Are "Silicon Lottery" CPU's really just CPU's with a good heat transfer?

[WeeemRCB]

I've seen quite a few videos recently where people were de-lidding their CPU and replacing the stock thermal compound under the IHS with either a better compound or in most cases, a liquid metal type of TIM.

 

Every one of them see an improvement in their CPU temps and as a result more solid overclocking (up to 25C)

Much less so with soldered chips, but still, there's an improvement over stock (about 5C)

 

One guy even did a GTX1080 and the card was able to regulate it's own clock higher before he did anything with MSI as it was running cooler.

 

And before anyone says it - yes I've seen the LTT video about delidding and Linus was pretty scruffy in his process... and still got an improvement over stock.

It seems that the more meticulous you are using a safe delidding tool from der8auer (in EU) or Rockit Cool (USA) and just ....taking your time all seems to pay dividends in the resulting CPU temps.

Expecially so if the stock TIM degrades over years of use.

 

So, is the Silicon Lottery just a better appliance of thermal paste rather than the quality of the chip underneath?

I imagine there will be some minor tolerances in the manufacturing process, but improving the contact between die and IHS seems to be a big part of making your CPU work more efficiently and faster.

 

 

Referenced Links:

 

Upgrade Addiction: 1080 + liquid metal

 

 

LTT Delidding:

 

 

der8auer: 

 

Rockit Cool modest example:

 

 

 

[LUUD18]

It really depends on the chipset. Some Intel CPU's have already good thermal compound.

[WereCat]

It is not just about heat transfer, it is about how much clock can you push on a given voltage.

[iamthecpu]

Also the silicone lottery is more about how fast the individual cpu will overclock to. For example some Intel cpu's will clock to 4.4 and some will clock to 4.6 it all depends on the luck of the draw. 

[mr moose]

Yes and no, Yes because all overclocking is thermally dependent, so improving the thermals will more often than not result in better overclocking.  However some yields are just so average that even the best thermal scenario (from factory or aftermarket) will not give a better result.   How varying the yields are can be seen as many intel chips are the exact same die, it's just they disable under performing features and then stamp it as a  lower spec chip rather than a xeon/i7 etc. 

[Princess Luna]

I'm pretty sure silicon lottery is more towards how pure the silicon is allowing it to function optimal on higher frequencies, heat is a great weight factor but not the sole one, even if you delid two identical CPUs one can still be better than the other given the silicon purity.

[Jurrunio]

Silicon Lottery refers to the chance of getting a chip with too-good quality (high purity of silicon in transistors of the core). Those chips can run at higher frequencies than others under the same conditions. Getting better heat transfer helps achieving higher frequencies, but chips with better silicon purity still goes faster than others.

[JoostinOnline]

Delidding has nothing to do with the TIM.  I explained it here.

[WeeemRCB]

2 hours ago, JoostinOnline said:

Delidding has nothing to do with the TIM.  I explained it here.

Yes. I can see that being a factor and there was a video by der8auer (I think) where he warns that you have to be careful with the replacement glue as may be too thick and prevent good contact between the die and the IHS.

 

Perhaps its a balance of all things? Closer/Direct contact with the IHS being the biggest factor and a better compound that removes more of the micro-air pockets that could create tiny hotspots being to icing on the cake.

 

I've heard of some nutjobs people not putting the IHS back on and instead mounting the cooler directly to the die.

But I think that's taking it a bit to far....

[Mira Yurizaki]

The lottery has to do with how well the processor is made to specifications. Manufacturing leads to different variances (i.e. defects) and Intel only tests to make sure the processor can perform within a specified tolerance. A "bad" processor had enough defects to not overclock well, yet still be within the tolerance of what its stock performance is. The defects in the manufacturing could affect things like the variance of switching, if the transistors can handle a faster clock signal, etc.

 

Thermals have nothing to do with it. All thermals have to do with overclocking is it has a relationship with this principle: P = CV^2f, where P is the amount of power (or heat) dissipated, C is the capacitance of the transistor, V is the voltage, and f is the frequency. And you need to dissipate heat in order for the processor to remain stable. Also, while you might think voltage can play a huge role because it's squared, it's very small compared to f, like at least 8-9 orders of magnitude.

 

Also the title should be "silicon", not "silicone", unless we're talking about exploding breast implants.

[done12many2]

3 hours ago, M.Yurizaki said:

Thermals have nothing to do with it. All thermals have to do with overclocking is it has a relationship with this principle: P = CV^2f, where P is the amount of power (or heat) dissipated, C is the capacitance of the transistor, V is the voltage, and f is the frequency. And you need to dissipate heat in order for the processor to remain stable. Also, while you might think voltage can play a huge role because it's squared, it's very small compared to f, like at least 8-9 orders of magnitude.

 

Also the title should be "silicon", not "silicone", unless we're talking about exploding breast implants.

 

It's a combination of silicon quality and thermals.  The silicon quality part is self-explanatory, but trust me when I say that thermals absolutely play a role in clock speed.  A cooler chip will require less Vcore to maintain the same or higher clock speed.  

