SOLVED Would like feedback on overclocked 8700k for system with intense workloads

[ceslen]

****Problem found and updated on post #5

I'm looking for feedback on my overclock for intense workloads.
System is a Hackintosh
8700k 1.335v 4.8ghz 0AVXoffset
Memory is overclocked to 3200mhz via xmp
Processor is delidded with liquid metal, kryonaut thermal paste on ihs, ek custom loop with supremacy block. Currently using a 240mm ek coolstream pe radiator with fans in push/pull.
I have removed and replaced the thermal paste/liquid metal multiple times to make sure it is applied properly and temps are always consistent.
Radiator is located at the top of the case with fans exhausting out.
The gpu is a gtx1080 that is also in the same loop, not overclocked.
Package power usage on the 8700k is 124w and the 1080 per Nvidia's website says 180w.
EK website lists the 240mm PE as able to cool 350w/10c @ 1800rpm (@ amb room temp of 25c) with it's Vardar fans in a push configuration.
Running full load cpu test (Mac yes test in terminal) with all cores/threads pegged and taxing the gpu via heaven at the same time, max temps are 82c on the cpu in the loop after temps stabilize.

If I added a second 240mm radiator, what kind of decrease could I see in temperature approximately and would it even be worth it as long as my temps while under full load are at 82c?
I've read many posts online stating the throttle temp for the 8700k is 100c. Is 82c still fairly safe for the system to run at for hours or possibly days with this load?

I am editing these numbers as I run Prime95 for stability/Torture Test small FFTs*
I am also open to any tips or suggestions that I might be overlooking.
Thanks in advance!

 

[trents]

82c is perfectly fine for the CPU. What is the GPU temp running?

Adding a second radiator of the same caliber I would guess would reduce temps 5-10c.

As long as the GPU temps are within safe/sub throttle range I would leave well enough alone. Unless, that is seasonal variation in ambient temps will change the equation significantly.

Tell us about your case and fans. This sounds like a high end system so I assume the case is a high end unit with good air flow.


What is the purpose for this machine? How much RAM are you running?

 

[ceslen]

Thank you for the response.
GPU temps are lowish (mid 50s to 60s range with the above heaven load test). I'm sure that would climb with a more intense gpu load and I will be testing that soon.

Case is an InWin 303 with 1 rear fan and 4 radiator fans in push/pull at the top exhausting out. Intake is filtered at the bottom but with no fans there.(Negative case pressure). Air inside the case seems to stay cool on the intake side and really warm on the exhaust side before leaving the case itself.

Fans are Cougar Vortex PWM 120mm (800-1500 RPM 70.5cfm/119.8cmh max Hydro-Dynamic)

I am currently still adjusting voltage/frequency as Prime95 is throwing fits and shutting down the computer after a short time. (Unsure if this is a vcore issue, vdroop issue, or something I'm missing). It is seeming to improve as I increase vcore however I am about to test 1.335v for 4.8ghz which seems a bit high for that frequency. I have overclocked for a number of years however I am in a bit of uncharted territory when it comes to stress testing this hard.
I will update this post with current settings for llc, etc.
**CPU LLC Level 5
** Svid Best Case
**Long Duration Power Limit 4095
**Short Duration Power Limit 4095
**Package Power Time Window 127

The Ram is 16gb of G.Skill 3200 Ripjaw V series, running xmp at 3200. Memtest has completed successfully. This will likely get upgraded to 32 or 64 in the near future (would be sticking with G.Skill though)

The purpose of the machine is for heavy photoshop and Lightroom work, and some lighter video work (final cut pro & blender) with the ultimate goal just to have a rock solid system that I can start large batch processes or exports and walk away from for hours without having to worry about coming back to a crashed system.

***Prime95 shut down the pc again at 1.33v at 4.8ghz. Attempting 1.33v. I'm starting to wonder what else this might be.

 

[ceslen]

*** I reset defaults and Prime95 is holding. It's definitely one of my settings. Just have to narrow it down.

Prime95 is still shutting down the pc at 1.33v 4.8ghz. I'm going to post/repost some of the power values in hopes that someone may see something I'm missing.
*Turbo Parameters
CPU LLC Level 5
Long duration power limit 4095
Short duration power limit 4095
Package power time window 127
IA AC Load Line 0.01
IA DC Load Line 0.01

SVID is enabled and Best Case Scenario
CPU Core/Cache Current Limit Max 255.50

Cpu Core/Cache is Adaptive
Turbo Core Voltage 1.335v with a negative offset of 0.001v
DRAM Voltage 1.3530
VCCIO Voltage 1.1
System Agent Voltage 1.1

 

[ceslen]

!!! It’s the VRM’s! Even with the heat sink on the Z370-A they are scalding hot. Repositioning and adding fans for vrms and retesting.

