Delidding Intels 7700k drops temps by almost 30C

[Dabombinable]

1 hour ago, ConiferousJelly said:

Exactly why they would fix it. Intel should start getting scared and wearing and actually do their best, that's what we need. We don't need Intel to fail, but to be challenged.

We also need AMD to challenge them more often, not every decade or so.

[patrickjp93]

1 hour ago, ConiferousJelly said:

Exactly why they would fix it. Intel should start getting scared and wearing and actually do their best, that's what we need. We don't need Intel to fail, but to be challenged.

What we need is the software world to get with the times and get with up-to-date tech. There's way more performance in current CPUs that games and Joe Consumer software can use if they're programmed by skilled people.

[Dabombinable]

12 minutes ago, patrickjp93 said:

What we need is the software world to get with the times and get with up-to-date tech. There's way more performance in current CPUs that games and Joe Consumer software can use if they're programmed by skilled people.

The main problem is, a lot of people still use old shit (eg, Phenom II, Core 2 Duo/uad).

 

Spoiler

Also, I'm having to learn C++ to make sure that I do not screw up compiling the AVX test, but judging by the amount of power AVX2 causes my 4790K to draw at only 4GHz (Prime95), it does show signs that it squeezes a lot of performance out of it

 

[patrickjp93]

4 hours ago, Dabombinable said:

The main problem is, a lot of people still use old shit (eg, Phenom II, Core 2 Duo/uad).

 

  Reveal hidden contents

Also, I'm having to learn C++ to make sure that I do not screw up compiling the AVX test, but judging by the amount of power AVX2 causes my 4790K to draw at only 4GHz (Prime95), it does show signs that it squeezes a lot of performance out of it

 

Code multiversioning is a thing (if you're not using Microsoft's compiler)

 

Prime95 uses all the heavyweight division and shuffle instructions and is tuned to not have an idle cycle. My program's nowhere near that abusive.

[porina]

2 hours ago, patrickjp93 said:

Prime95 uses all the heavyweight division and shuffle instructions and is tuned to not have an idle cycle. My program's nowhere near that abusive.

It is highly optimised to maximise throughput so it can create quite a load on a CPU. I wouldn't use the term "abusive" though.

[patrickjp93]

Just now, porina said:

It is highly optimised to maximise throughput so it can create quite a load on a CPU. I wouldn't use the term "abusive" though.

I would. Prime95 isn't the most efficient prime-finding algorithm. It, like Intel Burn Test, is designed to be a torture test to get as more power and heat coursing through a CPU as possible.

[Dabombinable]

6 minutes ago, patrickjp93 said:

I would. Prime95 isn't the most efficient prime-finding algorithm. It, like Intel Burn Test, is designed to be a torture test to get as more power and heat coursing through a CPU as possible.

Only if you select that option.....

[porina]

1 minute ago, patrickjp93 said:

I would. Prime95 isn't the most efficient prime-finding algorithm.

Prime95 isn't an algorithm, but software implementing one. Its original purpose was to find Mersenne primes, and in that it has been successful. The largest known prime number has been found by Prime95 for a lot of years, and they take most of the highest positions. http://primes.utm.edu/top20/page.php?id=3

 

If you're implying it isn't the best way to find ANY prime number, maybe. But that isn't its goal. Still, the math library as used in Prime95 is available for use elsewhere, in software such as LLR and PFGW. LLR is arguably the most used prime number testing software and implements tests for different types of prime numbers. You'll see it appear frequently in the top found prime numbers list also. At the time of writing, LLR is credited towards finding 4829 of the top 5000 known prime numbers. http://primes.utm.edu/bios/page.php?id=431

 

So basically, the hard working code underlying Prime95 is responsible for finding the vast majority of largest known prime numbers. It is no under-statement to say if you know how to do it better or faster, there will be a LOT of people who would be very interested.

 

1 minute ago, patrickjp93 said:

It, like Intel Burn Test, is designed to be a torture test to get as more power and heat coursing through a CPU as possible.

Just pain wrong. That it happens to be well optimised and a heavy stress had lead to its adoption as a stress tool also, but that is not its primary function. It does not try to maximise heat or power use for the sake of doing so, but it is optimised to do as much useful work as it can as quick as it can. In this community we might use it more for stress, but that doesn't change facts.

[Lays]

On 12/15/2016 at 6:24 PM, brighttail said:

 

 

NOTE ** This is not a recommendation to delid ANY CPU!  It is advanced knowledge and if you have any questions I suggest you google the video Linus did on the procedure.  The point of this post was to show that the thermal compound Intel is using on these pre-production and review CPUs MAY result in much higher temperatures.

