I ended up with the best temps using a Rockitcool IHS with LM on my 7960x
1.22v 4.8ghz 68-72c under blender/cinebench runs with a lapped ek supremacy.
 
I had no luck going direct die even with a lapped block and i tried 3 different waterblocks from heatkiller and ek.  The die on my 7960x is very convex and i would always have a huge delta between the cores when delidded with no ihs no matter what kind of LM application. With the rockitcool ihs i see about 5-7c between the cores, i went through about 4g of lm testing many thermal solutions. If anyone wants a iceman delid die gaurd or a derbau8r die guard i will let them go cheap, the Iceman unit is actually much nicer and retains the stock backplate.

DEJ915

boinger
Update. 
 
I was getting frustrated with the IHS. So I ended up just doing direct die. 
 
I got the derbuar direct die guard and set it all up. Strange no improvement in temps. In fact worse than ihs and when ihs was not cleaned.
 
I am really suspecting the waterblock is not making good contact or something is getting in the way. 
 
I am going to try a new waterblock and see how that goes. 

Before you buy new waterblock you could try to use some washers to increase mounting pressure.

I did try that didn't help 
 
I don't have pictures of my liquid metal application as I have already removed it. 
 
I did end up swapping the water block to a ek-supremacy evo and most of the temps are back within reasonable range.
 
Except I have 3 turd cores that just refuse to clock high at all. I can run 11 at 49x but 3 at 42x and even then they overheat on a stress test. 
 
I really thought direct die would help. But it has not.

are you applying liquid metal to your block and die? do you have a screenshot of hwinfo64 showing the hot cores? if you have some cores overheating and some cores at reasonable temps then that is an application or contact issue, take the block off and use some regular thermal paste reinstall/ pull the block back off and look at the imprint, you can also try rotating the block. if you are delidded and have cores hitting tjmax it sounds like not enough liquid metal or it wasn't applied to both surfaces, even with terrible contact you shouldn't be getting anywhere close to tjmax on any core once delidded.

jollydet
are you applying liquid metal to your block and die? do you have a screenshot of hwinfo64 showing the hot cores? if you have some cores overheating and some cores at reasonable temps then that is an application or contact issue, take the block off and use some regular thermal paste reinstall/ pull the block back off and look at the imprint, you can also try rotating the block. if you are delidded and have cores hitting tjmax it sounds like not enough liquid metal or it wasn't applied to both surfaces, even with terrible contact you shouldn't be getting anywhere close to tjmax on any core once delidded.

I am delidded and direct die, thermal grizzly on die. Take a look at the screenshot I have attached and the voltages. for the respective turd cores are highlighted in red. 
 
Look at the max temps as those are what it reaches consistently on a stress test will go higher but it crashes. 
 

did you apply lm to the block and the die? you either dont have enough LM or it wasnt applied to both surfaces.
 
also how are you applying your overclocks your core voltages are all over the place and those hot cores never moved past 4200mhz which would indicate a bios setting even with a 92c peak they should have gone over 4200 at some point.
 

I am not using liquid metal since I have no IHS. I am just using thermal grizzly tim. I just applied it on die only. 
 
I am using adaptive voltage of 1.25 with an +75mv offset. As using override just overheats the three hot cores instantly and occt stops. 
 
I have the multipliers set to 42x for the hot cores. Adaptive voltages allow them to run at a lower voltage for the respective oc. 

If you have no IHS you certainly should be using liquid metal otherwise there is zero point and you will have close to or worse temps than stock.
 
Whats happening is your die is convex and there is portions of the die not touching the block.  When you use LM this issue is not as significant. Rotate the block and see if you get better temps however i suspect you will not.
 
At this point you need to use some conducatnaut on the die and the block. Your bios settings are trying to compensate for a poor thermal bond which is not ideal. Also just set Adaptage voltage with no offset or a static voltage.
 
TJmax will not protect your cpu from degredation when you are pulling 400 watts through it at such high temps and voltages which huge temp discrepancy's across the die. pull the voltages and clocks back until you get the thermal solution sorted as you do not want to degrade your now out out of warranty chip.
 
 

I have ordered some more liquid metal and waiting for delivery before testing. I figured Tim would be enough on direct die. 
 
I am not loading up the chip constantly that was just on stress tests and I have it set to immediately cancel if it exceeds 100C The Tjmax is 102 for this chip I think. 
 
Anyhow I have lowered the clock to 4500mhz and 1.22v override. 
 
I really wish that evga had the ability to enter per core voltages. 

With LM you need to be putting it on the contact area on both the chip and cooler (doesn't matter if you are using the direct frame or with the IHS put it on both) - if you were doing it on just the chip then that is part of the problem. Watch Buildzoid's video I linked above

also make sure you put some clear nail polish or some liquid electrical tape on the capacitors on the chip so no lm can get on them accidentally