Prime95 really is a murder of a stress test.

 

[EarthDog]

Consider setting an avx offset of a couple hundred mhz as well. Maybe 4.5ghz avx and 4.8 ghz non.

 

[ceslen]

Consider setting an avx offset of a couple hundred mhz as well. Maybe 4.5ghz avx and 4.8 ghz non.

When you have an avx offset like that, how does the system handle the voltage?
Example:
If I have tested stable at [email protected] and I setup my overclock at 1.345@5ghz with a -2 avx offset and its stable as well, how do you know what voltage is being applied at 4.8 when an avx load is used?

 

[EarthDog]

Should be the same. Look at the voltage in software like you do without using an offset.

 

[trents]

AVX offset should not affect vcore. What motherboard are you using? Is this an Asus motherboard? Doesn't look like you won the silicon lottery with that CPU.
Please create a Signature. Go Advanced>Settings>Edit Signature.

 

[ceslen]

**This did not resolve the issue.

It would seem I have possibly found the reasoning for the vrms overheating. If I am reading this past post correctly, it appears that adaptive voltage allows the cpu power control unit to over volt the cpu by .1v
If this is the case, when my stable oc voltage is found for a particular frequency and I begin an avx load the processor increases my stable voltage by .1v and creates heat issues with the vrm.
I am copy/pasting the post discussion regarding this below.

Adaptive Mode: Adaptive voltage affects voltage for Turbo multiplier ratios only. Unlike Offset, using Adaptive does not affect idle/light load Vcore. Therefore, Adaptive mode is the preferred method for overclocking Haswell processors if one wishes to retain dynamic voltage changes according to processor load without running into issues with idle Vcore becoming too low..

There is one issue with Offset and Adaptive Mode that needs to be taken into account. The processor contains a power control unit which requests voltage based upon software load. When the PCU detects AVX instructions, it will ramp Vcore automatically beyond normal load voltage. There is no way to lock Vcore to prevent this if using Offset or Adapative Mode. This is pre-programmed by Intel into the PCU.

As an example, a CPU is perfectly stable at 1.25V using a manual voltage (static), if Adaptive or Offset Mode is used instead, it is impossible to lock the core voltage when running software that contains AVX instruction sets – stress tests such as AIDA and Prime contain AVX instruction sets. When the AVX instructions are detected by the PCU, the core voltage will be ramped an additional ~0.1V over your target voltage – so 1.25V will become ~1.35V under AVX load. If you intend to run heavy load AVX software, we recommend using Manual Vcore, NOT Adaptive or Offset Mode.

Most of us do not run AVX related software, so this is a non-issue. Either way, dialing in an overclock using Manual Vcore to determine how much voltage the processor needs under full load is best - Adaptive or Offset mode can be used to match the stable voltage later on. Simply type the target load voltage into the entry box "Additional Turbo Mode CPU core voltage" to set adaptive voltage.

 

[ceslen]

AVX offset should not affect vcore. What motherboard are you using? Is this an Asus motherboard? Doesn't look like you won the silicon lottery with that CPU.
Please create a Signature. Go Advanced>Settings>Edit Signature.

I'm still researching on the relationship between avx offset and vcore to try and get a better understanding of how they work. With my goal being the lowest stable vcore for avx loads, I'm unsure of how much room that will leave me for increased non avx load overclocks.

I thought I had made progress with the swap to manual but it appears Prime95 still shuts down the system. I'm at a loss. No other software creates these issues (and I have been running 1.345v@5ghz for a while). It's just Prime causing these issues. I do wish I could find the cause but I'm not sure I'll be able to.

I have read a few posts pointing to a failing power supply so I may investigate that possibility. I'm running an EVGA Supernova 850W B2. I wouldn't think that would be failing but I've seen stranger things.

Making a note here for future me:
Prime95 shut down computer at all voltages and frequencies beyond 4.6ghz set on auto vcore @79c(currently has run for 30 minutes)
Temps never exceeded low 80s even when pushing 5ghz 1.34v
Prime95 did allow me to dial in vcore and remove errors for the time it ran before shutdown
Prime95 is currently the only software causing a shutdown. Even the Mac Yes test pushed temps but did not cause shutdown.
VRM's were running quite hot and a fan was installed over their heatsink, however I do not believe they are the root cause of the shutdowns.