 

The shitty procedure he did with a stock cooler and not using CLU?  The whole point of delidding is to put CLU under the IHS, because it's as close to solder as we can get.  I dropped nearly 25c when I delidded my chip and put CLU on it.   I could never benchmark at 5ghz before with my 6700k, in XTU it would hit 90+ celsius and needed 1.455 volts to even make it through 1 run, now I can run 5 ghz 24/7 and prime95 it at 5 ghz 1.43v with temps in the high 60s.

[Energycore]

On 12/15/2016 at 9:08 PM, Notional said:

Or you could wait for ZEN, which are soldered, instead of paying overprices for few cores and shitty thermal paste?

While the 8-core Zen will be soldered, I highly doubt the <$500 smaller chips will.

[Okjoek]

So was it the CPU lid or the Thermal Compound causing higher temps? Or was it both?

[Energycore]

12 hours ago, patrickjp93 said:

I would. Prime95 isn't the most efficient prime-finding algorithm. It, like Intel Burn Test, is designed to be a torture test to get as more power and heat coursing through a CPU as possible.

No it was not. Prime95 was designed as a way to look for primes. It just so happens that when they adopted AVX, it was really great for stress testing a CPU both in stability and temperatures.

 

Originally it's just a code implementation of the Lucas-Lehmer test for a number's primality.

[Energycore]

2 minutes ago, Okjoek said:

So was it the CPU lid or the Thermal Compound causing higher temps? Or was it both?

We don't know because tweaktown's procedure was bullshit. They literally tested with a less powerful cooler before delidding so the temp drop could all be due to them using a Kraken X62 on the second go.

[Dabombinable]

12 minutes ago, Lays said:

 

The shitty procedure he did with a stock cooler and not using CLU?  The whole point of delidding is to put CLU under the IHS, because it's as close to solder as we can get.  I dropped nearly 25c when I delidded my chip and put CLU on it.   I could never benchmark at 5ghz before with my 6700k, in XTU it would hit 90+ celsius and needed 1.455 volts to even make it through 1 run, now I can run 5 ghz 24/7 and prime95 it at 5 ghz 1.43v with temps in the high 60s.

What could I expect with my 4790K?

[Lays]

Just now, Dabombinable said:

What could I expect with my 4790K?

AT LEAST 15 with CLU, if you go look at the OCN thread literally everyone doing it gets 15-30c

[Dabombinable]

1 minute ago, Lays said:

AT LEAST 15 with CLU, if you go look at the OCN thread literally everyone doing it gets 15-30c

Ok then. That kind of drop combined with new fans means that I should be able to run mine at 4.8GHz safely+at a comfortable temp.

[MageTank]

3 minutes ago, Lays said:

AT LEAST 15 with CLU, if you go look at the OCN thread literally everyone doing it gets 15-30c

Shaved off 22C myself, on a 45mm vapor chamber on my 6700k. 

 

On 12/15/2016 at 9:24 PM, brighttail said:

NOTE ** This is not a recommendation to delid ANY CPU!  It is advanced knowledge and if you have any questions I suggest you google the video Linus did on the procedure.  The point of this post was to show that the thermal compound Intel is using on these pre-production and review CPUs MAY result in much higher temperatures.

Ha... that was a good one. Linus "That liquid metal stuff they use" Sebastian, Delid pro. Pays little attention to his own methodology (as flawed as it is) to notice the per-core temp difference window dropped from 13C on average, to 5C in that very video, even on that god awful stock cooler. 

 

I do want to point out that Dows Corning is NOT a bad thermal compound, even on the die. It's designed for longevity, and is still used by many companies. The biggest impact indeed comes from using high performance compounds/liquid metals on the die, but the biggest reason people delid, is core temperature uniformity. Where one core runs 10C hotter than the rest, and is a problem. The number one cause of this, isn't the compound itself, but actually the glue they use on the IHS/substrate. A buddy of mine had a chip so bad, that one core ran 16C hotter than the other 3 on average. Upon delidding, we noticed one side had way more glue in a specific spot than the others. He replaced it with AS5 (all he had at the time) and put the IHS back on with no glue (all scraped off) and while he didn't see the 20C difference most of us other delidders saw on every core, that other core came back down to match the others. 

