The quest for 24x7 5.5ghz: one man's descent into a world of sex, drugs, and madness (in that order).

[done12many2]

2 hours ago, DildorTheDecent said:

Even that score puts you one point ahead of @done12many2

 

haha  

 

Daily 5.5 would be nuts. 

 

I do what I can on water.  Not everyone can be as fast as you.  

[DildorTheDecent]

5 minutes ago, done12many2 said:

 

I do what I can on water.  Not everyone can be as fast as you.  

Just need it chillier  

[done12many2]

4 minutes ago, DildorTheDecent said:

Just need it chillier  

 

For daily use I don't want the noise. I don't mind it when I'm benchmarking but not everyday for sure. 

 

I'm working on some plans for a permanent and long term solution that will let me have both, but it's going to involve noisy parts in another room with pipes coming into my office. 

 

With that said 5.3 GHz is far from slow.

[Tam3n]

That's insane, good stuff.

 

That pressure film is something I have never heard of before, looks like it worked great for this.

[AnonymousGuy]

3/21 Status update:

 

I've spent the past few days working on getting the chip stable at 77F water, which is where it's actually going to live 24/7.  Of course I *can* run chilled water 24/7 but especially during summer electricity pricing the rent is too damn high!

 

The chip seems happy (passing 8 hours of RB stress test) at:

• 5.3 Ghz, 1.45V (1.464V with LLC) vcore.  Auto for the other voltages (IO 0.95V, SA 1.05V)
• Max temperature: 76C during RealBench
• 5.0 Ghz uncore
• 4133 Memory 18-19-19-39, but there's some benchmarking discrepancies going on so I need to verify that it isn't pulling secondary timings way down to make it stable on boot.

 

The Gigabyte board has been really good at recovering itself from failed overclocks.  It sometimes will shit itself and completely forget all the profiles I saved if it fails hard, but complete system shutdown + restart gets it back.

 

Next up I need to test whether the backside waterblock actually does anything or not, as well as replacing a barb that isn't sealing well to the block (EK, in their usual sucking-a-dick fashion didn't tap the threads on the VRM waterblock all the way through).  I also need to try dialing the vcore back a tiny bit (it crashes at 1.425 and passes at 1.45, so the limit is somewhere between there)

[AnonymousGuy]

Update: the watercooled backplate on the CPU is completely fucking worthless.  I couldn't even find a single degree difference in max core temperatures after running 8 hours RealBench.  Water temperature was locked to 77-78F degrees with chillers absorbing everything. I was very surprised that even after reapplying CLU (had to disassemble everything to get the backside waterblock off), the max temperatures were bang on again.  Score 1 for direct die being consistent as fuck.

 

5.3Ghz at around 1.475V is 24/7 stress test stable with max temps around 80C survived back to back 8 hour runs of RB and 12 hours of Aida64.  Running XMP profile on the memory for 3866.  

[AnonymousGuy]

I'm back!

 

5.3Ghz, 100Mhz AVX offset, 1.48V, 47k cache multiplier, XMP memory.

 

Linx MKL with all the AVX fixings, 30GB problem size:

   96 degrees on the hottest core, no thermal throttling (disabled in BIOS)

   335W from the wall (system includes a low end discrete GPU).

 

Casually doing laundry while running.  It's a multi-purpose room!

 

[done12many2]

Big nuts running LinX MKL on that beast at that voltage/clock speed!

 

Congrats.

[AnonymousGuy]

Next update coming soon:

 

Ordered a tube of Thermal Grizzly Conductonaut which claims to have about double the thermal conductivity of CLU (something like 75 for TG vs 35 for CLU vs 8 for white paste).  We'll see.  This will be the third time I've taken the waterblock off the die.  First was checking mounting pressure, second was to remove the backside waterblock, third will be with Conductonaut.

 

Direct die is the best case scenario for thermal paste to make a difference.  Long story made short, if your IHS is 99% of the distance between your cold plate and your heat source, it doesn't really matter how much you improve the thermal conductivity of the 1% which is your thermal paste layer.  This is why you won't see much of a difference replacing the Intel TIM with TG instead of CLU.  With direct die, 100% of the distance is the thermal paste so it's 1:1 improvement.

 

However, at this level of gameplay you have to realize that heat needs to migrate through the silicon too to even get to the thermal paste, so it's probably not going to be 1:1 improvement.