Memory OC = +70% CPU power usage?

[DreadPirateDuck]

Here are my core system specs first:

 

Ryzen 2700X

Asus Crosshair Hero VII

2x8GB G.Skill FlareX 3200 CL14

EVGA GeForce GTX 1070 FTW3

Corsair H100iV2 with Noctua NF-F12 iPPC 2000 fans in push+pull config

Seasonic Prime Titanium 750w

 

I'm seeing a strange behavior that I just noticed recently when applying my memory overclock.  Since I've had my 2700x (about 14 months) I just ran my memory overclock according to DRAM calculator (picture of Ryzen Timing Checker attached) and set Performance Enhancement to level 2 in the Asus BIOS and called it a day.

 

Recently, I've decided to try to get some extra performance and attempt an all-core overclock since my 2700X runs around 3.8-3.9Ghz with an all core load.  Such as Rainbow Six Siege + OBS stream or Folding@Home.  I wanted to get at least 4.1Ghz all-core since I am almost always running fully loaded which, again, puts me around 3.8Ghz.

 

So, I load into BIOS, load optimized defaults, and then start pushing clocks.  I found that 4.15Ghz puts me in a good place in terms of voltage (1.3375v) and actually lowered my CPU temps significantly (warning sign that I notice later).  During the OCCT burn-in test I noted ~100-110w total CPU package power.

 

Ok.  I got my CPU OC dialed in.  Time to put my memory OC back in.  As you can see in the attached pic, my memory/memory controller can handle 3533 CL14 with fast sub timings + GDM on.  The SOC voltage is 1.0125v, by the way.

 

Time to stress test with both a CPU OC and memory OC.  I load up OCCT, run the same test and.....185w CPU package power!!!!

 

Within 2 minutes my system crashes.

 

Ok.  I know faster memory and tight timings increases CPU performance at the same clock speed.  More performance and a faster Infinity Fabric requires more power, but...70-80% more power?

 

I revert back to my previous config, 3533CL14 + fast subs and PE2.  I run the same OCCT stress test and, sure enough, the power usage is high.  About 160w even though the CPU is running around 3.85Ghz.  My CPU temps are about 17*C higher even though the clock speed is 300Mhz slower and the reported CPU voltage is lower.

 

I load the 4.15Ghz all-core OC + memory OC again and start to dial back the clocks.  Hoping that I can manage some sort of all-core performance boost.  Nope.  I am under 4Ghz before I am stable, almost entirely defeating the purpose of this exercise.

 

Has anyone else seen this behavior before?  Is this something with Ryzen+ and IF?  Maybe this is an Asus CH7 behavior?

[Guest]

So the SOC voltage on auto gets bumped when you OC the memory. That's why the higher package wattage. No big deal.

 

You can pull more than this. I'm not sure where the issue is? I did kinda speed read through this, looked like we where trying to compare some wattage figures at X clock speeds with no mention of voltage used, quite a few details missing. Can't even compare screen shots side by side while we have a few random with no placement. 

Can you do your 4.1ghz OC with the memory at lower speeds, then do the 3850mhz on with the memory OCed but have screen shots for both tests? And can you include some cpu-z and some voltage monitoring or something?? I really think we can get a lot closer to 200w pulled.... Performance Enhancer level 3 is pretty sweet setup, and 4 if you can cool the damn thing. After all it is the Flagship Zen+ and happens to be hot by nature.

 

 

 

 

[DreadPirateDuck]

Please don't "speed read" through my post then state that details are missing.  All the details that were "missing" are there.  Voltages, when not auto, were stated.  The SOC voltage is not auto.  It is set to 1.0125v when my memory OC is applied.  With just a CPU OC at 4.15Ghz, the vcore is 1.3375v.  When not running a CPU OC (PE2), the vcore is auto.

 

The problem is that I can't do both 4.15Ghz all-core AND run my memory at 3533CL14.  Even if I crank up the vcore.  1.4v vcore at just 4.0Ghz with my memory OC is still unstable.  But 1.3375v vcore at 4.15Ghz without a memory OC it is stable.  The power draw with just the CPU OC is ~100-110w.  The power draw with JUST the MEMORY OC and PE set to level 2 is 165w (see OCCT screenshot attached to initial post).  With the memory OC and PE2, the CPU clocks down to about 3.85Ghz during an all-core load.  The reported vcore for PE2 with an all-core load is around 1.26-1.28v.  The power draw with both the CPU and memory OC (before it crashes) is 185w.  I can get a screenshot of that later, but not sure what that will show other than what I've stated here.

 

Again, as I stated in the initial post, I am aware that a memory OC will increase power draw due to the increase frequency of the Infinity Fabric.  But 70-80w more?  You're telling me that my memory OC is causing my 2700X SOC to draw more power than a stock 2700 non-X draws altogether?  Does anyone else have their memory running at 3533CL14?  What is your SOC voltage?

 

I disagree with the 2700X being "hot by nature."  Fully stock (optimized defaults in BIOS), this CPU doesn't get much above ~51*C with the H100iV2 attached.  Heck.  With just the CPU OC at 4.15Ghz and 1.3375v the CPU temps stayed around 63*C.  With just the memory OC and PE2, my CPU runs around 78*C.  And for the short time I can run both the CPU OC and memory OC, temps are in the 80's before it crashes.

 

I COULD run my memory OC at a slower speed but I don't think I should have to.

[DreadPirateDuck]

Here is a screenshot of HWiNFO64 while OCCT is running.  Medium data set, no AVX, memory OC to 3533CL14 with fast subs, Performance Enhancement level 2.  I will post more screenshots in a bit.  I'm waiting for a F@H CPU work unit to finish before I start messing around with settings.

[DreadPirateDuck]

Here are some additional screen shots, plus the one above, with text labels.  BTW, I pulled the CPU OC back to 4.1Ghz since AVX work loads (Blender in this case, see below) pushed 4.15Ghz too far.  F@H also uses AVX.

 

I added CPU and SOC power usage to my HWiNFO view, which weren't present before.  As you can see, just adding the memory OC significantly increases core power usage.  Even with the CPU at optimized defaults with no PBO.  The additional power usage is definitely not coming from the SOC power.

 

I also ran Blender (the BMW demo) to measure the performance difference between the 3 configs.

 

Optimized defaults - 3:54

CPU default, mem OC - 3:49

CPU 4.1Ghz, mem default - 3:41

CPU PE2 + PBO, mem OC -  3:41  (CPU ran around 3.9Ghz when loaded)

 

I'm going to try fiddling with LLC to see if I can get 4.1Ghz stable with the memory OC.