Review Index:
Feedback

The Intel Core i7-4790K - Devil's Canyon Review and Overclocking

Author: Ryan Shrout
Subject: Processors
Manufacturer: Intel

Overclocking Experiences and Temperatures

Impressive out of the box performance aside, the other big mystery is how much the new thermal interface and capacitors on Devil's Canyon would change the way Haswell overclocks. I can say, based on my experience with this chip and after talks with motherboard vendors that have used quite a few more engineering samples of the Core i7-4790K, the answer is likely not very much.

View Full Size

Temperatures during load of Core i7-4770K at stock speeds

View Full Size

Temperatures during load of Core i7-4790K (DC) at stock speeds

Both processors were run on the ASUS Z97-Deluxe motherboard with a Corsair H80 self-contained water cooler.

Take a look at the two temperature readings above. Operating at 3.5-3.7 GHz, the 4770K was measuring about 53-55C. While that isn't a bad temperature range, the Core i7-4790K, running at 4.0 GHz, was only measuring at 50-51C. Seeing a couple of degrees improvement in load temperature with an increased clock speed is a welcome change and can be attributed to the improvements Intel designed into Devil's Canyon.

Let's see how the processor overclocks and what that might change. 

View Full Size

The overclocking process with the Core i7-4790K is identical to that of other Haswell parts.  The theory has been that, thanks to these improved thermals, we should see improved scaling. That did not happen for me. My peak overclock for all cores was 4.7 GHz running at 1.36v and no amount of added voltage or changes to the overcurrent protection or cache voltages, seemed to matter. I could not reach 4.8 GHz for a standard stability run. 

View Full Size

To be clear, even for a standard Haswell processor, this result is pretty average. Setting your core voltage to 1.36v is pretty damn high and I've had much better luck with our single Core i7-4770K, as you'll see below. By itself, this does not mean that Devil's Canyon is a worse overclocker, or that it isn't a better overclocker, just that the single sample we were provided didn't show the improvement

On this same motherboard, on this same day, I was able to take a Core i7-4770K ES processor and hit 4.8 GHz stable with a voltage of just 1.25v. Clearly THAT result is well above average and could be considered one of the better Haswell parts I have seen. So how did the temperatures compare in our best overclocks with the two CPUs?

View Full Size

Temperatures during load of Core i7-4770K at 4.8 GHz

View Full Size

Temperatures during load of Core i7-4790K at 4.7 GHz

Interesting stuff. Even though I wasn't able to get higher than 4.7 GHz, the limit does not appear to be temperature on my Core i7-4790K, just crappy luck. Under load with a voltage of 1.36v, temperatures on Devil's Canyon didn't go past 84C. On the other hand, the original Haswell part hit 91C with a lower voltage (1.25v) and higher clock speed (4.8 GHz). 

With just a single CPU sample in my hand its hard to make any definitive statement, but my guess is that the changes Intel made on the Core i7-4790K did in fact have a positive effect on the temperature of the core. It just so happens that our sample from Intel didn't perform as highly as we might have hoped, purely by chance. I am hoping to pick up another 4790K to do some more testing but even then, with a sample size of two, we will be depending on the community to really make the final decision on the benefits of Devil's Canyon.

June 7, 2014 | 01:35 AM - Posted by Anonymous (not verified)

"It's not "budget" but most people's standards but getting this kind of performance with a $339 CPU helps everyone and Intel's good will gesture to the community at least indicates that the lack of competition on the high-end of the market isn't totally damning us all."

There is something wrong with this sentence.

June 7, 2014 | 10:07 AM - Posted by Ryan Shrout

Fixed the typo, thanks!

June 7, 2014 | 01:45 AM - Posted by JxcelDolghmQ (not verified)

Thanks for the review. Decent pricing, for an Intel part, but they do not seem to be binning these chips for overclocking like some people were hoping or expecting. Funny to see the 8350 still topping the performance per dollar charts despite its age and lack of price drops. Really demonstrates the lack of progress in the higher end of late.

Couple of potential typos:
On the Test Setup page, you list the A10-5800K but it does not appear in any of the benchmarks.
In the conclusion, second to last paragraph, you write: "It's not "budget" but most people's standards", I think you meant "by most people's standards".

June 10, 2014 | 11:14 AM - Posted by Patrick P (not verified)

Eh, the performance for dollar rankings have never included the average power usage in the calculation. In reality 4690 > 8350 for perf/$ in the long run.

