Help: High Temps on New Liquid Cooling Loop

[spiceygas]

I just built a new liquid cooling loop.  At idle, temps are 32C.  When I put load on it, temps instantly spike up.

• Prime95 Smallest FFTs (high heat): 82C
• Prime95 Small FFTs (max heat): 87C

 

Temps jump up instantly, then when I stop the test they instantly drop back to idle ~32C.  I hadn't expected them to spike that high.  Any suggestions what to do?

 

Things I've tried thus far:

• Reseat the CPU block.  4 times.  With different thermal pastes (both from Thermal Grizzly).  And different application methods.
• Flipped the block upside-down.  It's mounted symmetrically, so I figured it's worth trying.

 

Other things of note:

• 2080 TI is liquid cooled.  Idles at 27C.  During load it slowly rises to ~48C.
• Chipset is liquid cooled.  Idles at 30C.
• Case is currently open during testing, so no restriction on airflow (at radiator or anywhere else).
• Fans & Pumps are running at maximum during all the above tests.
• Liquid temp is sitting between 26-28C.  It really doesn't move much.
• Flowmeter is moving well.  Pretty sure there's no kink or cutoff in flow.
• Temps measured in Ryzen Master.
• The water is definitely flowing in the right direction through the CPU block.
• I'm using the most recent BIOS (v1.80).

 

Hardware:

• ASRock Taichi x570
• Ryzen 3900x
  • EKWB Velocity Waterblock
• Asus 2080 TI OC
  • EKWB Strix Waterblock
• 560mm EKWB CE Radiator in push/pull (8x fans, EK Vardar static pressure fans)
• 2x D5 pumps.
• 32 GB Corsair RAM

[nick name]

You could try rotating the water block or cold plate (upside down will orient it more or less the same you need a 90* rotate).  Other than that it looks more or less normal for the 3900X from all I've seen.  

[spiceygas]

1 minute ago, nick name said:

You could try rotating the water block or cold plate (upside down will orient it more or less the same you need a 90* rotate).

The mounts for the CPU block have 180* symmetry, but not 90*.  So a 90* rotation is not possible.

 

Quote

Other than that it looks more or less normal for the 3900X from all I've seen.  

Really?  I assumed that temps would get that high under load?  Maybe I was just naive, but I had expected load temps to be more like 50-60C.

 

Is the issue just that heat can't transfer from the block to the water fast enough?

[nick name]

4 minutes ago, spiceygas said:

The mounts for the CPU block have 180* symmetry, but not 90*.  So a 90* rotation is not possible.

 

Really?  I assumed that temps would get that high under load?  Maybe I was just naive, but I had expected load temps to be more like 50-60C.

 

Is the issue just that heat can't transfer from the block to the water fast enough?

Then can you rotate the cold plate 90* degreses?  It's a theory some folks have had, but I don't know how well it will actually work.  

 

And yeah, heat dissipation is much harder on a 7nm chiplet design.  

[spiceygas]

Just now, nick name said:

Then can you rotate the cold plate 90*C?  It's a theory some folks have had, but I don't know how well it will actually work.  

 

And yeah, heat dissipation is much harder on a 7nm chiplet design.  

No, rotating 90* is not possible.  Only a 180* rotation, which had no effect.

 

(I didn't expect 180* to matter, but was running out of ideas on what to try)

[nick name]

Just now, spiceygas said:

No, rotating 90* is not possible.  Only a 180* rotation, which had no effect.

 

(I didn't expect 180* to matter, but was running out of ideas on what to try)

You can't pull the cold plate out of the block to rotate it?

[nick name]

It's a square block . . . you can't rotate it 90 degrees?  Or are you tubes hard line?

[spiceygas]

Just now, nick name said:

It's a square block . . . you can't rotate it 90 degrees?  Or are you tubes hard line?

Soft tubes.

 

Wouldn't disassembling the coldplate from the block leak fluid?  I'm not excited to drain the whole loop.

[nick name]

Just now, spiceygas said:

Soft tubes.

 

Wouldn't disassembling the coldplate from the block leak fluid?  I'm not excited to drain the whole loop.