 

For people that don't have this rare issue, or people that don't intend on pushing high overclocks, delidding isn't necessary. However, it's a flat out lie to say it's not beneficial, or offers insubstantial results, a narrative Linus tried to push with that video.

[Dabombinable]

1 minute ago, MageTank said:

Shaved off 22C myself, on a 45mm vapor chamber on my 6700k. 

 

 

Ha... that was a good one. Linus "That liquid metal stuff they use" Sebastian, Delid pro. Pays little attention to his own methodology (as flawed as it is) to notice the per-core temp difference window dropped from 13C on average, to 5C in that very video, even on that god awful stock cooler. 

 

I do want to point out that Dows Corning is NOT a bad thermal compound, even on the die. It's designed for longevity, and is still used by many companies. The biggest impact indeed comes from using high performance compounds/liquid metals on the die, but the biggest reason people delid, is core temperature uniformity. Where one core runs 10C hotter than the rest, and is a problem. The number one cause of this, isn't the compound itself, but actually the glue they use on the IHS/substrate. A buddy of mine had a chip so bad, that one core ran 16C hotter than the other 3 on average. Upon delidding, we noticed one side had way more glue in a specific spot than the others. He replaced it with AS5 (all he had at the time) and put the IHS back on with no glue (all scraped off) and while he didn't see the 20C difference most of us other delidders saw on every core, that other core came back down to match the others. 

 

For people that don't have this rare issue, or people that don't intend on pushing high overclocks, delidding isn't necessary. However, it's a flat out lie to say it's not beneficial, or offers insubstantial results, a narrative Linus tried to push with that video.

I'd probably just scrape off the glue to start with, and use a silver based thermal compound as the TIM.

[Lays]

1 minute ago, Dabombinable said:

I'd probably just scrape off the glue to start with, and use a silver based thermal compound as the TIM.

CLU buddy, CLU.

 

There is no better option for a delid. 

[MageTank]

1 minute ago, Dabombinable said:

I'd probably just scrape off the glue to start with, and use a silver based thermal compound as the TIM.

I didn't bother resealing mine. It's very easy to hold the IHS in place as you lower the retention arm, and makes it easier to work with later on. Most people advocate replacing CLU once a year, so not resealing is helpful in that regard. That being said, I've seen people run CLU for a couple years+ and not have a single issue. Mostly depends on how many thermal cycles it faces. 

[Dabombinable]

1 minute ago, Lays said:

CLU buddy, CLU.

 

There is no better option for a delid. 

It costs uite a bit more though, and I've seen what Galium can do....

[Lays]

Just now, MageTank said:

I didn't bother resealing mine. It's very easy to hold the IHS in place as you lower the retention arm, and makes it easier to work with later on. Most people advocate replacing CLU once a year, so not resealing is helpful in that regard. That being said, I've seen people run CLU for a couple years+ and not have a single issue. Mostly depends on how many thermal cycles it faces. 

 

 

Nor did I.

 

The safest practice when putting the CPU back in is to loosen the torx screws, put the arm down when there's no tension on it, then slowly tighten the torx screws back, that way the IHS doesn't move that tiny bit it does when you try the "push down method" lol

[Lays]

1 minute ago, Dabombinable said:

It costs uite a bit more though, and I've seen what Galium can do....

it's harmless for delidding, it's like $13 for 10c better than any other pastes you're going to find lol

[MageTank]

2 minutes ago, Dabombinable said:

It costs uite a bit more though, and I've seen what Galium can do....

Don't use aluminum, problem solved. Copper is superior anyways, unless weight is important to you. That being said, you can get away with nickel plated aluminum (if any heatsinks exist) as long as you don't accidentally scratch through the plating. 

 

EDIT: As @Lays said, for the bare die, it's perfectly fine. If using it on top (don't really recommend, as it doesn't do much) follow my advice above. Though, I personally just use gelid on top, lol. 

[Dabombinable]

3 minutes ago, Lays said:

it's harmless for delidding, it's like $13 for 10c better than any other pastes you're going to find lol

what's the difference going to be though? The silver based compound is also liquid metal.

2 minutes ago, MageTank said:

Don't use aluminum, problem solved. Copper is superior anyways, unless weight is important to you. That being said, you can get away with nickel plated aluminum (if any heatsinks exist) as long as you don't accidentally scratch through the plating. 

This is what I plan on using:

https://www.jaycar.com.au/heatsink-compound-25-silver-1-5g/p/NM2013

https://www.jaycar.com.au/medias/sys_master/images/h27/h5e/8886994862110/NM2013-msdsMain.pdf