June 7, 2014 | 02:22 AM - Posted by renz (not verified)

so clock for clock how much improvement we are going to see from sandy bridge to this new line up?

June 7, 2014 | 02:45 AM - Posted by JxcelDolghmQ (not verified)

Same as Sandy to Haswell, IPC is not any better on these than regular Haswell.

June 7, 2014 | 10:08 AM - Posted by Ryan Shrout

Correct, it's the exact same architecture, just higher clocks (than we have seen before) and new TIM and added caps.

July 26, 2014 | 04:39 AM - Posted by Carolyn (not verified)

Hi there, i read your blog from time to time and i own a similar one
and i was just wondering if you get a lot of spam responses?
If so how do you prevent it, any plugin or anything you can advise?
I get so much lately it's driving me mad so any assistance is very much
appreciated.

Review my web blog ... sms blast singapore - Carolyn,

June 7, 2014 | 03:17 AM - Posted by Havor (not verified)

/Offtopic

Why do you use the X series when comparing price performance ratio's of S2011 CPUs, hell i would not even recommence a S2011 CPU for 95% of the people that buy them.

And i know my 3930K + R4E dose not have the best price performance ratio, but using 3970X and 4960X CPUs that no one in his right mind would buy over the x930K series is just muddying the waters.

June 7, 2014 | 10:09 AM - Posted by Ryan Shrout

Well, to be honest, these are the parts we have in-house for testing.

But the comment is noted and we'll try to include more of the product range when doing future reviews. Maybe for Haswell-E, for example.

June 7, 2014 | 04:16 AM - Posted by Martin Trautvetter

I really like that Intel went with a much higher default clock on this one, not just a 100 MHz bump, and agree with your assessment that that makes the 4790K a great choice even if you never plan to overclock. 400 MHz extra for just $25 over the regular 4790? Who wouldn't buy the K?

Interesting results, though the (sometimes) stilted scaling and conspicuously higher idle power consumption look a bit odd. Hope the latter will be fixed by newer UEFIs, and is not an inherent trait of DC Haswells!

I wouldn't worry about the pricing, the 4770K will be sold off, with the 4790K replacing it at the same price soon.

Personally, I can't wait to see when (if?) Gigabyte updates my Z87's UEFI to include DC support.

June 7, 2014 | 04:51 AM - Posted by arbiter

if they don't then its gigabyte that decided to, I have an Asus Z87 board and they have bios for DC cpu out.

June 7, 2014 | 08:27 PM - Posted by John H (not verified)

The K parts miss some of the virtualization features the non-Ks have..

June 8, 2014 | 07:29 AM - Posted by Martin Trautvetter

While that was true for the old Ks, VT-d and TSX-NI are enabled on the new DC Ks:

http://ark.intel.com/compare/80807,80806,75123

_
BTW: How cool is it that the ARK-references for the i7 Haswell refreshes could be read as "eighty-eighty-six" and "eighty-eighty-seven"? /geek-out ;)

June 8, 2014 | 04:16 PM - Posted by Anonymous (not verified)

Wow that's good to know.

June 9, 2014 | 07:03 PM - Posted by John H (not verified)

That is good to know! I want a i666 inside my system now!

June 7, 2014 | 05:54 AM - Posted by Anonymous (not verified)

Ryan,

What would your temps on the 4770k have been had you put it at 4.7GHz (the same as the 4790k) and reduced the voltage to the bare minimum to sustain that frequency?

If the temperatures in that case would be in the 70's or 80's, which is warm but still acceptable, I really don't see the point in this product from an overclocking prospective.

June 7, 2014 | 10:10 AM - Posted by Ryan Shrout

I would agree with that statement if my overclocking results were to be the definitive results, but it is possible that the 4770K that I got happens to be a GREAT overclocking part and the 4790K I got happens to be a poor overclocking part. There is still a lot of luck in this business.

June 7, 2014 | 03:39 PM - Posted by Anonymous (not verified)

The point is that they cost the same, but the 4790k has better thermal properties, which is huge for overclocking. Even if it can't clock higher, it can reach the same speeds with much more stability without requiring a risky delid.

June 7, 2014 | 07:48 AM - Posted by CB (not verified)

Ryan, any chance you could tell us what the fastest stable OC you got was without a voltage bump? It'd be interesting to see that OC temp without voltage increases...

