7900x OC stability issues.

qdub

qdub

qdub

Associate
Joined
26 Apr 2018
Posts
17
Hi guys,

I purchased the pre-binned 7900x @4.7ghz a few months ago but cannot for the life of me get a stable OC (not new to overclocking, but new to this platform).

I've tried numerous settings suggested by others (offsets, adaptive, per core, sync all cores) but just wondered if anyone could post their current settings for me to play around with again or help me out please?

Settings tried recommended by RSR but with XMP for my ram and bumped voltages slightly higher:

Ai Overclock Tuner [XMP]
XMP [XMP DDR4-3603 16-16-16-36-1.350V]
CPU Strap [100]
BCLK Frequency [100.0000]
AVX Instruction Core Ratio Negative Offset [3]
AVX-512 Instruction Core Ratio Negative Offset [5]
CPU Core Ratio [Sync All Cores]
ALL-Core Ratio Limit [47]
Min. CPU Cache Ratio [32]
Max. CPU Cache Ratio [32]
DRAM Frequency [DDR4-3600MHz]
CPU SVID Support [Disabled]
CPU Core Voltage [Manual Mode]
CPU Core Voltage Override [1.178]
DRAM Voltage(CHA, CHB) [1.3500]
DRAM Voltage(CHC, CHD) [1.3500]
CPU Load-line Calibration [Level 6]
CPU Current Capability [140%]
Long Duration Package Power Limit [4095]
Short Duration Package Power Limit [4095]

Specs:

i9-7900x
EVGA 1080Ti FTW3 w/ EK waterblock
ASUS Rampage VI Extreme
G.Skill Trident Z RGB 3200mhz CL14 4x8GB - (actually bought this recently as was using 8 sticks before and thought it could be affecting OC)
Samsung 960 PRO 2TB w/ EK heatsink
Corsair AXi 1200w

Cooling:
2x 480mm EK rads, EK-D5 Pump, EK-XRES 250, 8x Noctua NF-12 PPC 3000 PWM, 2x Noctua AF-12 3000 intake and a biggun as exhaust.

Re-seated monoblock a few times due to seriously high temps, still high during bench (cancelled when hitting over 90oc, max has been 96oc with ambient 16-18oc). GPU max 40oc.

Sorry if I missed anything, just ask and I'll post.

Cheers!
 
what are you using to test and how quickly does it fail? do OCUK give you a core voltage they say works?

for what its worth you can improve temps a lot by undervolting the cache. for example mine is at 33x and 1.075v. however, while trying to get the core stable id suggest sticking it at 30x auto volts for now.
 
did some real world testing, games like far cry 5 whilst streaming with OBS will blue screen anywhere from 15 minutes to 1 hour

Ran 3DMark, PCMark and Realbench today BSOD within 1 - 10 minutes

edit: result! found the parcel, sticker was stuck in the bottom reading 4.7gamang 1.235v
 
Last edited:
Just an update, here are the current settings with changes made underneath each. Gone at it most of the day with no result.

Probably just overlooking something so fresh set of eyes would be great.

Ai Overclock Tuner [XMP]
XMP [XMP DDR4-3200 14-14-14-34-1.350V]
CPU Strap [100]
BCLK Frequency [100.0000]
ASUS MultiCore Enhancement [Auto]
AVX Instruction Core Ratio Negative Offset [3]
AVX-512 Instruction Core Ratio Negative Offset [5]
CPU Core Ratio [Sync All Cores]
ALL-Core Ratio Limit [47]
Min. CPU Cache Ratio [30]
Max. CPU Cache Ratio [30]
BCLK Frequency : DRAM Frequency Ratio [Auto]
DRAM Frequency [DDR4-3200MHz]
Xtreme Tweaking [Disabled]
TPU [Keep Current Settings]
CPU SVID Support [Disabled]
BCLK Aware Adaptive Voltage [Auto]
CPU Core Voltage [Manual Mode]
CPU Core Voltage Override [1.235]
CPU Cache Voltage [Auto]
Uncore Voltage Offset [Auto]
Uncore Voltage Offset Prefix [+]
CPU Input Voltage [2.100]
DRAM Voltage(CHA, CHB) [1.3500]
DRAM Voltage(CHC, CHD) [1.3500]
CPU VCCIO Voltage [1.10000]
CPU System Agent Voltage [1.10000]
PCH Core Voltage [1.10000]
PLL Reference Offset Mode Sign [+]
PLL Reference Offset Value [Auto]

