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

[AnonymousGuy]

Hey, I have a flair for dramatic and over the top titles.

 

I've done my best over the past several years to push conventional and subambient 24/7 cooling as far as it can go.  Water chillers, 25 gallons of water, 125 feet of hose, 4 pumps, a broken marriage, dead hookers...all of this pretty much sets my status as knowing what the fuck I'm doing.  So hey it turns out the 7700k is really good at overclocking, and I happen to have a cherry picked one that can handle 5.2ghz with a average AIO.  Plus there's going to be a new Optane SSD coming out shortly, so that leads me to conclude it's a great time for a new build and to see how far this thing can get pushed.  I don't care what happens.  I don't care if we kill someone.

 

Desktop Parts:

Case: Corsair 450D

Power Supply: EVGA Supernova G3 750W (eventually the AX1500i from my Workstation).  

Ram: G.Skill TridentZ DDR4-3866 4x8GB.  Is this a waste of money for practical usage? Yeah, but we're trying to win some benchmarks here!

CPU: 7700k, pre binned at 5.2Ghz.

Motherboard: Gigabyte Aurus Gaming 9 Z270.  TBH I sorted price high to low and this one was #1, had a PLX chip, and 22 phases of CPU power is redonkulous.  Don't give a fuck about the RGB but oh well it has it too.

Video Card: NV GT 210?...who cares not the point.  Eventually 2xTitan X maxwell's.

Storage: Samsung 840 Evo, to be replaced with Intel 750 series successor.

 

Game Plan (the cooling):

• Direct Die Cooling.  Aquacomputer Skylake Spacer (probably hopefully Kaby Lake compatible).  Coolaboratory Liquid Ultra TIM.
• EK True 115X backplate.
• Aquacomputer kyros NEXT VARIO VISION .925 silver.  This is their top of the range flagship block.  It looks baller as fuck, and it's got awesome performance.  And the vario part makes EK look like a bunch of retards with their thumbs up their asses.  Oh and just to top it off, fucking sterling silver cold plate for 10% more thermal conductivity!!!1!!1  Boy let's hope CLU doesn't wreck this things asshole.
• Watercool Heatkiller IV pro.  This block is going to be substituting for the kryos while I'm waiting for delivery.  First world problem: I have to use a IV Pro while I wait for my $300  waterblock.  Smallest violin....
• Phobya UC-1 Extreme.  This block will be replaced by the Watercool block, serving the dude of chilling behind the CPU socket.  Yes, friends, that's right.  We're going to be making a waterblock sandwich around the CPU and watercooling the back of the PCB, just to spite Mr. Linus and his "does it make a difference" test.
• Integrated EK G-Frost VRM waterblock to the Gigabyte Gaming 9.
• Hailea HC-500 + HC-350 water chillers.
• 5 gallons of water, 1L of Mayhem X1.
• There's a bunch of nuance to everything that I'll post about as I go.

Where are we at:

Motherboard delivered.  Waiting on case, power supply, ram, cpu, and waterblocks.  All in flight.

Waiting on the hardware needed for making a waterblock sandwich happens.  Basically waiting on some M4 threaded 80mm long screws.

Various shit needs to happens to get a 5 gallon chilled water loop set up.  Boring details, but there will be pictures.

 

 

Update 2/20:

• Converting a 5 gallon bucket to an insulated 5 gallon bucket.  1.5" thick polyiso something or other reflective insulation.  Wood frame.  Pretty much have to do this to reach 5C water temperatures.
• The pump is sitting in the bottom of the bucket.  EcoPlus 1267gph submersible pump.  $70 on Amazon.  Works great for this application but it's not so good for 24x7 usage.

 

2/21 Update (and in the OP), we've reached 45F/7.2C, which is as low as the chiller thermostats can go.  Going lower is possible but there's not much point since the build will just turn into a rain forest of condensation.  Ambient: 80F, 25% humidity.

 

Here it is with the lid on, all buttoned up and ready for prom.  It *has* to be insulated like a fucking polar bear, or else it won't work.  Every seam is sealed with aluminum foil tape, the top lid is gasketed with foam weatherstripping and held down tight with velcro straps.  Even with 1.5" of insulation on the bottom, it's chilling the tile floor!:

 

 

This is just the chilled water portion handled.   Next up we need to plumb in the heat exchanger to be able to chill the build itself.  More pumps, more fititngs!

 

2/22 update: got the backside waterblock set up and mounted (hehehe...mounted...) 

 

EK True backplate is a nice piece of zinc plated flat steel about 1/16" thick.  I needed a flat backplate for better thermal contact.  Most waterblocks "cheap out" and use thin sheet metal folded and rippled to give it strength.

 

Phobya UC-1 Extreme is a good waterblock quality-wise for the price ($40).  It won't be around for the "final" configuration but it'll be a good stand in.

