Intel Devil's Canyon: The Good, the Bad, the Ugly @ [H]

dumping x86 for a new isa is not in the cards, too much would have to change unless they offered the new chips for free, and bankrolled the software for it

then again arm might be viable...
 
It's in the cards, they're just putting it off as long as possible. While they can switch to carbon rather than silicon for the fab process, they're still going to run out of miniaturization room for X86 by 2020 on the current roadmap.

I understand why they're stringing this ancient trash along, I just hope the alternative has legitimate performance gains for the first time in decades (by the time we actually get it).
 
Azaran said:
There are some who would say that Intel hit the wall with Sandy, and has been spinning their tires ever since. When a 920, a 6 going-on-seven year old CPU, is still competitive for most tasks, that's a sign that something has gone very wrong in the engineering department.

Increasing tick-tock delays (The roadmap still had 10nm in Q1 2015 as recently as 2011), and stagnant performance increases makes me wonder how long before Intel just stops trying like AMD, or finally dumps the architecture they've been miniaturizing for 20+ years and releases something new.
Click to expand...

920 is nice and all but it is only competitive in some areas. Sandy was a big increase to the architecture for various reasons, some of the biggest was power related.

I'm guessing you are thinking they are on the same architecture for 20+ years now, which isn't true. Intel knows there is a massive drop off of software reliance on the CPU. For a mass majority of the users of these consumer based products their new stuff is insanely over kill as a C2D would handle almost all tasks without taking much of a sweat. The Prosumer market is seeing some need for additional computing but they are getting help from GPGPU, CUDA and OpenCL.

Intel has been refining a product it knows it doesn't need to advance. They have been dumping all their R&D into mobile products and doing quite a good job. My hope is broadwell is a new architecture that brings something fancy to the table.
 
Trimlock said:
920 is nice and all but it is only competitive in some areas. Sandy was a big increase to the architecture for various reasons, some of the biggest was power related.

I'm guessing you are thinking they are on the same architecture for 20+ years now, which isn't true. Intel knows there is a massive drop off of software reliance on the CPU. For a mass majority of the users of these consumer based products their new stuff is insanely over kill as a C2D would handle almost all tasks without taking much of a sweat. The Prosumer market is seeing some need for additional computing but they are getting help from GPGPU, CUDA and OpenCL.

Intel has been refining a product it knows it doesn't need to advance. They have been dumping all their R&D into mobile products and doing quite a good job. My hope is broadwell is a new architecture that brings something fancy to the table.
Click to expand...
I thought everyone following cpus knew that Broadwell is basically just a shrink of Haswell.
 
Azaran said:
It's in the cards, they're just putting it off as long as possible. While they can switch to carbon rather than silicon for the fab process, they're still going to run out of miniaturization room for X86 by 2020 on the current roadmap.

I understand why they're stringing this ancient trash along, I just hope the alternative has legitimate performance gains for the first time in decades (by the time we actually get it).
Click to expand...

Why would dumping x86 be necessary or beneficial? The CPU itself doesn't run x86 instructions anyways, it just decodes them to uops because CISC instructions aren't very amenable to running natively on a CPU. That's why they just add on more instructions all the time so x86 has a zillion instructions, as long as the decoding hardware doesn't take up much die real estate (it doesn't) I don't see why there would be a need to switch to something new.
 
Thanks for the review on the cpu. Guess money will stay in the bank account till intel actually releases something worth while to upgrade to. Tired of getting nickled and dimed for small upgraded that in reality don't do squat, for us folks.
 
Well, preordered me a 4790K for my new build. I've been using a 870 for almost 4 years now. Going to be a decent upgrade as I can't manage to OC my 870 much.
 
im not gonna bother upgrading till i can double performance....its only been 6 years lol...why should i bother with a side grade?
 
Given that I'm moving from an old Lenovo box with a dual core Athlon, I'm really looking forward to this build. Would be Like ditching a Mini for a pagani zonda. Can't wait to get it...
 
Insula Gilliganis said:
Kyle above talked about the possibility of getting some "retail" 4770Ks to test but I couldn't find any evidence he posted any tests on these and seemed to still using his original 4770K ES until his mentioning of the "electromigration" problem in the July 30, 2013 Gigabyte Z87X-UD3H review.
Click to expand...


Just got my first retail part today. And yes, by "retail" I mean we purchase those from retailers.