Thanks to everyone for the help.

I'll get the Sig taken care of asap

 

[trents]

Don't make this more technical than it needs to be. The relationship between vcore and AVX offset is irrelevant. What matters is that recent versions of Prime95 uses second generation AVX (AVX 2) which is brutal and will cause stress test failure when other stress testers and real life computing will not. Older versions of Prime95 (like 27.7) use AVX 1 which is not as demanding and still older versions of Prime95 (26.6) don't use AVX at all.

Doesn't seem like to me you have a CPU that is going to be truly stable at 5.0 with no AVX offset. So try 4.9 or 4.8 or use an AVX offset. Temp is not always the issue. Some 8700k CPUs just are not 5.0 ghz capable. At least not without an AVX offset. You have put this thread in the Cooling section of the forum but it doesn't seem that temps are turning out to be your issue, either with regard to the CPU or the VRM, since you put a fan on the latter and that didn't help. I think you just need to lower your expectations as to overclock level.

Siliconlottery.com sold a lot of "5 ghz plus" CPUs but they were not tested with long trials of recent version Prime95. Their stress test was much less demanding, passing one hour of Realbench I believe.

 

[ceslen]

Don't make this more technical than it needs to be. The relationship between vcore and AVX offset is irrelevant. What matters is that recent versions of Prime95 uses second generation AVX (AVX 2) which is brutal and will cause stress test failure when other stress testers and real life computing will not. Older versions of Prime95 (like 27.7) use AVX 1 which is not as demanding and still older versions of Prime95 (26.6) don't use AVX at all.

Doesn't seem like to me you have a CPU that is going to be truly stable at 5.0 with no AVX offset. So try 4.9 or 4.8 or use an AVX offset. Temp is not always the issue. Some 8700k CPUs just are not 5.0 ghz capable. At least not without an AVX offset.

True. I definitely agree that this is not a 5ghz avx chip. I really don't need it to be anything more than stable with heavy avx loads, which it is at 4.6 on auto Vcore (so far so good). Prime95 just threw a curve ball at me since I've had it running at 1.34@5 for the last 4 months (for gaming and light editing on personal photos). I thought I had a decent grasp on its oc ability and I could just knock it down a notch or two and turn it into a productivity machine. Prime showed me I was wrong
Even with the processor not being what I thought it was, I probably will still replace the psu. I have had it in various systems for the last 4 years. It's probably time, and if it fixes the Prime shutdowns over 4.6 that'll be a bonus.

- - - Auto-Merged Double Post - - -

Don't make this more technical than it needs to be. The relationship between vcore and AVX offset is irrelevant. What matters is that recent versions of Prime95 uses second generation AVX (AVX 2) which is brutal and will cause stress test failure when other stress testers and real life computing will not. Older versions of Prime95 (like 27.7) use AVX 1 which is not as demanding and still older versions of Prime95 (26.6) don't use AVX at all.

Doesn't seem like to me you have a CPU that is going to be truly stable at 5.0 with no AVX offset. So try 4.9 or 4.8 or use an AVX offset. Temp is not always the issue. Some 8700k CPUs just are not 5.0 ghz capable. At least not without an AVX offset. You have put this thread in the Cooling section of the forum but it doesn't seem that temps are turning out to be your issue, either with regard to the CPU or the VRM, since you put a fan on the latter and that didn't help. I think you just need to lower your expectations as to overclock level.

Siliconlottery.com sold a lot of "5 ghz plus" CPUs but they were not tested with long trials of recent version Prime95. Their stress test was much less demanding, passing one hour of Realbench I believe.

Also true regarding the deviation from cooling being the subject. The post started out just being about temps during stress tests and worked its way into hardware troubleshooting while I was trying to stress the temps with Prime. I appreciate the help posted as I was discovering these kinks and am happy to let this thread become inactive, or moved to an appropriate alternate location if a mod so desires.

 

[EarthDog]

Please edit the title back to what it was. Otherwise, the problem cant be searched and found.

Our forum has a feature that allows one to place a 'tag' on thread marking it "Solved". Edit the first post, go advanced and it is in prefix dropdown.

Their stress test was much less demanding, passing one hour of Realbench I believe.

It changed a while ago to something more strenuous. Non avx prime, linpack, and some other software iirc.

 

[ceslen]

Please edit the title back to what it was. Otherwise, the problem cant be searched and found.

Our forum has a feature that allows one to place a 'tag' on thread marking it "Solved". Edit the first post, go advanced and it is in prefix dropdown.

Gotcha Done and done