UK-based system builders have claimed that retail Haswell processors run significantly hotter and with lower overclocking potential than pre-launch engineering samples.
System builders have reportedly been blind-sided by Intel's latest fourth-generation Core series processors, with retail models proving significantly different to pre-production versions previously supplied.
Intel's latest Haswell chips represent a tock - the changing of the microarchitecture, rather than process node - in Intel's
annual tick-tock development cycle. Based around an improved architecture, the processors include boosted graphics performance and improved speeds along with new instructions designed to further increase software performance. While the company is keeping its best innovation, its high-performance GT3e 'Iris Pro' graphics technology, for OEM-only parts supplied in ball-grid array (BGA) packages, it's clearly still hoping to make an impact with its more traditional land-grid array (LGA) parts.
Sadly, the impact it's made so far has been somewhat negative - at least where system builders are concerned. Anonymous individuals from four UK-based computer manufacturers have spoken to our sister publication
PC Pro with the news that retail Haswell parts are considerably less capable than the pre-production versions on which they had been basing their engineering work.
Prior to each processor release, Intel produces a number of engineering sample chips. These chips are, theoretically, identical to the finished retail product, minus a few last-minute fixes that may be inserted to work around some bug or other. They are provided to hardware partners in order for pre-release engineering work - designing desktops, laptops, tablets and mobiles around the requirements of the part - to be carried out before mass production begins. This allows manufacturers to launch their products day-and-date with Intel's chip release.
Sadly, it doesn't appear to have worked out quite that way for Haswell. According to PC Pro, manufacturers who produce pre-overclocked gaming systems have found that the retail chips they've just received are only hitting speeds of 4.2GHz to 4.4GHz stably compared to up to 4.8GHz with pre-production engineering samples.
One manufacturer claims that pre-release chips marked as 3.5GHz parts were tested completely stably at 4.8GHz, but of the 40-50 retail units the company has tested not a single one has managed to reach above 4.2GHz without hitting unsafe temperatures or requiring too-high levels of voltage. Another firm has stated that it has had to drop plans to offer pre-overclocked Haswell systems running at 4.5GHz - a figure, again, planned based on engineering work carried out on pre-production samples provided by Intel - to 4.3GHz in order to ensure stability. '
There is a big difference in the overclocking potential between early Haswell samples and retail,' the unnamed manufacturer claimed.
Another manufacturer claims that retail Haswell parts are proving too hot to handle, stating that even while running at stock speeds the chips reach higher temperatures than the pre-production engineering samples - by around 15 degrees Celsius, according to the unnamed company's tests - or even the retail models of Intel's last-generation Ivy Bridge chips.
Intel, for its part, has refused to comment on the specifics of the manufacturers' claims, stating only that overclocking is not covered by the its warranties and that companies - or individuals - doing so are boosting performance at their own risk. The company has also stated that the overclocking potential of a given chip will vary from batch to batch, but has provided no explanation for the apparent vast difference in heat output between pre-production and retail Haswell models.
Our review of Intel's Core i7-4770K
can be found here, and while we were able to get our chip up to 4.7GHz it wasn't without difficulty: despite using a
Corsair H100i sealed-loop cooler, the temperature of the chip hit a whopping 98 degrees Celsius - giving us cause to question just how far a Haswell chip can be pushed for long-term use.
103 Comments
Discuss in the forums ReplyI guess they must be saving a lot of money by using cheap thermal paste instead of soldering IHS.
At the moment, if it turns out that Ivy Bridge E or Haswell E are stupidly hot as well then I will stick with my SB-E setup or switch to back AMD.
There is no way I can currently use LN2 or Dry Ice to cool a cpu, and since that's, pretty much what I would need to get these cpu's up to a decent speed Intel are doing something wrong.
Yeah
It wasn't done to save money.
For a moderate OC, SB/IB shuold run 4Ghz on stock voltages and with stock (boxed) cooling, does Haswell?
Well it saves me money as there is no way I'm buying a Haswell.
Why was it done?
If the type of thermal paste used under the heatspreader is the cause (as is the case with Ivy Bridge), compounded by the current heat generation situation then I would like to think Intel will address this ASAP with a new stepping.
I hope this is not a return to the leakage suffered in the days of Prescott cpu's, because this time round, in terms of outright cpu performance, there's no alternative and no reason to buy this one....I wonder how long IB will remain available as a result?
+1 !
The only thing that springs to my mind would be to reduce the chips lifespan !
between the indium and the indium-gold intermetallic became uncontrollable leading to voids forming in the solder.