I do think I have one retail 4770K here now we are having electromigration issues with.

http://hardocp.com/news/2014/06/25/intel_devils_canyon_core_i74790k_in_hand
 
xorbe said:
What's the symptom? Falling clock speed? Requires pegged to static frequency / dynamic freq locks up?
Click to expand...


Starting to fail on long term stability tests at our stock OC settings.
 
Kyle, my previous post on pg 15 discussed that when the 4770K was first previewed by you and other websites, there was hardly any mention that a "Engineering Sample" (ES) was being used or that retail 4770Ks would behave any different than those tested ES. And once the retail 4770K was available, there was no rush to test to see if it behaved any differently than ES chips.

But a year later, just about every website who tested the 4790K emphasized, or at least made known, the fact that an engineering sample was being tested. And now this rush to test the retail 4790K, which didn't occur with the 4770K. I can't believe all this was done independently without at least some strong "hinting" from someone (perhaps the person(s) giving out the ES) that the retail version would behave slightly/significantly different than the ES version. Kyle, can you shed any light on this?

Originally Posted by xorbe
What's the symptom? Falling clock speed? Requires pegged to static frequency / dynamic freq locks up?
Click to expand...

To elaborate on what Kyle already mentioned 2 posts up..

"The Intel Core i-7 4770K that we have been using for testing over the last couple of months. It seems that electromigration has damaged our CPU, at least in an overclocking sense when stressed for long periods of time. It seems that it has also been damaged for high GHz (4.8) for short periods of time. That all said, we had the same results with this Gigabyte motherboard that we have had with the two previous high end motherboards we have reviewed in that these would not hold a 48 hour 4.5GHz overclock. Since this review we have pulled our two month "old" testing 4770K and replaced it, and are now back on track with our stress testing." http://lucca.hardforum.com/rewrite/?url=http%3A%2F%2Fwww.hardocp.com%2Farticle%2F2013%2F07%2F30%2Fgigabyte_z87xud3h_lga_1150_motherboard_review%2F7&id=1&match=1&source=none&destination=none

"We no longer run any testing at default CPU speeds, and we have recently purchased a new Core i7-4770K to test with as our original engineering sample CPU from Intel seemed to be affected by some of the long term torture testing we do as it would not longer hold a 4.5GHz overclock for any considerable amount of time." http://lucca.hardforum.com/rewrite/?url=http%3A%2F%2Fwww.hardocp.com%2Farticle%2F2013%2F09%2F05%2Fgigabyte_z87xud4h_lga_1150_motherboard_review%2F7&id=1&match=1&source=none&destination=none
 
Insula Gilliganis said:
Kyle, my previous post on pg 15 discussed that when the 4770K was first previewed by you and other websites, there was hardly any mention that a "Engineering Sample" (ES) was being used or that retail 4770Ks would behave any different than those tested ES. And once the retail 4770K was available, there was no rush to test to see if it behaved any differently than ES chips.

But a year later, just about every website who tested the 4790K emphasized, or at least made known, the fact that an engineering sample was being tested. And now this rush to test the retail 4790K, which didn't occur with the 4770K. I can't believe all this was done independently without at least some strong "hinting" from someone (perhaps the person(s) giving out the ES) that the retail version would behave slightly/significantly different than the ES version. Kyle, can you shed any light on this?
Click to expand...

To give you the short answer. When a new architecture CPU is released, we do expect to see chips later in it life behave differently than the ES samples. That is why we purchase retail CPUs to test with. I have purchased 5 or 6 4770K since introduction, and honestly I have not seen that much variability from those. All mine would do 4.5/[email protected] out of the box on water cooling, as did the ES sample. Overclocking further beyond that has not shown us any real variances other than seeing people get a "lucky" processor that was good at overclocking, which is par for the course.

In terms of Devil's Canyon, we are not seeing a new architecture release, but a re-release of Haswell specifically for enthusiasts. That said, for the "regular guy" that just throws one in his box at stock speeds is getting a nice CPU when you compare to a stock clocked 4770K. So what is DC then actually beyond what Intel has told us about the package itself (ie TIM and power supply)? Dc is nothing more than culled "4770K" dies that have been set aside to be 4790K processors. I have it on good information that Intel has been sorting for 4790K since March of this year to find the best silicon. So if Intel has been very tight on sorting, it would leave me to believe it knows "golden" processors and sent those to people for review that they thought those would be best with for publicity. So for those of us in the media, and I will speak only for myself here, I was very wary as to the ES 4790K processors possibly being very different from the retail 4790K processors. I know this is why you saw my emphasis on this, and likely with other real reviewers as well.