 

3M thermal tape 25mil (i think) is holding the backplate to the mobo.  It was a bitch to figure out how to cut the stuff and apply it.  It's very sticky and stretchy (hehehe...sticky) so it doesn't allow much room for fuckups. 

 

I stacked .5mm + 1mm thick thermal pads on the capacitors to build it up to the height of the backplate, then put .5mm thermal pad on all of it.

 

I need my 70mm M4 screws to be able to cover the whole sandwich of backside and frontside waterblocks.  55mm was just a little bit too short.

 

The case will not be able to close with the backside waterblock installed.  Not really a huge deal, but it sticks out about 1/4" to 1/2" too much.

 

 

03/02/2017 Picture Update:

 

 

Picture Update:

 

Socket Before Being Molested (you have to shave the corners down because they're taller than the rest of the socket.  I used a Dremel with a carbide burr 5k RPM, you have to vacuum + blow out the socket because the plastic gets flung in small particles all over.  No big deal.):

 

 

How the chip fits with the corner shaved.  This was the first attempt. 

 

 

Tape installed (provided with the spacer to add .1mm of thickness to the .3mm thickness of the spacer.  .4mm is the height of the die off the PCB (verified it myself with calipers accurate to .01mm).

 

 

Spacer installed.

 

 

Very ghetto set up, but we're aiming for 5.5Ghz this setup won't be permanent anyways.

 

Temperatures not so bueno so I took everything apart and here's a series of photos of the contact patch.  It looks decent, but the center seems to be getting more pressure:

 

 

I touched up the edges of the socket even further so the waterblock shouldn't touch them at all.  It'll touch the other areas first.

 

This actually worked incredibly well for blowing coolant out of the loop so I could take it apart.  Blow gun + wrapped in tape + 3/8" to 1/2" adapter.

 

 

 

I'm using a set of precise digital calipers to measure how far the top of the 4 corners of the waterblock sit above the mobo.  By doing this with no chip installed, I can tell the z-height where the waterblock is "bottoming out" on the socket itself.  Then when I install the chip, I can crank the nuts down until I hit that same height.  Very smart and precise way of doing this without fucking up.  I'm turning the nuts 1 revolution at a time in a criss cross pattern.

 

Next update will be tomorrow with very precise pressure sensitive film pictures.  I will see if my corner touchup has improved the situation and how much clamping force I'm getting.

 

Update 3/2,

 

I apparently had to do more engineering legwork than Aquacomputer did for their spacer.  They need to deburr the top of the spacer too to get rid of the burrs from the laser cutting process.  They only did the bottom, probably to keep the top "looking nice".   The tape they provided is too fucking thick.  Yeah it technically is spacer + tape = exact die height, but the tape and spacer metal have tolerances on them.  Cut the tape thickness in half and holy shit get good die contact.

 

This Fujifilm pressure film is fucking awesome.  It costs $80 per square foot, so it's not for poor people, but that gives me about 40 attempts and incredible visibility into what the fuck is going on between the chip and the waterblock.

 

 

Here's a bonus photo of what the first attempt (far left of the above pic) looked like on top of the package:

 

 

Update 5/9: 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.

 

5.3Ghz at around 1.475V is 24/7 stress test stable with max temps around 80C.  Running XMP profile on the memory for 3866.  

 

Also probably the last update.  In conclusion, 24/7 5.5Ghz is plausible but even with this cherry picked CPU it's not easy to get to and pretty much impossible without chilled water.  5.4Ghz if you get a unicorn can be done 24/7 room temperature.  5.3 requires a cherry picked 90th percentile CPU.

[Droidbot]

holy shit boys he's gone insane

[steffeeh]

2x Titan X Maxwells? Why not the Pascals, and shove in that sweet new 4k 144Hz display while you're at it.

[AnonymousGuy]

4 minutes ago, steffeeh said:

2x Titan X Maxwells? Why not the Pascals, and shove in that sweet new 4k 144Hz display while you're at it.

Cause NV can suck a dick after the shit they pulled on me with the 980Ti.  Oh so sorry you paid 2 grand so we can fuck you in the ass and release something at 3/4 the cost with more performance and custom PCBs....3 months later.

 

1080Ti's are what I'm waiting for.

[PrimeSonic]

This is gonna be a wild ride.

[NoobCase]

9 hours ago, AnonymousGuy said:

Cause NV can suck a dick after the shit they pulled on me with the 980Ti.  Oh so sorry you paid 2 grand so we can fuck you in the ass and release something at 3/4 the cost with more performance and custom PCBs....3 months later.

 

1080Ti's are what I'm waiting for.

Oi....  So many first world problems in one thread.....!  Count  me in for the ride!