For a more detailed study this make very interesting reading The Material Optimization and Reliability Characterization of an Indium-Solder Thermal Interface Material for CPU Packaging (warning PDF)
People need to start slicing the tops off their Haswells to get those temps down. Then again they could always use liquid and Aquarium chiller.
To be honest, you'd probably be better off with a 2500K, the 3570K runs hot too!
When users delid their chip and replace the TIM, they use a thinner layer so they see a reduction in temperatures. Even if they used the TIM Intel is using, they'd still see an equivalent reduction.
To bad you have void the warranty on the cpu to fix the mistake Intel made.
Reminds of the Rampage III Extreme issue, where if you took of the heatsink for the north and south bridge chips and applied new thermal paste, you fixed a issue where the board would report your cpu as overheating even running at stock speed.
Atleast Asus fixed that issue though.
And those that do fall into the magical moderate category... at what point does it stop being moderate? Over a certain voltage threshold? Over a certain ghz?
I have friends that want an overclock... none are interested into what I'd guess is called, a moderate overclock. Surely there is everyday and extreme? If I was to dish out labels that is. And this depends on cooling capabilities... and moderation was required to reach those limits. Perhaps those are moderate overclocks...
It seems things are kind of going downhill with Intel.
All they need to do now is kill off CPU sockets and start welding every CPU to motherboard.
Actually they could extend that to welding the CPU cooler to the CPU...
:)
Which is pretty similar to the overclock on my i5 3570k... I run at 4.7ghz 24/7 using offset overclocking method and as far as I am aware have not seen anything too shockingly high either. (I went from the i5 2500k @ 4.7ghz also)
I was convinced I was going to have to do the IHS removal... but, in the end, I was comfortable with the temps as is until I approach the end of my warranty.
And when I hit the end of the warranty period, I will be more than happy to remove that IHS and pump it up to ~5ghz. (It's not too bad a chip overclock wise looking around at others on the web)
You would have to be insane to do it. I could understand if you had a slower system than what AMD has launched by the x79 sandy systems destroy any AMD system in every test ive seen. You know something the rest of us dont and AMD making a wonder chip for us to all envy ?
Whats classed as regular overclocking 5ghz? 1.5vcore? seen alot of people assuming 5ghz is normal but if what i remember is accurate only early sandy chips hit 5ghz most hit 4.9 and that was there wall and required a high vcore to get it which resulted in 100c temps on most of those chips.
No ivy chip has ever hit 5ghz without a resonable voltage increase 1.4-1.5 is the general amount found around the web. Even then the Wall for Ivy was 4.7ghz.
So what are they putting through these chips to hit 100c? Seems to me like Manufactures might actually have to get some effort into overclocking again instead of the crap they usauly spam, Scan ocuk are two of the worst at this in the uk ( anandtech runs rampant on companies that do the same thing as well), People review there systems and none of them are using offset voltages and all could be undervolted from 0.1 to 0.9 vcore on the stock and tested that they banged into it.
From my point of view anything north of 4.4ghz and 1.3vcore requires a full on loop not a poxy h100i which despite claims to the world is not a replacement for a full blown water cooling setup.
Most of the reviews of pre built systems with overclocks always state the same thing the overclock is very poorly done and not using offsets, If you take 10mins yourself you can reduce the vcore and set some offsets.
Most people ive overclocked for do not need the performance they want it as a figure to brag to there m8s my cpu does 5ghz ect.
And those pre-overclocked bundles, like you say, time and time again come overvolted and not using the offset mode.
My ivy 3570k though is sat around ~1.3v for 4.7ghz. But as you say, it requires a wc loop to tame it.
already been done on a blind test.
A guide if anyone owns one also has explain on chip quality pretty interesting read in truth.
Where? (out of curiosity, not argumentativeness.)
It's all relative. Like I was trying to say in the Richland review thread, nothing is that simple. Sometimes, good enough and a specific feature set is all that is required.
I'm probably in the 10% because I notice immediately if my CPU is down-clocked to 4.2ghz. Everything is more sluggish. Not slow... just not lightning fast I get at 4.9ghz.
"Why does everyone threaten to switch back to amd from socket 2011 platforms thats just asking for a 50% downgrade in prorgrams outside of games. " << because intel have threatened to weld CPU's to motherboard and perhaps get rid of user overclocking completely.
Well, there's a simple solution to that:
Put more damn cores on the chip to make it bigger!
So you would go to a slower system that's slower than all socket 2011 systems even if they stock and the AMD Chip is fully oced you enjoy that I would not.
