Living dangerously; delidding your i7-7700k

Subject: Processors | January 30, 2017 - 02:29 PM |
Tagged: kaby lake, core i7 7700k, overclocking, delidding, risky business

Recently [H]ard|OCP popped the lid off of an i7-7700k to see if the rumours that once again Intel did not use high quality thermal interface material underneath the heatspreader.  The experiment was a success in one way, the temperatures dropped 25.28%, from 91C to 68C. However the performance did not change much, they still could not reach a stable 5GHz overclock.  They did not let that initial failure discourage them and spent some more time with their enhanced Kaby Lake processor to find scenarios in which they could reach or pass the 5GHz mark. They met with success when they reduced the RAM frequency to 2666MHz, by disabling Hyperthreading they could reach 5GHz with 3600MHz RAM but only when they increased the VCore did they manage to break 5GHz. 

Of course you must exercise caution when tweaking to this level, a higher VCore will certainly reduce the lifespan of your chip and delidding can have a disastrous outcome even if done carefully.  If you are interested in trying this, The Tech Report has a link to a 3D printed tool to help you in your endeavours.

View Full Size

"Last week we shared our overclocking results with our retail purchased Core i7-7700K Kaby Lake processor. We then took the Integrated Heat Spreader off, replaced the Thermal Interface Material and tried again for 5GHz with 3600MHz memory and failed. This time, less RAM MHz and more core voltage!"

Here are some more Processor articles from around the web:

Processors

Source: [H]ard|OCP

January 30, 2017 | 02:55 PM - Posted by willmore

I'll print you one if you'll test it!

January 30, 2017 | 03:22 PM - Posted by remc86007

Every intel processor I've had did not like high clocked memory and overclocks at the same time. It appears that lately games scale with higher clocked memory which makes maximizing performance a bit more interesting than it used to be. It will be interesting to see how Ryzen does in this regard.

January 30, 2017 | 05:04 PM - Posted by Anonymous (not verified)

I want PCper to look at some Zen server SKUs also for those that may want to do some rendering workloads and such! Let's see how AMD's Zen workstation/server SKUs perform with 16 and 32 Zen cores or maybe some lower cost 16 core Zen server SKUs on dual socket AMD Zen/Server variant motherboard SKUs. I'll bet that there will be some 16 core Zen server variants that will still be priced less than Intel's 10 core E series Systems.

The Ryzen 8 core(2 CCX units) SKUs are really just half a 16 core(4 CCX units) Server variant in the first place as AMD has not had enough time to really spend on its consumer tuned Consumer Ryzen SKUs(APUs, other CPU only variants). There may be, after AMD gets some Ryzen/Vega APU SKUs on the market, time for AMD to maybe offer some 6 core Ryzen(Non APU) Variants but anyone can see that AMD is really focusing first on the server market Zen(CPU cores only) variants and that the first consumer Ryzen SKUs were derived directly from half a 16 core Zen server SKU! And for HEDT consumers that is not a bad thing in and of itself.

AMD needs more time the get its Ryzen APUs to market, hopefully with at least one APU variant that has at least one HBM2 stack that acts like an HBM2 cache to at least 16 Vega NCUs on an interposer based Zen/Vega/HBM2 APU SKU. Even 4 GB of HBM2 could leverage a larger amount of slower DIMM based DDR4 DRAM in such a way that the integrated Vega graphics will operate from the HBM2 memory! That Vega High Bandwidth cache controller(HBCC) IP should be able to manage any amount of HBM2 to allow for the integrated graphics to operate from the HBM2 and not the slower DDR4 DIMM based DRAM with the HBCC managing any slower DDR4 to HBM2 transfers in the background from DDR4 memory and virtual memory page swap/other onto the HBM2. The New Vega full reveal is not to far off and I’d expect that by that time AMD will be ready to talk more about its Raven Ridge Ryzen/APU line, including AMD’s workstation class Zen/Vega/HBM2 SKUs with even larger Vega dies.

PCIe 4.0 is going to be great for laptops and PCs when it finally gets to market. For any server APU variants I’ll wait until AMD fully reveals their infinity fabric IP and how it relates to AMD’s entire line of consumer/workstation/HPC products.

Also AMD is a founding member of OpenCAPI and that's IBM/AMD/others getting together and making an even faster standard around IBM's CAPI(Now OpenCAPI) standard than even PCIe 4.0 or maybe even NVlink. So the OpenCAPI IP looks to be very interesting also along with AMD's infinity fabric IP.

April 28, 2017 | 08:02 AM - Posted by TheTolsonator

SUPER relevant to the topic at hand.

