FRONTPAGE Intel i7 4790K Devil's Canyon CPU Review

outhouse · Jul 16, 2014

My IHS has a small gap betwen the IHS and the PCB when sitting on the die.

Just enough room for silicone to hold it in place, and yet still sit square on the die.

paper is way to thick and would not let the IHS sit on the die.

I think a slight swipe of silicone on just the ears might hold it perfect.

My only fear here with CLU is having the IHS move around when closing the latch holding the CPU in, scraping off some of the tim off the die.

pqwoerituytruei · Jul 18, 2014

rge said:
lol...yeah my cpu either is leaky or tim1 challenged.

1.3v, 4.7gh, cinebench R15 single run was max temps 65C.

lot of people are using tests other than prime with avx/small ffts because temp issue though.

when you use certain instructions say prime's AVX FFT the CPU actually ups its own vcore, so when set a vcore of save 1.38 and you run prime it ups to something like 1.4 or something, there are more extreme examples like something linpack uses https://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download/
at some points that will jump the temp instantly from ~49C to ~60C (stock i5-4690k w/ NH-D14)
what we really need is a way to disable this internal voltage bump, it is a nice feature for non-K-skew chips, but... 2ed option is disabling the instruction set

EarthDog · Jul 18, 2014

I have never heard that before!

I would think that a .02v is a margin of error for voltage fluctuations on load. I also wouldn't imagine it to bump temps over 10C...? Are you sure the CPU actually does that when working with specific instruction sets? Do you have any links I can chew on?

Thanks!

pqwoerituytruei · Jul 18, 2014

I am running the benchmark on linux, currently only CPU thermal sensors are working
i can make you a recording in a few minutes showing the temp change

EarthDog · Jul 18, 2014

The temp change is not what I am looking for. Simply hitting those instruction sets and different sized FFT's will do that. I am looking for some information on the voltage going up by a significant amount (.02 is not significant to me) when using those instructions sets.

ATMINSIDE · Jul 18, 2014

0.02V on the core is most likely just a bit of overshoot by the LLC.

pqwoerituytruei · Jul 18, 2014

https://imgur.com/pQ10BXI

- prime95

https://imgur.com/TufRoSC

- linpack
http://pastebin.com/GcJX78qA -linpack output for the curious
EDIT:
reference for last post:http://linustechtips.com/main/topic/45230-prime95-aida64-intel-burn-in-test/#entry595725

EarthDog · Jul 18, 2014

I cannot see 3rd party hosted images from the office... can you host them internally?

pqwoerituytruei · Jul 18, 2014

edited them into the last post

EarthDog · Jul 18, 2014

Thanks for doing that!

As I stated earlier, .02v change is nothing and really considered 'normal' for varying loads. .02v also wouldn't raise CPU load temps by 10C. I would bet that is the tests in the applications that are doing it. I am not sure how a still photo shows the voltage going up when using a different instruction set though...

I am not too familiar with the details of how those applications work as far as if it always uses all instruction sets or cherry picks per test...etc. P95 I thought used them all and just the size of the FFT's varied...

What you would want to show to help prove your point would be some evidence that a specific test/fft length uses a different instruction set, and then see that voltage jump up... and in a more significant manner than a mere .02v.

Again, a .02v variance between different kinds of loads would be normal.

Thanks again.

RGE · Jul 18, 2014

a week ago I was measuring vcore on my board via readouts with a fluke multimeter, after many had noted higher vcore upon load with these chips. And just spent some time rechecking.

There is an aggressive LLC logic built into fivr on cpu which end user has no control over. The higher the load, the higher the vcore.

CPUZ is reporting the max vid, and does not report vcore. HWM reports both vid and vcore.

With power savings off, vcore manually to 1.292 volts bios, multimeter reads (just ran this with different bios, different settings, but similar results):
1.295v at idle (multimeter reading vcore off leads UD5H GB board)
1.303v with AIDA64 load CPU only
1.308v with AIDA64 CPU + FPU
1.308v with prime 25.1 blend and small ffts same
1.308v with prime 27.9 large ffts with avx
1.311v with prime 27.9 small ffts with avx
1.309v with prime 28.5 large ffts with avx/fma3
1.314v with prime 28.5 small ffts with avx/fma3
1.311v , prime 28.5 increases a little more, running small ffts a little more.