And they if they wanted to do so, they'd do it anyway... and no-one at intel would lose any sleep over the matter...
i'm with rollo... overclocking has been taken for granted... the Core 2 Duos and early i7s [like my 920] could be pushed so high, so easily that the act of overclocking was reduced to typing a couple of numbers into the bios [for most people]... and in the event you do go too far... a quick button press and all is forgiven...
think back to when overclocking required short circuiting stuff on the processor and you ran the very real risk of breaking stuff if you messed it up... when overclocking required thought, skill and often a smidge of luck... all for a few Mhz... the overclockers of yesterday would most likely tell you to quit your whining... any speed bump you can get out of the chip is welcome and never guaranteed...
so... quit your whining...
As things stand at the moment I would still stay with Intel. I've never owned an AMD cpu but I was 2 inches from buying one some years ago until Intel released the first core duo or something (it's all vague and wishy washy in my head now). I do read up on AMD once in a while to see where they're headed.
I just think that if Intel ever do completely kill manual overclocking then I might just go to AMD.
Regarding this particular thermal issue I'd actually quite like the challenge of seeing how far I can cool the Haswell. That's probably not a good enough reason for me to ditch existing mobo and cpu though.
If the AMD chip could still cope, then probably yeah.
Some people want different things out of computers.
I'll say no more :)
Cheers for this Rollo I have an i5 with H100i to setup this weekend and will post my results in the hardware thread. I'm not expecting miracles but a nice overclock will do for now.
Dude... I was overclocking back in the days of DX25 or whatever it was called. I used to do extreme overclocking wearing nothing but underpants for protection.
Is that water cooled?
My 2700k (see sig), hits 85c when I overclock to 4.9ghz. That's on air (Thermalright silver arrow cooler).
I reckon I could easily clear 5ghz, if I could keep the heat down.
Yup, liquid. 85c is pretty good at 4.9ghz with HT I would say. Presumably the temps would drop about 10c if HT was disabled? (not that you'd want to but just for comparison).
I might try later if I get a chance. The only problem is its actually quite warm this evening, for a change, so it will mess up my ambient temps.
I looked up your air cooler... It's a monster. :)
Yes, it practically fills my whole case lol
I love my 3770k, but I couldn't dream of running it as I do if I hadn't de-lidded it. Once I did, the load temps dropped by 20 degrees. Running at 4.6ghz the max I hit is 60deg now.
If Haswell runs slightly hotter than Ivy, then I'm really not surprised at this report. It's a shame though as they are such good chips.
Damn I wish AMD could compete, doesn't help they've got less total income than Intel does Research funding.
As far as i can see the heat issues are only going to get worse, as the die's keep shrinking it gets harder and harder to remove the heat from the smaller surface area.
It makes me wounder how far Mores law can go before physics becomes a problem :(
(remember the 1Ghz barrier?....naah you probably wouldn't)
On topic:
A ceiling of 4.2-4.4GHz means an overclocked SB chip can perform faster. Whoops.
My bargain basement 2500K is looking pretty good right now.
Well I did some AIDA64 benchmark tests last week and my 4.9ghz 2500k is still at the top of a lot of the benchmarks.
Sadly i do but as far as i remember that wasn't because of heat issues, but for all i know dementia could be setting in :)
The problem has already caused Intel to stop using solder for the TIM, and its a issue cause by physics in that you can only remove X amount of heat from Y surface area.
Unless a better way of removing heat from the ever decreasing size of CPU die's is found, or we change the laws of nature we are all Doomed, Doomed i tell you :)
I just bought a FX 8320 for £113 and I'm having tons of fun with it. Really shines in Crysis 3 and haven't met a game it can't run at acceptable levels :)
I think people will go with whatever gives the most performance. At one time that was AMD. I remember when all my friends had AMD processors. I was one of the few with intel and I lost in most benchmarks to their CPUs.
http://www.cpubenchmark.net/cpu.php?cpu=AMD+FX-8320+Eight-Core
Then I read this.
http://www.eurogamer.net/articles/digitalfoundry-future-proofing-your-pc-for-next-gen
So I bought one and reviewed it :)
http://linustechtips.com/main/topic/22908-amd-fx-8320-and-asrock-990fx-extreme3-high-end-on-a-budget/
Everyone wishes AMD could compete but it won't happen anymore.
See my price to performance comparison above. AMD are winning right now, and offer two of the best value CPUs per pound/performance in the world.
What's nice is you don't even notice that they're slower in games. In Crysis 3 I demolished my old Xeon E3 (I5 2400) with my overclocked FX :)
Games just don't need any more and won't for quite a while. Outside of gaming then yes, there are faster options if you need them.
Hmm. With this impasse Intel find themselves in it would be nice if AMD can improve upon Piledriver (with Steamroller) the same way that they improved upon Bulldozer.