 

Bonus question: do you happen  to know if there's much of a performance difference on the CLU and the thermal grizzly kryonaut? 

 

And Hey...  Seen several benchmarks show a single titan x (m)  betingelser the 980tis in 4k benchmarks and due to the 12 GB v ram...  That was my comfort....

 

Not that it helped much...

[TidaLWaveZ]

I know a guy who chased a 5.5ghz stable 24/7 overclock. Everything seemed to be going his way, he had everything going for him, his whole life ahead of him.

 

Now he sleeps on a bench in the park and talks to pigeons.

[done12many2]

@AnonymousGuy

 

I like your style bud.  I know a guy running 5.3 GHz on ambient all day long.  

 

Hell of a project and I'm looking forward to seeing more of it.

[AnonymousGuy]

Added "Update 2/20" to the OP.

[AnonymousGuy]

2/21 Update (and in the OP), we've reached 45F/7.2C, which is as low as the chiller thermostats can go.  Going lower is possible but there's not much point since the build will just turn into a rain forest of condensation.  Ambient: 80F, 25% humidity.

 

Here it is with the lid on, all buttoned up and ready for prom.  It *has* to be insulated like a fucking polar bear, or else it won't work.  Every seam is sealed with aluminum foil tape, the top lid is gasketed with foam weatherstripping and held down tight with velcro straps.  Even with 1.5" of insulation on the bottom, it's chilling the tile floor!:

 

 

This is just the chilled water portion handled.   Next up we need to plumb in the heat exchanger to be able to chill the build itself.  More pumps, more fititngs!

[AnonymousGuy]

2/22 update: got the backside waterblock set up and mounted (hehehe...mounted...) 

 

EK True backplate is a nice piece of zinc plated flat steel about 1/16" thick.  I needed a flat backplate for better thermal contact.  Most waterblocks "cheap out" and use thin sheet metal folded and rippled to give it strength.

 

Phobya UC-1 Extreme is a good waterblock quality-wise for the price ($40).  It won't be around for the "final" configuration but it'll be a good stand in.

 

3M thermal tape 25mil (i think) is holding the backplate to the mobo.  It was a bitch to figure out how to cut the stuff and apply it.  It's very sticky and stretchy (hehehe...sticky) so it doesn't allow much room for fuckups. 

 

I stacked .5mm + 1mm thick thermal pads on the capacitors to build it up to the height of the backplate, then put .5mm thermal pad on all of it.

 

I need my 70mm M4 screws to be able to cover the whole sandwich of backside and frontside waterblocks.  55mm was just a little bit too short.

 

The case will not be able to close with the backside waterblock installed.  Not really a huge deal, but it sticks out about 1/4" to 1/2" too much.

 

 

Next update may not come until Tuesday, but things are going to go very fast where it's wham bam and everything put together.  You'll get pictures of the CPU being installed, the full assembly of frontside and backside waterblocks, plumbing the frontside, backside, and VRM waterblocks together, getting it all in the case, etc.

[done12many2]

Shit's getting very real in this thread.  Looks good man.

[AnonymousGuy]

2/28 Update (more like a status tweet than a full update).  

 

Everything is together, POSTed first try with no drama.  Windows 10 and drivers installed.  Pictures will follow later tonight.

 

Temperatures are a bit weird.  I think it's not happy with thermal paste instead of liquid metal.  Spikes up to 90C but in Realbench it runs 60C under full load.

[done12many2]

2 minutes ago, AnonymousGuy said:

Temperatures are a bit weird.  I think it's not happy with thermal paste instead of liquid metal.  Spikes up to 90C but in Realbench it runs 60C under full load.

 

At 5.5 GHz?  If so, that's nothing to sneeze at.  

[PrimeSonic]

How much voltage are you pumping into that CPU?

[DildorTheDecent]

The amount i spent on watercooling i could probably have got a chiller

 

Looks fast as fucc tho.5.5%GHz dream is being wished for on this end. 

[dexT]

On 2/22/2017 at 11:02 AM, AnonymousGuy said:

 

 

[AnonymousGuy]

2 hours ago, PrimeSonic said:

How much voltage are you pumping into that CPU?

Still too preliminary to start really pumping for the finish line.  I think I have mount issues going on with the socket and the waterblock (fuck me direct die adds a lot of pain in the ass), so I'm going to be attempting to clean that up first in the coming days before I start getting crazy.

 

Consulting with done12many2, my temps are way too high for the setup I'm running.  They're not shit temps, but they should definitely be 15C+ lower.  I've got some precision pressure recording film on the way so I'm going to see exactly where the interference is coming from.