June 7, 2014 | 10:11 AM - Posted by Ryan Shrout

The only data I have on that is at 1.2v, which is where I started overclocking testing. On that, I could hit 4.5 GHz, but no more.

June 7, 2014 | 10:47 AM - Posted by Robogeoff (not verified)

"Temperatures during load of Core i7-4970K at 4.7 GHz"

4790K or 4970K?

I would think it weird for this Haswell refresh to get a model number that's higher than my Ivy-E 4930K.

June 7, 2014 | 01:52 PM - Posted by Morry Teitelman

Fixed, thanks for pointing that out...

June 7, 2014 | 10:52 AM - Posted by polu (not verified)

Can you pls cut the IHS and see if there is a weak TIM or is it soldered as SB, pls?

June 9, 2014 | 04:40 AM - Posted by Earnest Bunbury (not verified)

YES, er... YES!

June 7, 2014 | 11:36 AM - Posted by Anonymous (not verified)

Could you run tests using a z87 motherboard also? Would really like to see if there are any limitations on using the new chips with a z87 board.

June 7, 2014 | 11:51 AM - Posted by geekthem (not verified)

http://geekthem.com/future-of-intel-based-smartphone-chip/

June 7, 2014 | 12:04 PM - Posted by Anonymous (not verified)

Why am I not seeing 8core cpu for decent price yet.

fuk this shit, fuk intel, goto devil canyon hell.

June 7, 2014 | 04:40 PM - Posted by Anonymous (not verified)

Hopefully they skip Broadwell on the PC. Can't see them releasing Broadwell this year if they're dropping a Haswell "Refresh" this late. Not to mention it's going to be a disappointment for desktops anyways.

Can't even get excited for the Haswell 8 core part being that it's only for the $1,000 model. My fricken $500 i7 3930K is an 8 core part with 2 cores disabled. No reason why they shouldn't offer an 8 core part for the $500 range either. Intel being Intel.

June 8, 2014 | 12:49 AM - Posted by Humanitarian

Lel, "[Intel]8core cpu for decent price"

Are you high?

June 7, 2014 | 12:47 PM - Posted by Anonymous (not verified)

Ryan,
Could you maybe make a test with an overclocked Sandy bridge chip versus an overclocked Devil canyon?

June 7, 2014 | 04:42 PM - Posted by Anonymous (not verified)

Why it doesn't matter this has already been said the architecture is the exact same with only overclocks being different.

4Ghz SB = 100%
4Ghz IVB = +5% over SB
4Ghz HSW = +5% over IVB

June 7, 2014 | 05:50 PM - Posted by Keith Jones (not verified)

There seems to be a big difference between the cores on the Core i7-4790K 84c on core 1 and 58c on core 4 that seem a bit strange i know there is normally a difference but 26c seems way to much

June 7, 2014 | 10:11 PM - Posted by Dazed-n-Confused (not verified)

Anyway to know how to buy chips that have a good yield? Bought two 3770k's at the same time, and I think they are from a poor yield. Can't get either one past 4.4ghz with high-end water cooling.

June 8, 2014 | 12:55 AM - Posted by Humanitarian

step one: live in a country with good distance selling regulations

Step two: buy CPU from vendor 1

step three: put it in a system, if it sucks send it back

Repeat step two and 3(vendor+1) until you get the desired OC

June 8, 2014 | 02:38 AM - Posted by Anonymous (not verified)

How would you like it if you sold something new-in-box and then had a buyer return it used and now worth $50 less? It's a dick move and leads to retailers setting super-strict return policies and restocking fees and other shit that hurts people who want to do a legitimate return.

June 8, 2014 | 09:38 AM - Posted by Martin Trautvetter

You could always do the nice thing and order tray versions, no box to rip open, no apparent damage to the shop.

Of course, chances are tray versions were already picked clean by the nice people at said shop. ;)

June 8, 2014 | 05:08 PM - Posted by Humanitarian

>$50 less

What planet are you on? $10 less is more accurate.

And if it's a legitimate return, like damage or wrong item, it has no effect on the returns process. You've been able to return items you don't want for decades.

Please, read back your incoherent babbling before you post.

June 8, 2014 | 12:23 PM - Posted by Atexy (not verified)

Hence why Germany is planning to modify its Distance Selling Directive based off a new EU Directive - because of shitty people like you abusing vendors like that.

June 8, 2014 | 05:12 PM - Posted by Humanitarian

>because of shitty people like you abusing vendors like that.

wow, top retard. Go read the proposed legislation, it's not even in force now so how could I be abusing vendors?

June 12, 2014 | 06:19 PM - Posted by Atexy (not verified)

It is from today on, way to be uninformed scum.

February 24, 2015 | 10:53 AM - Posted by scotty (not verified)

Great advice. It seems obvious, what with the silicone lottery, but I didnt think of it, and have not read a similar comment.

June 8, 2014 | 10:32 AM - Posted by Mike S. (not verified)

What BIOS did you use for these tests? ASUS recently (a few days ago) updated the BIOS for it's Z97 line to aid in Devil's Canyon compatibility and overclocking. If you were using the stock BIOS that might make a difference.

June 8, 2014 | 11:11 AM - Posted by Anonymous (not verified)

Thanks for the review Ryan.
It's a bit underwhelming to see it only go to 4.7 GHz. I just hope that your sample is a dud and doesn't resemble the average OC of those chips. But as I've seen in several forums those CPUs don't have much headroom left on them.

June 8, 2014 | 04:18 PM - Posted by Anonymous (not verified)

Do you think that DC chip that did 5.5GHz on air at Computex was a cherry-picked part?

June 12, 2014 | 10:51 AM - Posted by KngtRider

They used LN2 chilled Air, ie they 'cheated'.

It's still air coolng, just how cold the air is re-defined...

LN2 Pot -> Vapour -> Liquid Cooler Raidator or High end Air Cooler

Not many media who were not there physically paid attention to what was going on.

Going by what was said on the video stream of the event, supposedly Intel set the rules.

June 8, 2014 | 05:46 PM - Posted by Damir (not verified)

Hay everebody its worth to uprgrade from i7 3770k to i7 4790k ?

June 8, 2014 | 10:02 PM - Posted by Anonymous (not verified)

No

June 8, 2014 | 08:33 PM - Posted by Anonymous (not verified)

Not so Impressive Temperatures

for a "next gen" TIM..

June 10, 2014 | 11:22 AM - Posted by Patrick P (not verified)

At 1.33 volts that's pretty good for just an engineering sample.

June 10, 2014 | 11:23 AM - Posted by Patrick P (not verified)

At 1.33 volts that's pretty good for just an engineering sample.

June 10, 2014 | 11:24 AM - Posted by Patrick P (not verified)

At 1.33 volts those are good temps for an engineering sample. Until mass-production of graphene is possible, TIMs will not be getting much better. Quantum mechanics and entropy dispersion have their limits.

June 9, 2014 | 04:24 AM - Posted by Joe (not verified)

Always get to get 2 different chips from 2 different countries. Send the 1 back the doesn't OC as good. I expect an easy 5ghz on all 4 cores full time oc turbo disabled.

If broadwell's version of devils canyon can hit a good 5.4ghz or so I may even give up on x99 and build a badazz mini itx gaming rig in the ncase m1 with a 45mm double thick 120mm rad in push pull for the cpu and a 45mm double thick 120mm radiator in push pull for the GPU (Preferable the GTX 880 or 980 maxwell flagship) or maybe just a triple slot version since the ncase m1 can handle triple slot gpu's even tho it's a mini itx case. And with maxwell being so incredible power efficient a triple slot gpu in that architecture would have the fans running so slow from that massive of a heatsink it would be so quiet. Like tripling the amount of heat the gpu can dissipate with what it needs to dissipate ='s very low rpm quiet gpu fansn

June 9, 2014 | 10:13 AM - Posted by Zealotki11er (not verified)

Stock speeds really dont matter much in a K part because 100% of the time it will be OCed or else it would not be a K part. If Intel wanted to push performance to average person they could have made 4790 non K @ 4GHz also. The way i see it these K CPUs are getting less and less impressive. 4.7GHz OC even if not a amazing one still shows how far this CPU can go. 200-300Mhz more will not make it a amazing overclocker unless 90% of the CPUs can achieve it. Feels like Intel is killing overclocking with this CPU. I mean a 300MHz OC, why even bother. Why spend money in high end MB, cooling for this CPU.