I hope this answers your question. If it has not, please be specific to what you are asking. Thanks.
 
They may have been sorting them for even longer than that, I've read of guys that are receiving 4790K's with 2013 manufacture dates. (assuming they follow the historical batch number assignments)

My "new" 4790K chip from Newegg was L352C119 and it's crap. :)
 
=Gunslinger= said:
They may have been sorting them for even longer than that, I've read of guys that are receiving 4790K's with 2013 manufacture dates. (assuming they follow the historical batch number assignments)

My "new" 4790K chip from Newegg was L352C119 and it's crap. :)
Click to expand...


Extremely possible. When looking into this in the past month March was all I was able to verify. But if folks have "2013" CPUs....well, there is your solid answer. :)
 
THANKS Kyle for your response and for upholding the "overclockers" portion of [H]OCP by punishing your retail 4790K so we don't have to.. unless we want to.

picard_clapping.gif


I wouldn't have cared squat about the 4790K EXCEPT for Intel's 5.0GHz claims. Something must have made Intel believe that changing the TIM and adding some capacitors was going to give the 4790K an extra 200+MHz over a 4770K.

Now time to stop thinking about old tech and start hoping that Haswell-E, X99 and DDR4 all come out sooner than later.
 
Insula Gilliganis said:
THANKS Kyle for your response and for upholding the "overclockers" portion of [H]OCP by punishing your retail 4790K so we don't have to.. unless we want to.

I wouldn't have cared squat about the 4790K EXCEPT for Intel's 5.0GHz claims. Something must have made Intel believe that changing the TIM and adding some capacitors was going to give the 4790K an extra 200+MHz over a 4770K.

Now time to stop thinking about old tech and start hoping that Haswell-E, X99 and DDR4 all come out sooner than later.
Click to expand...


Thanks for the kind words.

I still have not heard what shaked out at Intel over this whole fiasco, but obviously some folks in engineering where thinking the 5GHz wall was going to easily be splashed with a bit of cooling and power. Guess mean old Mr. Physics got in the way of their electrons.
 
GeneO said:
Wow, degradation at 1.28v. Many people are claiming 1.3v+ 24x7 is "safe". YMMV I guess.

As for the DC culling and DC lots from 2013, now I know why my late 2013 4770K is a poor performer (4.3 GHz wall) - the odds were stacked against me by Intel's culling.
Click to expand...

Well, let's keep in mind that we have used this CPU since it first retail availability for long term torture testing for a long time now. It is not unusual for it to run a solid week at full load on a motherboard with no fans on it.
 
Sounds like temperature may have contributed a lot to the electromagnetic damage then.
Thanks again
 
gigatexal said:
not to nitpick, but is that a good representation of what a desktop workstation environment would be? Say a render node running full tilt for a week, it'd likely have plenty of fans.

What were the temps during these torture tests?
Click to expand...


Not unless you run your workstation processors overclocked and overvolted.

And keep in mind that we water cool the CPU during this, as these are motherboard review torture tests we are running.

I usually see per core temps in the mid-70s to mid-80s during these tests.
 
Wow. Just came from a I7 920 and the temps are very scary here.

I put my 4790K with a Corsair H60 and at load it quickly goes all the way up to 100C. Like, very quickly. Running Cinebench it fluctuates between 90-100C pretty quickly. Obviously I could have put the heat sink on badly but it seems solid, and ive certainly put many on before.

Very uncomfortable.
 
More uninteresting processor from Intel. Nothing new to see here, move along.
 
I received my 4790K (batch #: L418C169) from Newegg today and, for now, have just bumped the cache to 4.2GHz and all cores to 4.4GHz using just the MB Auto voltages for both (currently 1.182v for the cores and 1.223v for the cache). It is running at about 60C while crunching for BOINC on all 8 threads right now, using my H100i (ambient ~74F).

So far, I'm happy with this chip. Since I crunch almost 24x7 on this rig, I don't go for maximum overclocking and am usually happy with 200-300MHz over stock. I try to strike a balance between overclocks/voltage/temps (which is tough on Haswell!!!), so that I have a 100% reliable and stable computer that will happily crunch for a couple of years or more. That being said, I am hoping to get 4.5GHz out of this chip. I will be trying that later tonight...
 
Last edited:
Cali3350 said:
Wow. Just came from a I7 920 and the temps are very scary here.