Well the person that claimed a huge temp drop didn't seem to account for ambient, but it did appear to reduce temps a decent amount.  

 

So I guess it's just how much you wanna try to maybe see lower temps.  Because if you can't just rotate the block 90 degrees than fussing with the cold plate is definitely gonna take some work and care.  

[nick name]

Sorry, but it seems odd that you can't rotate your block 90 degrees.  The mount should allow for that.  

 

Edit: 

 

Disregard.  I meant to say the block looks like it should account for it, but looking again it doesn't.  

[Tristerin]

What is your ambient temp?  OC with high voltage?

[spiceygas]

1 minute ago, Tristerin said:

What is your ambient temp?  OC with high voltage?

24C ambient.

 

All settings and voltages in BIOS are stock.

[Tristerin]

1 minute ago, spiceygas said:

24C ambient.

 

All settings and voltages in BIOS are stock.

Ive read and watched videos a couple weeks ago about how the voltages in certain boards have to be tinkered with.  Link to read example:

 

https://www.techpowerup.com/forums/threads/ryzen-9-3900x-voltage-and-temperature.257295/

[spiceygas]

13 hours ago, Tristerin said:

Ive read and watched videos a couple weeks ago about how the voltages in certain boards have to be tinkered with.  Link to read example:

 

https://www.techpowerup.com/forums/threads/ryzen-9-3900x-voltage-and-temperature.257295/

At stock (read from the BIOS):

• CPU VCore Voltage = 1.488
• DRAM Voltage = 1.200
• VDDP Voltage = 0.936
• VPPM Voltage = 2.528
• CPU VD 1.8V = 1.824
• CPU VDDCR_SOC = 1.024
• PREM_VDDCR_SOC = 1.045

 

Offset CPU Voltage -5mV

• Instant crash when I fire up Prime95

Manual CPU Voltage 1.35v

• Verified CPU voltage in BIOS displays as 1.36v.
• No other voltages sizeably changed.  (Minor fluctuations within margin of error)
• Prime95 Smallest FFTs (High Heat) = Instant Crash

Manual CPU Voltage 1.40v

• Prime95 Smallest FFTs (High Heat) = 83*C
  • One degree worse than @ stock
  • Seems within margin of error
• Prime95 Small FFTs (Highest Heat) = 94*C
  • 7*C worse than @ Stock.
  • Probably outside margin of error
  • Ryzen Master has 95*C as the limit, and I assume the CPU is throttling.

 

So I would conclude that lowering voltages substantially causes instability, and lowering a tiny bit does nothing, and possibly hurts thermals.

==============================

 

Just for shiggles, I tried disabling Precision Boost Overdrive in the BIOS.  Fired up Prime95 and... instant crash.

 

Oddly, if I disable it in Ryzen Master then it's fine (see below).  So maybe I suck at using the BIOS...?

==============================

 

Ryzen Master set to "Eco Mode."

• After reboot, it passed the AMD Ryzen Master Stress Test.
• Prime95 Smallest FFTs (High Heat) = 48*C
  • Wow!  Huge difference.  (At a cut to performance, of course)
• Passmark CPU Score = 31145
  • 2.1% below average.  (Link)

Ryzen Master Set to "Default" Mode (i.e. Not Precision Boost Overdrive)

• After reboot, it passed the AMD Ryzen Master Stress Test.
• Prime95 Smallest FFTs (High Heat) = 72*C
  • 10*C improvement over stock
• Passmark CPU Score = 31802
  • 0.08% below average.  (Link)

Ryzen Master Set to Precision Boost Overdrive

• Passmark CPU Score = 32019
  • 0.60% above average.  (Link)

 

===============================

My Conclusions Thus Far

===============================

1. Leave the voltage settings alone in ASRock Taichi x570.  Things become unstable really quickly.
   • Your mileage may vary on other motherboards.
2. If temps are running high, set Ryzen Master to either Eco or Default (i.e. Disable Precision Boost Overdrive), depending on your comfort with losing CPU performance.
3. The difference in performance between Default and Precision Boost Overdrive is very small.

[Tristerin]

1 hour ago, spiceygas said:

-snip-

Nice detail - Hopefully with BIOS updates things get better for this awesomely powerful chip