Given there is an aggressive LLC built into cpu vcore logic that is ample reason for increase in vcore without the less logical specific instructions somehow causing an increase.

Mpegger · Jul 18, 2014

When I ran OCCT, what I believe to be the vcore was reading 1.30 when testing, when vcore is set to 1.28 in bios (uefi). I assumed this may have been the LLC being set to max.

EarthDog · Jul 18, 2014

CPUZ is reporting the max vid, and does not report vcore.

If its reporting VID, why would it change on load? VID is stock voltage = static value... right?

Woomack · Jul 18, 2014

EarthDog said:
If its reporting VID, why would it change on load? VID is stock voltage = static value... right?

I think you are right, VID in this case is base voltage for max ratio , in most cases it's between 1.1 and 1.2V ( max for haswell is 1.25V I think ).
CPU-Z is showing actual voltage and you can see the same voltage in various monitoring programs. When power saving options are enabled then depends from load, CPU-Z is showing any voltage between lowest possible value and max VID or max value set manually.

RGE · Jul 18, 2014

Stasio said it was max vid, given his gigabyte, etc connections, he likely knows. But cpuz is definitely not reporting vcore on my 4790k, nor on any of the many people that tested same on oc.net. There people use HWM to show vcore.

My vid changes in bios, and all software reading it with each different setting I boot. My stock vid, stock settings is 1.19. If I boot at 35multi/35 uncore, all rest auto/stock, my vid in bios reads 1.05. Each overclock I boot higher vid in bios and software tracks upwards, and corresponds to cpuz.

I have never seen cpuz change value from load to idle. Only HWM shows vcore and vid, and shows vcore change from idle to load which corresponds to multimeter readings.

Hardware monitor shows both vcore and max vid.
compare cpuz.

Idle vcore bios is 1.292v, hwm shows 1.296vcore idle, 1.292vid, multimeter 1.295v, cpuz reads 1.292v.

load prime 28.5small ffts vcore (bios still 1.292), hwm shows 1.320v (sensor reads in steps), vid in HWM still 1.292, cpuz 1.292, multimeter reads 1.314-1.315.

vids are different for different settings.

Woomack · Jul 18, 2014

Recently CPU-Z has many various issues depends from hardware. Sometimes is reading all correctly, sometimes not. The same as it's reading XMP profiles wrong or timings and memory clock. No matter what it's showing it's not a good tool to check actual voltage.

EarthDog · Jul 18, 2014

Im confused.. I always thought the term VID was the stock voltage for the CPU. Although its range varies, the value for ONE CPU will not.

Somewhere over the past couple of years 'stock vcore' changed to 'VID', which now is being used as 'current vcore'...

... so confused.

RGE · Jul 18, 2014

There is no cpuz out that reads vcore on 4790k that anyone on oc.net has been able to find. I have 1.69.2, 1.69.3, 1.7 (just out), 1.67, gigabytes OC1 and S versions, asus versions.

In past, mobos only showed stock vids and I think most defined vid as stock vid hence stock voltage. But newer software that shows vids, and bioses now show change in vids with sensors that read. So I think users definition changed, although intels has always been same.

here is similar discussion that I joined in 2008, same place I stole that sceenshot from...too lazy to find it in intel specs, im opt33 there.
http://www.overclock.net/t/344663/what-would-make-a-cpu-vid-change

Woomack · Jul 18, 2014

Really every voltage can be called VID no matter what CPU so I see no point of looking for an answer to correct naming the same thing that we all talking about. CPU sees every voltage as some VID. Who cares if it's .001V higher or lower ... on multimeter you have slightly different values anyway.

RGE · Jul 18, 2014

FRONTPAGE Intel i7 4790K Devil's Canyon CPU Review

Senior Member

Member

Gulper Nozzle Co-Owner

Member

Gulper Nozzle Co-Owner

Sim Racing Aficionado Co-Owner

Member

Attachments

Gulper Nozzle Co-Owner

Member

Gulper Nozzle Co-Owner

Senior Member

Member

Gulper Nozzle Co-Owner

Benching Team Leader

Senior Member

Benching Team Leader

Gulper Nozzle Co-Owner

Senior Member

Benching Team Leader

Senior Member

Similar threads