If they do they'll be right up Intel's chuff, and I bet their chips will be as cheap as chips (groan).
im not so sure I5 2500K isnt cheaper than haswell. Id like to see the comparison though between the overclocks posbbile on air, not your stupid overclocks just bog standard everyday achievable ones.
something like I5 2500k @ 4.6Ghz vs a 4670K @4.2-4.
Just chucking a date and time out there. I've marked it down in my calendar also so I can confirm nearer the time.
Yeah exactly.
I'd definitely go AMD if Intel stopped allowing overclocking entirely. All you'd get for £113 Intel wise is a dual core IB 3420 - I know which one I'd rather have, after spending another £100+ on my X-OC-ROG-Boost-DrMOS-etc motherboard.
There may not be a huge difference in new prices (maybe a tenner?), but the marketplace is flooded with second hand Sandy Bridge chips.
Yeah that is utter madness. I did some price comparisons when I bought mine and you're spot on, all you can get for that is an I3 that's locked. I have a Xeon sitting here that I clocked to 3.5ghz using the FSB and the FX (clocked to 4.2ghz) doubles the framerates I get in Crysis 3.
What made it more fun was how unstable the multi was, forcing me to do it the hard way with the FSB and then offsetting my memory etc. Haven't had that much fun in ages :D
It's a very prospective purchase in some ways (relying on the new consoles to give it oxygen) but even with older games the lowest min FPS I saw was 35 with Hitman running 8xmsaa.
Gotta agree with Shirty. There's bargains to be had in the market place.
I disagree totally.... It reminds me of the days people used to say games will never need 64k ram.
I dunt know what games you run but I run stuff that uses considerably more. Some people mod skyrim to start with. Some people run flight sims...
If I drop my i5 2500k from 4.9ghz down to 4.2ghz (I did last week for testing purposes) I can see the difference immediately in windows.
I have sims that can easily max a quad core on overclock.
Flight sims are (and have been for a long time) in huge decline, and the proportion of people playing heavily modded games pales into insignificance compared to those who play the vanilla games (or gently modded versions).
By your own logic you should be gaming on a dual processor rig with 64GB and quad Titans.
I've had the 2500 up to 5.1 but in reality leaving it there will kill it much more quickly, use far more electricity and heat up the whole rig.
4.6GHz for the most part with a decent £35 cooler is the norm. The same as ivy-bridge.
I Think it is a bad thing the dy shrinking, it may be smaller but seems like it will generate more heat instead of less. Wasn't it the whole purpose of shrinking the processors anyway, to get less heat?
If it runs 3.8Ghz or 4.4Ghz you never going to notice the difference unless you like viewing numbers on benchmark software all day. the 2600 still can keep up with all the generations after it.
There is just no innovation in chip design that makes you go wow!... Unless you go mobile...
I already booted at 4.8-4.9ghz with 1.4v, but it crashes alot...
At 4.8ghz i reach 100ºc and it starts to throttle down... I have to delid the cpu... :(
Haswell is Intel's Bulldozer yet no one can admit it because they've spent the last few years praising Intel.
It's very difficult to be brutally honest about something that was supposed to be amazing, yet failed to deliver. After Sandybridge they were always going to struggle to get the sorts of yields that would impress, and as each new CPU comes along they are getting more and more disappointing.
They've simply hit a big brick wall in that rush to get out more and more new CPUs to woo people with.
I leave my 2500 at 4.9ghz using offset voltage. The power consumption is minimal for normal day to day use. 112 watts for the computer. I use offset voltage so it drops to normal low levels when idle.
It's not that flightsims are in huge decline... they are as much in demand now as they were 10 years ago. It's just that they aren't played by casual or console gamers. There are people who run flightsims as part of their flight training and just for fun.
It was never something that was played by everyone and suddenly the majority have stopped playing them. It was always a niche area and that area is still going strong for flight enthusiasts.
Loads of vids on youtube for skyrim modding and again not everyone is going to be modding skyrim but there are plenty of people who are into it and want to make it look better.
I personally don't like how windows runs when I'm clocked down to 4.2ghz. I am used to 4.9 and dropping to 4.2 for example, the difference is very noticeable.
That doesn't mean everyone should OC to 4.9 but I can certainly tell the difference.
If your defending AMD but own a 3770k what's the point you did not buy there product. Your not saving them from going broke by buying Intel are ya?
Lets get this straight computer desktop sellers are lazy and just want a quick buck, scan OCUK in the uk are both awful overclockers who don't know what offset means.
Anandtechs reviews of computers from the states shows its a widespread problem.