June 9, 2014 | 10:24 AM - Posted by ancientscream (not verified)

It's about time they addressed the die to heat spreader issue post sandy bridge is this Tim better than soldered ? Energy efficiency are always good, but that better be a bad binned part compared to average were looking for a cool <>5ghz, and the fact there's been transistor count or logic improvement on the cpu side excluding energy efficiency and igpu for so long, makes one worry they lost the skill to advance it, and they're for core count increase to resolve the power advancement issue for them, Moore law since sandy bridge has been skewed.

June 9, 2014 | 10:37 AM - Posted by ancientscream (not verified)

Devils canyon: It's about time they addressed the die to heat spreader issue post sandy bridge is this Tim better than soldered ? Energy efficiency is always good, but that better be a bad binned part compared to the average, were looking for a cool <>5ghz, and the fact there's been no transistor count or logic improvement on the cpu side excluding energy efficiency and igpu for so long, makes one worry theyve lost the skill to advance the xx design, and they're waiting for core count increase to resolve the power advancement issue for them or the need to enhance the design, Moores law since sandy bridge has been slooowwed. And the use of the stream units in the igpu is dependant on people coding for it ? Whereas plain increases in x86 could from logic enhancement that is largely static at the moment. Corrected version

June 9, 2014 | 01:27 PM - Posted by jkreykes

Thanks for the review, I have mine on order from micro center.

June 9, 2014 | 08:31 PM - Posted by teddy2003 (not verified)

where's the gaming benchmark we love to see?

June 10, 2014 | 01:28 AM - Posted by AK (not verified)

"Performance of the Core i7-4790K is impressive, even if you choose to not overclock at all. The 500 MHz clock speed increase (at both base and Turbo speeds) when compared to the Core i7-4790K makes the choice between the two parts pretty much a no-brainer. "

I think you meant "when compared to the Core i7-4770K".

June 10, 2014 | 01:49 PM - Posted by Martin Trautvetter

I think the proper comparison should be the 4790 (non-K). For an extra $38 (list) you get a guaranteed 400 MHz boost (that's almost linear scaling!) and the unlocked multi on top.

Best value in a long time, IMHO.

June 10, 2014 | 11:18 AM - Posted by Patrick P (not verified)

If you still have the engineering sample, do you know if the lid is soldered on or not? I want to know if this can be de-lidded asap.

June 10, 2014 | 11:27 AM - Posted by Anonymous (not verified)

A big Thank You for including a 2600k in your review. Mine is still in service and it looks like it'll stay at least until Skylake. It's tough to find comparisons between old and new so it's great you included it.

Any take on why only Core #1 got really hot during overclocking?

June 12, 2014 | 04:05 PM - Posted by Crazycanukk

ok..maybe someone could help me out understanding something.

currently running 2 pcs ..1 gaming with an i5 3570k and gtx 780 and 1 streaming box for twitch with an i7 970. I want to give my 1 i7970 system to a friend who is disabled and loves to game but his system is dying. I want to re-purpose my i53570k for an Htpc.

Looking to build a system that i can play Swtor, ESO, BF4 and be able to Twitch Stream at 720p 60fps smoothly or 900p at 45 fps using Fast settings.(no worries on upload speeds..i have that covered at 15mb up)

looking at the Haswell 4790k x264 second pass it does VERY well against the 4960x 6 core. I was denating on going Haswell e 6 core at release but maybe i wont have to..

would the 4790k do what i want and give me good in game performance with my gtx780 + 27" 2560 x 1440 monitor) AND stream to twitch at 720p 60 fps or 900p 45 fps at fast encoding settings? (i use xsplit and obs..havent really settled on 1 yet)

June 13, 2014 | 01:10 PM - Posted by Anonymous (not verified)

well one thing, the reason why you where limited on the overclock is that you used asus motherboard, I've never had any asus board compete with a msi gaming or a power version of any of their motherboards, and that isn't using the auto overclock, try better componets when trying to overclock

October 6, 2014 | 11:33 AM - Posted by Marc Couture (not verified)

I have just purchased this CPU and currently shopping around for an aftermarket cooler (as Intel's HSFs are crap). I've run across a few forums where bad CPU thermal throttling occurs, including this one with nice details on testing procedures:

https://communities.intel.com/thread/54032?start=45&tstart=0

Ryan, have you heard about this? Thanks.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.