*Changes*

CPU Input Voltage: 1.8 to 2.1

Vcore: 1.235 to 1.250

VCCIO: Auto to 1.1

System Agent: Auto to 1.1

CPU Load-line Calibration [Level 6]
CPU Current Capability [140%]
CPU VRM Switching Frequency [Auto]
VRM Spread Spectrum [Auto]
CPU Power Duty Control [Extreme]
CPU Power Phase Control [Extreme]
CPU Power Thermal Control [120]
DRAM Current Capability [140%]
DRAM Power Phase Control [Extreme]
DRAM Switching Frequency [Auto]
CPU Input Boot Voltage [Auto]
CPU System Agent Boot Voltage [Auto]
CPU VCCIO Boot Voltage [Auto]
Enhanced Intel SpeedStep Technology [Enabled]
Turbo Mode [Enabled]
Long Duration Package Power Limit [4095]
Package Power Time Window [Auto]
Short Duration Package Power Limit [4095]
CPU Integrated VR Current Limit [Auto]
CPU Integrated VR Fault Management [Auto]
CPU Integrated VR Efficiency Management [Auto]

*Changes*

Tried all of these at default, LLC Auto etc.

Intel(R) SpeedStep(tm) [Auto]
Turbo Mode [Enabled]
CPU C-states [Auto]
Enhanced C-states [Enabled]

Intel(R) Speed Shift Technology [Enabled]

*Changes*

Tried default and turning off SpeedStep / C-states.
 
It's pretty much always dpc_watchdog_violation

Think there's only been 1x irql_not_less_or_equal with some different settings and the old ram
 
Last edited:
From looking at those settings and given it's a Watchdog debug I'd wager you need more vcore. 4.7 at 1.235v may well be a bit optimistic for that CPU. 4.7 stable is a good CPU none the less but increase to 1.25v and see if the problem persists.

Also undervolting the cache when already experiencing instability is questionable advice. Leave both cache ratio and voltage in auto till the problem has been resolved.
 
I would turn of speedshift as its new and may not have been catered for in their binning.
Also remove XMP whilst you trying to stabilise the CPU.

Otherwise as said keep raising the vcore.

If its stable at higher vcore then a decision to be made if you happy or to hold for the vcore spec you was advised would work.
 
Upped vcore to 1.25v, disabled speedshift and put the cache ratio / voltage to auto.

Realbench failed at 11 minutes, should I now try not using XMP and manually enter the voltage and timings?

Also, is realbench (benchmark/stress test) the best to use?

Got 3DMark, PCMark 8 Advanced Edition but not sure on what's best to run on it... only used it for benching in the past.
 
Realbench uses AVX. You really want to be establishing AVX stability before applying an AVX offset, too. As you're not sure on how much voltage is needed for AVX workloads at the applied offset ratio.


You can disable XMP to rule out memory instability, too. Alternatively if not wanting to test for AVX stability, use Intel Extreme Tuning Utility (Intel XTU). This test does not use AVX extensions and also pulls a fair amount of current, so is a good gauge for non AVX workloads.


Also reduce your input voltage to 1.85-1.9v. 1.21v is too high.
 
Last edited:
Tested again without XMP but still the same crash around 10 minutes in.

Thanks for the info, I feel like AVX is the issue... should I set the offsets to AUTO (with auto they're obviously matching the clock @4.7 so would this cause even more instability?) and try again?
 
4.7 with AVX workloads is asking too much of the CPU. If wanting to continue to use Realbench test stability without any AVX offset applied with a more realistic ratio. 43 or 44 for example. As I've already said, I'd recommend XTU for testing non AVX stability.
 
Ok, ran XTU for 20 minutes (hope that's enough for now) and it's stable with temps not going above 67oc.

Is the next step for AVX stability like you said, starting high and lowering 1 by 1?
 
That depends. Are you planning on running anything that uses AVX? If not, set a -3 and -5 offset respectively for 512 and see how you get on with simply using the machine for your daily workload.
 
To be honest, I'm not entirely sure what else uses it apart from OBS.

This build is primarily for gaming, streaming (OBS), photoshop, lightroom, sony vegas and media streaming (remote connection from phone / laptop to access files, view cctv and all of that jazz).

Using the previous settings with vcore 1.25v and AVX -3 / -5 the system still crashes whilst playing games and using OBS, guessing that's the culprit?
 
Yeah, you'll likely need a bigger offset in that case. I've experienced similar when using OBS on my 7940X. Failing that cut your loses and drop to 4.6.

This pre-binning depends on a multitude of conditions. 4.6 on that CPU is still good.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom