (spyware) down our throats plus the retroactive OneDrive quotas, Google's pioneering of new ways to invade our privacy... I want to see the whole house of cards come tumbling down.

 

[Burpo]

Price hikes? :\
6 years ago the 6 core Xeons were more than $1500!

http://ark.intel.com/products/47916/Intel-Xeon-Processor-X5680-12M-Cache-3_33-GHz-6_40-GTs-Intel-QPI

 

[Arachnotronic]

I'm starting to hope for a big market crash. The major players in the tech industry are becoming far too arrogant: Intel's attempts to crack down on BCLK OC plus upcoming price hikes to $1500 on HEDT, Microsoft's incessant attempts to jam Win10 (spyware) down our throats plus the retroactive OneDrive quotas, Google's pioneering of new ways to invade our privacy... I want to see the whole house of cards come tumbling down.

Guess you want people to lose their jobs (bad corporate financial performances -> workforce reductions) and for people to see their 401k retirement accounts crash...all over a $1500 halo CPU that nobody is forcing you to buy and dumb overclocking features because you're apparently too cheap to spend the extra $20 to buy a z170 board.

What is wrong with you?

 

[III-V]

I'm starting to hope for a big market crash. The major players in the tech industry are becoming far too arrogant: Intel's attempts to crack down on BCLK OC plus upcoming price hikes to $1500 on HEDT, Microsoft's incessant attempts to jam Win10 (spyware) down our throats plus the retroactive OneDrive quotas, Google's pioneering of new ways to invade our privacy... I want to see the whole house of cards come tumbling down.

A market crash would only worsen the competitive state of things (short term)... a lot of the smaller guys will go splat, while the larger ones will have economies of scale, assets, and cash reserves to weather any storm.