We are not taking about what somebody in these forums can do.
Haswell is about 20% faster than sandy clock for clock that I have seen under heavy overclocks.
So your 5ghz overclock is only around 4ghz on haswell ( evga forums someone has ran the overclock comparisons in game benchmarks using same gpus)
As the manufacturing process goes down heat will go up and power go down, the surface area that your CPU cooler can cool is alot smaller than compared to a i7920 for example.
Until Intel can develop a way to spread the heat more, its already been delid and found the tim was good but there's this big black stuff getting in the road of proper heat transfer, so even delid is at best 5-10c.
4.8Ghz, air cooled Haswell at 1.24v with 2800Mhz Corsair Doms in my system. Doesnt go above 80c after hours of prime, it hasnt had the lid removed and tim reaplied and who would have thought that one of those at OcUK who doesnt know what he is doing set the OC on this system?
Perhaps rollo you could give me a tutorial given how useless OcUK, Scan and those in the tech departments must be :)
There no need mate, so long as you know what your doing and dont just crank up the vcore like a noob then they dont run as hot as people let on :)
I'm not trying to get shot at, torn apart or anything here - this is pure curiosity - is yours a standard retail chip or one of the 'early' ones? If it's a standard retail chip, this is a pretty good result! :)
You seem to be a master overclocker, maybe you should post your rig details and bios settings and we can replicate your success
Standard run of the mill retail chip first batch of official sale worthy stock in the UK :)
That's not a bad result at all - it's a big clock speed jump. Although 80deg is a touch too high for me - I like to max at 65deg.
+1 to showing us your settings and set-up
Can you provide a screengrab of smallfft running for an extended period with core temp and cpu z being displayed?
Maybe it's worth the commute. OCUK Owned by a German company now int they :)
In fact given that this is a tock. We'll see.
But don't they run cooler when not pushed to the max and not over-clocked? They certainly use less power and therefore should be generating less heat.
I thought the whole purpose of this chipset was to lower the power requirement rather than to boot overclocking potential?
I'm not sure but i think the "normal" temperature is about the same as SB, and i think much of this is to do with the way Haswell is designed what with the upgrade in the IGPU and moving the VRM's on to the chip.
It seems more components are incorporate with each new design, Memory controllers, integrated GPUs, northbridges, and southbridges have all moved from the MoBo to on chip :(
The reasons for moving components into the CPU are always sold to everyone as being a good thing, and i dont doubt that they are.
But you have to question what impact this has had on temperatures within the core, and how much faster current CPU's would have been without all that extra heat :?
When the heat transfer isn't great, sustained temperatures will be a bit higher. Also, Haswell suffers from the same problem as IVB: it doesn't consume a lot of power but it's a small die, meaning it will be hotter per unit energy consumed.
EK Water Blocks has released a cheap kit that allows users of existing Supremacy CPU blocks to remove the IHS and let the water block make direct contact with the CPU die.
http://www.ekwb.com/shop/ek-supremacy-precisemount-add-on-naked-ivy.html
http://www.ekwb.com/shop/blocks/cpu-blocks/supremacy.html
http://www.xtremesystems.org/forums/showthread.php?280892-3770k-IHS-Removals-CPU-temp-dropped-from-79C-to-71C/page3
I'm sticking with my IB i5 3330, nothing to see here...move along.
There is nothing wrong with the quality of the TIM used, the application of the TIM leaves a lot to be desired and could be improved.
It has nothing to do with cost and everything to do with moving components from the MoBo to inside the chip, along with the smaller contact area available.
+1 !! Mine will NOT boot at 4.6 regardless of the vcore. It was even crashing in the bios when trying to dial it down. It's taking 1.35 to get it running at what seems to be a stable 4.5ghz. Getting temps of about 71 after a couple of hours of Crysis 3 with a Corsair h110. It's been a lot more hassle to OC than my 3770k was for sure!
this would enable longer low power sleep, lower power consumption at the same clock and maybe lower voltage all because of the lower temperature.
Settings vary from CPU to CPU and they vary board to board so if I provided my settings they will most likely not work with your kit.
thats a cop out and you know it
No but I have had 4 chips played with in this very board and all run similar clocks, similar memory but need different vcore and or chipset voltages hence the differences which negates the point of posting settings.
From the little I've seen/played around, they do tend to run a little hotter, that doesn't mean that every single chip does, merely that on average they do, which is all the article is saying I think?
Just because you've got chips which perform better doesn't mean other people's finding are worthless?
Also did 8-pack used to be involved in phase change systems? Under the name of Paz or something?
This has already been covered, Indium-Solder can not be used with smaller die sizes.