What's new
  • Please do not post any links until you have 3 posts as they will automatically be rejected to prevent SPAM. Many words are also blocked due to being used in SPAM Messages. Thanks!

Intel Haswell i7-4770K & i5-4670K Review Comment Thread

Dzzope

Well-known member
Joined
Oct 15, 2010
Messages
3,299
Location
Irishman in Kiev, wOOoo, I'm an alien...
What do you want to know? Load temp at a particular clock?
Pretty much, just like all the other review that have been done.. Basically if there is very little performance increase and no real oc ability increases and temperatures are the same then why spend the extra on haswell over Ivy?

Anyone else having issues seeing the graphs?

For this week's Richland review, I'll up the contrast but I need to know if the size is ok for everyone.
Graphs look fine to me, unless there was temperatures in there that I couldn't see :haha:
 

Eldonko

Well-known member
Joined
Apr 5, 2007
Messages
5,414
Location
Calgary, AB
From the 2 CPUs Ive tested temps and overclockability are very close to IB. Comparing a 3770 to a 4770
clock for clock the 3770 needed less vcore for 4.8ghz but the 4770 produced less heat allowing me to push to 4.9Ghz. There are other advantages and disadvantages however.. For example haswell can use adaptive voltages which is much better than manual in terms of power saving and heat on an oced machine. But on haswell it is more difficult to stabilize high cpu and mem clocks together than it is with IB.

I'll have a mobo review up pretty soon.
Pretty much, just like all the other review that have been done.. Basically if there is very little performance increase and no real oc ability increases and temperatures are the same then why spend the extra on haswell over Ivy?

Graphs look fine to me, unless there was temperatures in there that I couldn't see :haha:
 

10e

Well-known member
Joined
Jan 28, 2008
Messages
773
Location
Markham, Ontario
Nice review SKY. Very exhaustive and complete.

I bought a 4770K last week, apparently supposed to be a decent batch (L311B) according to the Overclock.net folks.

So far a gargantuan disappointment in overclocking.

With the 4770K, even a lowly 4.3Ghz is proving to be a challenge to balance voltage, stability and temps on a bigger Antec Kuhler H2O 920. The Antec is reporting fairly low water temps of 36 celsius, and yet the chip skyrockets to 100 celsius on IBT or LinX within 30 seconds of starting the test and then throttles down to 3.6 - 3.8 Ghz.

This above result is on both a Z87 TUF Gryphon and a cheap(ish) Z78M-Plus, both from Asus. The Z87M Plus is showing me voltage from Asus TurboV not CPU-Z so I believe it is as accurate as software can be, and it's showing me what the TUF Gryphon was showing me on CPU-Z in terms of VCore.

At 1.28 volts vCore the chip dies instantly/BSODs with Intel Burn Test and LinX (AVX binaries) and at 1.3 volts I can hold stability but the chip quickly throttles down.

And this has been with DDR3 set at 1333, 1400, and 1600 mhz speeds with 9-9-9-24 1T and 2T timings.

I obviously got a chip with a bad internal TIM application/interface but this is the WORST chip I've EVER seen since I've had my first I7 920 C0. I had an 7 920 D0 at 4.2 Ghz and my i7 970 at 4.2 ghz without much trouble.

This has been with lowered and raised cache speeds etc... I believe this chip will be a de-lid candidate judging from the above. I'm going to run IBT and LinX at stock just to see what the heck is going on and see where temps go.

By way of comparison, my 3770K was running at 4.3Ghz on stock voltage and 4.5Ghz with a slight vCore voltage bump or LLC setting of High, and less than 73 celsius on these settings using IntelBurnTest or LinX (AVX)

I've de-lidded this same 3770K and gotten it to 4.9Ghz at 68-70 celsius and 1.38 volts vCore on an old Antec Kuhler H2O 620. My i7 2700K is easily running at 4.8Ghz at relatively low 1.37 - 1.38 volts vCore.
 
Last edited:

Eldonko

Well-known member
Joined
Apr 5, 2007
Messages
5,414
Location
Calgary, AB
What kind of voltage you using, adaptive? What is your cpu voltage at 4.3?

Also for stability testing I recommend using aida64.
 

10e

Well-known member
Joined
Jan 28, 2008
Messages
773
Location
Markham, Ontario
What kind of voltage you using, adaptive? What is your cpu voltage at 4.3?

Also for stability testing I recommend using aida64.
I've tried 4.3 ghz at 1.2 volts where IBT 2.5.4 instantly stops working, and LinX crashes the PC or dies.

I've tried 4.3 ghz at 1.25 volts - same result.

I've tried 4.3 ghz at 1.3 volts - IBT runs and doesn't crash, but within one minute, when the test loads into RAM, it either dies or the chip bursts up to 99 celsius (as seen by CPU ID HW Monitor) and throttles down to 3.8 ghz.

Same with updated LinX using Intel's latest LinPack binaries.

Maybe it's a "power virus" issue (like furmark with video cards a few years back), so I'll try Aida64, but these temperatures are awful.

LLC was set to level 6 and this resulted in 1.328 volts according to CPU-Z and Asus Turbo V III for these runs, cache/Ring was manually set to 39, at one point I bumped the chip voltage to 2.0v from default of 1.80 volts, but the result was really no different.

I'll also try re-installing Windows and see if that works better. It's Win7 x64 with SP1, but it's an install carried over from an older 3770K so it may have something to do with it.