Back to the ACTUAL topic at hand, delidding the processor to ease down temps sounds like something I'd be interested in, but the hefty $250+ price tag outweighs the possibility of getting lower temps

January 30, 2017 | 04:01 PM - Posted by John H (not verified)

For some reason Kyle didn't run the chip with AVX offset. So the delidding was good for 100-200 MHz with AVX code, it may not have made a difference with non AVX code (games).

January 30, 2017 | 04:11 PM - Posted by Anonymous (not verified)

In past generations of Intel CPUs, the benefits observed in delidding have almost entirely come from reducing the gap between the IHS and the CPU. The idea that Intel's TIM is of low quality has - as far as I know - never been true.

Somebody even coined the term "gap lottery" to describe the way the thickness of the adhesive layer between the IHS and the CPU varies between individual CPU samples, sometimes resulting in an unusually cool/hot running CPU.

However, I'm not certain if any of this still applies to the Kaby Lake generation of CPUs.

January 30, 2017 | 04:48 PM - Posted by Anonymous (not verified)

"temperatures dropped 25.28%, from 91C to 68C"

Temperature values in Celsius are not measured from absolute 0, so it was not a "25.28%" drop. Simply saying they dropped from 91C to 68C is sufficient for people to recognize it as significant.

January 30, 2017 | 06:22 PM - Posted by Jeremy Hellstrom

The Kelvin scale is not what we use when discussing cooling PCs.

January 30, 2017 | 08:27 PM - Posted by Anonymous (not verified)

Yeah, talking about percentages with a relative scale is a pet peeve of mine, too. I just try my best to ignore it, because so many people do it.

January 31, 2017 | 11:58 AM - Posted by Jann5s

I agree, Jeremy, the drop of 23C is enormous and much more than the 25% indicates. And it really doesn't make sense to discuss percentages without have a well defined zero.

(I would say that zero is room temperature if I had to pick one, since it is the lowest temp you can reach with an air-cooler)

January 31, 2017 | 01:20 PM - Posted by Jeremy Hellstrom

The pedantry aimed at me, which is caused by a simple copy and paste from the source article never fails to amaze me.

February 1, 2017 | 05:14 AM - Posted by Jann5s

I'm sorry if my remark was perceived as pedantic, it wasn't meant as such. Also, it wasn't clear that it was a simple copy and paste as I didn't read the original article (I mean where are the quotation marks).

However, don't you agree that the % statement does the delidding result injustice. To me the temperatures dropped much more than 25%

February 1, 2017 | 01:14 PM - Posted by Jeremy Hellstrom

The pedant crack wasn't necessarily aimed directly at you, you were polite.  The math on that percentage is bad however you look at it, but it does give those who don't want to calculate the difference a quick and dirty stat to read so I included it.

It was more a general observation about the amount of people making comments in this post and others which add nothing to the conversation. I could add quotes but it won't really have any effect on those that come here just because they enjoy making disparaging comments.

February 1, 2017 | 04:52 PM - Posted by Jann5s

I understand, thx for the response

January 30, 2017 | 06:15 PM - Posted by Anonymous (not verified)

0 Celsius is not absolute zero -273 is. Thus the temperature dropped a mere 6.32%

January 30, 2017 | 06:25 PM - Posted by Jeremy Hellstrom

We could also use -40 as the assumed base.

January 31, 2017 | 06:18 AM - Posted by John H (not verified)

I thought Canada could go below -40

January 31, 2017 | 01:18 PM - Posted by Jeremy Hellstrom

I was making a temperature scaling joke in Fahrenheit and Celsius at the same time.

January 30, 2017 | 07:55 PM - Posted by Anonymous (not verified)

It IS true you are supposed to use Absolute Zero as the starting point, however I'm not sure that level of scientific accuracy is needed here. We get the point.

We could have said "relative to the freezing point..." and made it technically correct too likely but again we get the point.

January 30, 2017 | 10:45 PM - Posted by BlackDove (not verified)

Intel should use CVD diamond or diamond/copper diamond/silver metal matrix composite IHS electroplated with gold like other companies(Fujitsu) do.

January 30, 2017 | 10:46 PM - Posted by BlackDove (not verified)

And they should use indium silver solder TIM between the CPU die and IHS as well

January 31, 2017 | 08:18 AM - Posted by Master Chen (not verified)

Sandy-gods are laughing out loud.
I am Sandy-god. Are YOU?

January 31, 2017 | 05:47 PM - Posted by razor512

Someone should sue Intel for deliberately adding thermal insulation between the heat spreader and die. Consider that even toothpaste doesn't perform that much worst than good thermal compound.