 

Vcore requirement drops are very common with delidded chips and even more common with sub ambient cooling. 

[Mira Yurizaki]

Just now, done12many2 said:

It's a combination of silicon quality and thermals.  The silicon quality part is self-explanatory, but trust me when I say that thermals absolutely plays a role in clock speed.  A cooler chip will require less Vcore to maintain the same or higher clock speed.  

 

Vcore requirement drops are very common with delidded chips and even more common with sub ambient cooling. 

Thermals are only related to the fact your power dissipation increased. Just because you can cool your CPU down to -30C doesn't mean it can overclock to 7.0GHz if there's enough defects in the manufacturing process to stop it from doing so.

[done12many2]

Just now, M.Yurizaki said:

Thermals are only related to the fact your power dissipation increased. Just because you can cool your CPU down to -30C doesn't mean it can overclock to 7.0GHz if there's enough defects in the manufacturing process to stop it from doing so.

 

You were intentionally extreme with your example in an attempt to make a point, but the fact remains.  A cooler chip will absolutely require less Vcore then a hotter chip given the same load.  That's a fact. 

[NuclearKing]

1 hour ago, done12many2 said:

 

It's a combination of silicon quality and thermals.  The silicon quality part is self-explanatory, but trust me when I say that thermals absolutely play a role in clock speed.  A cooler chip will require less Vcore to maintain the same or higher clock speed.  

 

Vcore requirement drops are very common with delidded chips and even more common with sub ambient cooling. 

That makes sense, but how much of a difference does it make? Say we go from an air cooler at 4.6 GHz (example: 6600K) at about 65C, and then delid and watercool and the temperature goes to about 35-40C. What kind of voltage decrease can we expect? Lets say it starts at 1.35 on air as that is what mine requires.

[done12many2]

55 minutes ago, NuclearKing said:

That makes sense, but how much of a difference does it make? Say we go from an air cooler at 4.6 GHz (example: 6600K) at about 65C, and then delid and watercool and the temperature goes to about 35-40C. What kind of voltage decrease can we expect? Lets say it starts at 1.35 on air as that is what mine requires.

 

I have binned and delidded a lot of 7700k CPUs so I'll use a 7700k as an example.  

 

During my binning process I run the stock chips in Cinebench R15 starting at 5 GHz up to as high as they will go.  

 

The current chip that I'm using would run R15 at 5.2 GHz / 1.355v (74c) prior to delid and 5.2 / 1.325v (54c) after delid.  

 

At 5.3 GHz with the same test (load), the 7700k would require 1.46v (95c + throttling) prior to delid and 1.37v (64c) after delidding.  Notice how prior to delid, the voltage jump was no longer linear and became exponential, but after delidding, the voltage fell back into line.  Cooling matters.

 

All my binning testing is done with my test bench and a  Corsair H115i AIO cooler an not my full custom loop, which explains the higher then normal temps for a R15 type load.

 

Once I've found a good one, I then add it to the custom loop for further testing and increased cooling (more speed).

 

Silicon leakage and how it is impacted by high temperature increases greatly as you reach the silicon speed limits of a chip.

 

Hopefully that helps.

 

[NuclearKing]

50 minutes ago, done12many2 said:

The current chip that I'm using would run R15 at 5.2 GHz / 1.355v (74c) prior to delid and 5.2 / 1.325v (54c) after delid.  

 

At 5.3 GHz with the same test (load), the 7700k would require 1.46v (95c + throttling) prior to delid and 1.37v (64c) after delidding.  Notice how prior to delid, the voltage jump was no longer linear and became exponential, but after delidding, the voltage fell back into line.  Cooling matters.

That is actually quite significant, I wasnt expecting that big of a difference. I would imagine going from air (in my case) to a custom loop + delid would result in substantially better results.

 

Thanks for the data, it is really quite fascinating.

[done12many2]

Just now, NuclearKing said:

That is actually quite significant, I wasnt expecting that big of a difference. I would imagine going from air (in my case) to a custom loop + delid would result in substantially better results.

 

Thanks for the data, it is really quite fascinating.

 

You are very welcome.  By the way, I've done this dozens of times with similar results.  Some are better and some are worse, but the the one common thing is they all drop.  

 

The closer you are to the silicon speed limit of the chip, the more drop you'll see.  

 

Good luck.  

[mr moose]

5 hours ago, M.Yurizaki said:

Thermals are only related to the fact your power dissipation increased. Just because you can cool your CPU down to -30C doesn't mean it can overclock to 7.0GHz if there's enough defects in the manufacturing process to stop it from doing so.

 

I get what you are saying, but thermals are a huge part of overclocking and can't really be separated from the equation when dealing with silicon yield quality.  Yes some chips will just not overclock regardless of how well you cool them.  But that does not change the fact that the minute you overclock any chip it will produce more heat.

[WeeemRCB]

Thanks all for your replies to this thread so far.

Some great points made and I've learned a lot from it ?