I'll also try Aida64 for future runs. Updates will go in the proper sub-forum, thanks for the help.
 

SKYMTL

HardwareCanuck Review Editor
Staff member
Joined
Feb 26, 2007
Messages
13,410
Location
Montreal
IMO, AIDA64 isn't great for stability testing, especially since you have to pay for a license.

I'd recommend the usual: Prime95. 1 hour and you'll be good to go.
 

Eldonko

Well-known member
Joined
Apr 5, 2007
Messages
5,414
Location
Calgary, AB
What kind of voltage are you using though? 1.2v adaptive / offset is a hell of a lot different than 1.2 manual. With adaptive you will get a lot more under load. Grab cpuz 1.64.3 (CPU-z) to check load voltage or the latest beta of aida64 will be accurate as well as it is optimized for haswell. You dont have to pay, the trial version does the trick and the stability test works fine. Grab version 3.00.2505 here - Downloads | AIDA64

The reason why I suggest aida64 is it updated constantly and includes full instruction sets for haswell.
I've tried 4.3 ghz at 1.2 volts where IBT 2.5.4 instantly stops working, and LinX crashes the PC or dies.

I've tried 4.3 ghz at 1.25 volts - same result.

I've tried 4.3 ghz at 1.3 volts - IBT runs and doesn't crash, but within one minute, when the test loads into RAM, it either dies or the chip bursts up to 99 celsius (as seen by CPU ID HW Monitor) and throttles down to 3.8 ghz.

Same with updated LinX using Intel's latest LinPack binaries.

Maybe it's a "power virus" issue (like furmark with video cards a few years back), so I'll try Aida64, but these temperatures are awful.

LLC was set to level 6 and this resulted in 1.328 volts according to CPU-Z and Asus Turbo V III for these runs, cache/Ring was manually set to 39, at one point I bumped the chip voltage to 2.0v from default of 1.80 volts, but the result was really no different.

I'll also try re-installing Windows and see if that works better. It's Win7 x64 with SP1, but it's an install carried over from an older 3770K so it may have something to do with it.

I'll also try Aida64 for future runs. Updates will go in the proper sub-forum, thanks for the help.
 

Squeetard

"Quote This..."
Joined
Nov 15, 2007
Messages
3,984
Location
Hell
IMO, AIDA64 isn't great for stability testing, especially since you have to pay for a license.

I'd recommend the usual: Prime95. 1 hour and you'll be good to go.
2nd Prime 95. I've found that if I can get through 2 iterations of the torture test without an error, my system is dead stable. No need to run it for hours.
 

10e

Well-known member
Joined
Jan 28, 2008
Messages
773
Location
Markham, Ontario
What kind of voltage are you using though? 1.2v adaptive / offset is a hell of a lot different than 1.2 manual. With adaptive you will get a lot more under load. Grab cpuz 1.64.3 (CPU-z) to check load voltage or the latest beta of aida64 will be accurate as well as it is optimized for haswell. You dont have to pay, the trial version does the trick and the stability test works fine. Grab version 3.00.2505 here - Downloads | AIDA64

The reason why I suggest aida64 is it updated constantly and includes full instruction sets for haswell.

I always start with manual voltage. When I reach stability then I test with adaptive or offset carefully to understand how the VID is changing at certain speeds/loads.


Since I am wholeheartedly unhappy with my OC right now, I'm not at the offset/adaptive voltage stage yet.


Aida64 CPUiD says that even with Ultra-High or Extreme LLC I'm getting within .05 volts of what I've set manually. So 1.310 in the BIOS is giving between 1.312 - 1.315 according to Aida64 CPUiD.

CPU-Z is accurate on the Asus TUF Gryphon but not the Z87M-Plus so I needed Aida64 anyways.


I de-lidded the thing last night and put CoolLaboratories Liquid Ultra Pro between the die and IHS, and Vaseline over the contacts and resistors on the chip's PCB to ensure no shorting out from excess CLU. The chip runs fine, and no negative effects except, and about a 7 celsius package temperature reduction.


The AVX tests on IBT and LinX take another 20 seconds to reach 100 celsius and start throttling after the de-lid.

This is at 4.4Ghz @ 1.29 volts vCore (verified in Aida64 CPUiD)

This is my second de-lid, BTW. My i7 3770K runs all day at 4.9Ghz with a lower end Kuhler H2O 620 at less than 80 celsius on these tests at 1.38 volts vCore. 5.0Ghz at 1.43 volts VCore was fine too, just too much voltage for little performance increase.

At this point, I'm leaning towards having a dud of a 4770K. It's [email protected] right now with stability after eight hours, but it's using 155 watts of power (with no video card) for 25K PPD. My kill-a-watt has shown it go up to 220 watts power under the IBT and AVX tests. Something is up.

I also tried using 125 mhz bClk Strap and setting memory accordingly at a lower speed to see better stability, with little to no change.

The newest versions of AIDA64 run out in 30 days. :(
Yes, and not worth the license price IMHO. My OCing efforts right now look like they'll exceed this period.:doh:

2nd Prime 95. I've found that if I can get through 2 iterations of the torture test without an error, my system is dead stable. No need to run it for hours.
I found that with a proper set-up this is true, except for the older i7 8xx series which always seemed to die after 8 hours, likely due to too-high LLC and vid drift. So now I'll use less LLC and higher offset to ensure this doesn't happen at a slight cost of power consumption and general temps.

Following that practice I 100% agree.
 

Latest posts

Twitter

Top