Going topless or not?

[Kenrou]

HT was turned off as it doesn't benefit this task.

HT off will run 5c-15c cooler by default depending on voltage used, my 4.7ghz 1.38v 85c runs 4.8ghz 1.38v 70c. Or those specific instruction set bypass that ?

 

[mackerel]

Prime95 and similar tasks like the one I'm doing are highly optimised for the resources available. There's nothing left for HT to give. In the example above, what workload is that under?

 

[Kenrou]

Prime95 small FFT standard

 

[mackerel]

Ooh! Guess I'll have to try that. I optimise for performance (throughput) and basically HT doesn't improve things there, and you do twice the work in twice the time. So I never really looked at that as a realistic working condition.

 

[mackerel]

Just done a test... 54C HT on, 52C effectively HT off. 6700k @ 4.2 1.20v measured under load. Water cooled. I selected custom test, 64k FFT, 4 or 8 workers, in-place FFTs so it wont hit ram. I left HT on at all times but manually set affinity for 4 workers. So it does look like there's HT overhead even when it isn't productive.

 

[mackerel]

Things are going bad again... I was running real tasks equivalent to Prime95 small FFT, and found the system reboot. On starting it again I kept an eye on the temps and I have a hot core come back. They were hitting 78, 68, 94, 73. Could the IHS/heatsink have shifted or something? I've put it offline now and will take it apart tomorrow. As my VR system which I also use to take to friends places, I'm tempted to swap it with my 6600k and leave this in another less critical desktop.

 

[mackerel]

I've extracted the CPU. The thermal compound pattern seemed ok between IHS and HSF. The IHS fell off the CPU while I was working how how to unstick it, and the pattern there seemed to suggest a covering also. Is there such a thing as a better or worse compound in this application? For example, if this one is too thin did it get pumped out? I used the Noctua one as it is supposed to be one of the better ones. Not sure what else I have.

Or could something else have shifted? I'll clean off the glue for now and will skip it next time I think.

 

[Nebulous]

Are you trying to reglue the IHS back on after the delidding? When I delidded my chip I didn't reglue it back on. I used the same Noctua thermal paste and then just placed the IHS back on right before I locked the latch down. I held the IHS inplace with my finger tips so it was centered, then locked it down.

That's a little too much grease. Just clean off the edges around the die and slap that puppy back on.

 

[mackerel]

I had attempted to glue it back with superglue, but it didn't seem to set after some hours and I got bored and put the CPU back in anyway.

Anyway, after some discussion elsewhere, I found my old tube of Arctic MX-2 and have now applied that between die and IHS. On IHS-HSF I've kept the Noctua stuff. Running Prime95 small FFT on it now. Seeing less than 5C difference between cores, and it is a different pair of cores than before.

Edit: after about 10 mins, hottest temps for each core are: 72, 69, 70, 68C.

 

[trents]

Super glue breaks down PCB I understand. Use automotive RTV Black Ultra Silicone if you want to secure the lid back to the PCB.

 

[mackerel]

Superglue works by reacting to the surfaces, but there's a load of coating on the substrate before you get to the tracks.

Forgot to say, this time around I didn't bother gluing it and just left it to be held in place by the socket mechanism. After a total of 26 minutes when I stopped Prime95, the 3rd core had hit a max one degree higher than in the previous post.

I'm running small compute tasks through it again now (similar to Prime95 small FFT). It's bed time for me now, and if all is ok in the morning I'll move it to large FFT which is less stressful anyway as it gets somewhat ram bottlenecked.

I'm half wondering, if there is a non-OC Kaby Lake that'll hit similar clocks (around 4 GHz all cores active). I'm thinking I could move this 6700k to a dedicated OC box then, and not have to worry about my VR box.