[AnonymousGuy]

Picture Update:

 

Socket Before Being Molested (you have to shave the corners down because they're taller than the rest of the socket.  I used a Dremel with a carbide burr 5k RPM, you have to vacuum + blow out the socket because the plastic gets flung in small particles all over.  No big deal.):

 

 

How the chip fits with the corner shaved.  This was the first attempt. 

 

 

Tape installed (provided with the spacer to add .1mm of thickness to the .3mm thickness of the spacer.  .4mm is the height of the die off the PCB (verified it myself with calipers accurate to .01mm).

 

 

Spacer installed.

 

 

Very ghetto set up, but we're aiming for 5.5Ghz this setup won't be permanent anyways.

 

Temperatures not so bueno so I took everything apart and here's a series of photos of the contact patch.  It looks decent, but the center seems to be getting more pressure:

 

Oops, condensation.

 

 

I touched up the edges of the socket even further so the waterblock shouldn't touch them at all.  It'll touch the other areas first.

 

This actually worked incredibly well for blowing coolant out of the loop so I could take it apart.  Blow gun + wrapped in tape + 3/8" to 1/2" adapter.

 

 

 

I'm using a set of precise digital calipers to measure how far the top of the 4 corners of the waterblock sit above the mobo.  By doing this with no chip installed, I can tell the z-height where the waterblock is "bottoming out" on the socket itself.  Then when I install the chip, I can crank the nuts down until I hit that same height.  Very smart and precise way of doing this without fucking up.  I'm turning the nuts 1 revolution at a time in a criss cross pattern.

 

Next update will be tomorrow with very precise pressure sensitive film pictures.  I will see if my corner touchup has improved the situation and how much clamping force I'm getting.

[Droidbot]

That's fucking beautiful. The delidded die is so small it's amazing. Also I love that Thinkpad keyboard. 

[AnonymousGuy]

Update 3/2, I apparently had to do more engineering legwork than Aquacomputer did for their spacer.  They need to deburr the top of the spacer too to get rid of the burrs from the laser cutting process.  They only did the bottom, probably to keep the top "looking nice".   The tape they provided is too fucking thick.  Yeah it technically is spacer + tape = exact die height, but the tape and spacer metal have tolerances on them.  Cut the tape thickness in half and holy shit get good die contact.

 

This Fujifilm pressure film is fucking awesome.  It costs $80 per square foot, so it's not for poor people, but that gives me about 40 attempts and incredible visibility into what the fuck is going on between the chip and the waterblock.

 

 

Here's a bonus photo of what the first attempt (far left of the above pic) looked like on top of the package:

 

[AnonymousGuy]

3/3 Update:

 

We did it reddit LTT!  5.5Ghz is Cinebench stable at 1.6V.  5.4Ghz survived Realbench for 15 minutes @ 1.5V.  I tried 5.6Ghz 1.725V (yep, I went balls deep on this thing and YES I KNOW THAT'S A CRAZY HIGH VOLTAGE), but couldn't get past the login screen without a BSOD.  Pics will follow later tonight.

 

Temperatures are incredibly good.  1.5V 60C max core temp in Realbench @ 5.4Ghz.

 

I couldn't get 5.5Ghz Realbench stable even at 1.675V, but it was 3AM and I didn't have time to turn all the knobs of XMP and uncore and shit to try to make it work.

[DildorTheDecent]

1 hour ago, AnonymousGuy said:

3/3 Update:

 

We did it reddit LTT!  5.5Ghz is Cinebench stable at 1.6V.  5.4Ghz survived Realbench for 15 minutes @ 1.5V.  I tried 5.6Ghz 1.725V (yep, I went balls deep on this thing and YES I KNOW THAT'S A CRAZY HIGH VOLTAGE), but couldn't get past the login screen without a BSOD.  Pics will follow later tonight.

 

Temperatures are incredibly good.  1.5V 60C max core temp in Realbench @ 5.4Ghz.

 

I couldn't get 5.5Ghz Realbench stable even at 1.675V, but it was 3AM and I didn't have time to turn all the knobs of XMP and uncore and shit to try to make it work.

nice. 5.5GHz is pretty tasty. 

 

You get a good CB result for: ?

 

[AnonymousGuy]

1 hour ago, DildorTheDecent said:

nice. 5.5GHz is pretty tasty. 

 

You get a good CB result for: ?

 

The pic I have of CB is 1219 score.  Not sure what the run conditions were on it, but last night was just a quickie for getting everything up and running and laying down some tracks.  This weekend I can start getting stuff dialed in.

[DildorTheDecent]

1 hour ago, AnonymousGuy said:

The pic I have of CB is 1219 score.  Not sure what the run conditions were on it, but last night was just a quickie for getting everything up and running and laying down some tracks.  This weekend I can start getting stuff dialed in.

Even that score puts you one point ahead of @done12many2

 

haha  

 

Daily 5.5 would be nuts.