I put my 4790K with a Corsair H60 and at load it quickly goes all the way up to 100C. Like, very quickly. Running Cinebench it fluctuates between 90-100C pretty quickly. Obviously I could have put the heat sink on badly but it seems solid, and ive certainly put many on before.

Very uncomfortable.
Click to expand...

Ok. Reset the heatsink and now running 4400Mhz at ~77C, which is hot but much more inline with what I would expect. Cinebench CPU tops out around 66C.
 
=Gunslinger= said:
They may have been sorting them for even longer than that, I've read of guys that are receiving 4790K's with 2013 manufacture dates. (assuming they follow the historical batch number assignments)

My "new" 4790K chip from Newegg was L352C119 and it's crap. :)
Click to expand...

The date (codes) on the IHS are assembly dates, which can be dramatically different from the actual fab date. Wafers can sit around waiting to get chopped up, chopped up die can sit around in reels waiting to get packaged, packaged die can sit around in trays waiting to get boxed or fused as i3/i5/i7, etc etc. All depends on what the market needs for that product and what other products may be getting higher priority through the assembly sites. At any time up until they go out the door to customers stuff can be moved around / regrouped so I wouldn't be surprised to find 4790K's that were fabbed in July of last year.

The conclusion I would write for Devil's Canyon is that it is a 4770k guaranteed to overclock decently (4.6+) and be a bit cooler while doing it. And costing the same as a 4770k.
 
Last edited:
=Gunslinger= said:
They may have been sorting them for even longer than that, I've read of guys that are receiving 4790K's with 2013 manufacture dates. (assuming they follow the historical batch number assignments)

My "new" 4790K chip from Newegg was L352C119 and it's crap. :)
Click to expand...



Me and about 5 other people on overclock.net have l352c119 chips and they are all bad.
 
gwhall007 said:
I received my 4790K (batch #: L418C169) from Newegg today and, for now, have just bumped the cache to 4.2GHz and all cores to 4.4GHz using just the MB Auto voltages for both (currently 1.182v for the cores and 1.223v for the cache). It is running at about 60C while crunching for BOINC on all 8 threads right now, using my H100i (ambient ~74F).

So far, I'm happy with this chip. Since I crunch almost 24x7 on this rig, I don't go for maximum overclocking and am usually happy with 200-300MHz over stock. I try to strike a balance between overclocks/voltage/temps (which is tough on Haswell!!!), so that I have a 100% reliable and stable computer that will happily crunch for a couple of years or more. That being said, I am hoping to get 4.5GHz out of this chip. I will be trying that later tonight...
Click to expand...

Currently running stable at 4.5GHz @ 1.195v. Ran Prime95 for 2 hours and the highest temp I saw was 81C. I'm now running it with 8 threads active in BOINC and temps are stable at ~63C, which is perfectly acceptable to me. I am really happy with this chip!
 
Cozmo85 said:
Me and about 5 other people on overclock.net have l352c119 chips and they are all bad.
Click to expand...
Funnily enough, I'm in the process of building new 4770K PCs for friends (they managed to get the 4770Ks cheap, ~$230 IIRC).
Got a 3401xxx and a 3408xxx batch chip (Costa Rica 2014 week 1 and week 8). Both seem to be great overclockers (Prime95 28.5 stable - 4.4GHz @ 1.15Vcore fixed, 4.5GHz @ 1.185Vcore fixed)
 
whoa that's some low voltage right there, what cooling are you guys using?
 
Nothing fancy, just Antec Kuhler 920s with IC Diamond. Same for my own launch 4770K in my sig.
 
wow, that's impressive.

good luck, where'd you get your cherry 4770k's? amazon? new egg?
 
gigatexal said:
wow, that's impressive.
good luck, where'd you get your cherry 4770k's? amazon? new egg?
Click to expand...
I think they picked them up from Amazon (I didn't order them, just helping them build)
I wouldn't say that they're exactly cherry in the sense that the TIM sucks on both of those chips. Getting peak temps in the 80s after 100 standard passes of IntelBurnTest 2.54, which should not be the case at those low voltages.

They'll probably be monsters once delidded though.
 
mine is proving to be decent

4.8 so far @1.31 with highest core temp of 80c with silver arrow SB-e.

I could do 4.6 @ 1.225
 
I got mine in from Newegg on Friday, but I probably won't get to it until this weekend. Just moved into a new house so we're still unpacking. L418C169.

 
You must log in or register to reply here.
Back
Top