Looking for CPU fine-tuning OC advice

[Malfeasance]

Machine setup:

i7-9700K with a Noctua NH-D15
Gigabyte Aorus Pro bios version F11
16Gb G.Skill Ripjaws DDR4-3200
EVGA Black Gaming RTX 2070S with their stock cooler
Corsair RM 650x
Win 10 x64
Fractal Design Meshify with 2x 140mm intake, 2x 140mm top exhaust and 1x 120mm rear exhaust.

Having some trouble with the i7 and looking for some advice from people with similar setups and successful OCs if possible.

Bit of a disappointment this has been so far, but I may be missing something. Started off with your run of the mill 5.0Ghz attempt, 1.30v, turbo LLC, x45 cache ratio, C1E and all that disabled. Long story short, this simply wouldn't work. I had the voltage up to 1.38v at one point and it did okay in benchmarks but started to blue screen in games, starting with Elite Dangerous which is relatively light to run. Temps have been well within acceptable ranges, topping out at 68c, but getting up towards and slightly over 1.4v with even high LLC isn't where I want to be at for longevity's sake, and seems insanely high to me anyway for what these chips should be capable of.

I dropped down to 4.9Ghz (which should be doable) at 1.375v and was fine until today when Mon Hun crashed the PC, even though I've played it previously with no issue (DRM thread hell, thanks again Capcom). Dropped it down again where it's at now at 4.8Ghz which I don't really want to accept if I can help it, at 1.375v, still high LLC I believe.

My primary question is: is there anything I can do to try to stabilize at least 4.9Ghz without having to use an AVX offset? Drop cache ratio more perhaps? Do I bite the bullet and offset to try for 5.0?

I was also going to mention the Super overclock being really finicky buuuut I just realized I didn't have my fan curve activated. Afterburner could really use some better indicators for that, like simple check boxes or something. Hunt Showdown was giving me increasingly bad odd pulsing lighting and splotchy artifact problems about 20-30 minutes into a match, but it was also ramping up to 77c in the meantime, so we'll see how it goes at lower temps.

Been benching with Time Spy, Cinebench R20 and a couple other tools, didn't do any Prime95, never cared for it. Everything I ran passed though, only been having problems with gaming. Any input would be great. I've done a ton of poking around the web for info on similar setups and almost everyone can hit at least 4.9Ghz at like 1.2-1.3v even without offsets. Did I really just get that bad of a chip or am I missing something with the Aorus bios settings? There's a bios update that includes voltage stability improvements but I REALLY don't like updating bios if I can help it; I bricked an Asus board doing it once, wasn't fun. The majority of the overclocks I've seen have been on F9-11 anyway.

Thanks in advance.

Edit - After tinkering for a bit, I'm now unable to start any overclock. Board sticks in a boot failure after looping several restarts if I try to go back to my old settings. No idea what's going on, honestly about to scrap the entire project at this point because realistically 4.8-4.9Ghz on all cores isn't going to be enough of an overclock to matter that much. I was originally aiming for 5.0 or more, but that's clearly not going to happen. Still, any advice to give in my OP will be appreciated if I return to trying it someday.

[Jelly Donut]

I am not sure exactly how to overclock with an Aorus Board, but I did manage to successfully overclock mine (I use an ASUS Board, see below for specs). Mine was a pretty rare 9700k, was able to handle up to 5.2Ghz, AVX 2 and 1.375V, but it was toasty, so to keep my system silent, I used 5.0Ghz, AVX 2 @ 1.33V.

You also need to have luck with the Silicon Lottery; 9700k uses quite an inefficient silicon and it consumes almost as much power (and produces as much heat) as a 9900k, since you know, its actually a 9900k with HT disabled, since when binned at Intel it couldn't match the specs of the 9900k.

Try slowly to overclock from a low frequency. However, AVX workload (Prime 95) is something that's inevitable when the PC or game needs it, so you should bin it until it's stable and if you can't leave it on the default setting.

Also, Manual OCing is more efficient (once you hit the sweet spot) and make sure you aren't using auto overclocks - they suck.

Binning Hardware and environment:
Motherboard: Asus Z390 ROG Maximus XI Gene
Ram: G Skill Trident Z RGB (2 X 16GB) 32GB C16 3200Mhz, XMP II Enabled
CPU Cooler: Corsair H150i Pro with 3 x Noctua NF-A12x25s in PUSH config and 3 x Cooler Master MF120R RGBs in PULL config
Power Supply: Corsair HX750 80 Plus Platinum (2 x 4+4pin CPU connectors were used while binning)

Temperatures during testing
Ambient: 25C - 26C
CPU Package: 65C - 70C (During Normal 100% CPU Stress), 80C - 85C (During 4K (AVX) Prime95 Test onwards)
Liquid Cooler Temperature: 37C-39C

Motherboard Settings:
Ai Overclock Tuner: XMP II
BCLK Frequency 100.0000
ASUS Multicore enhancement: Disabled
SVID Behaviour: Intel's Fail Safe
AVX Instruction Core Ratio Negative Offset: 2
CPU Core Ratio: Sync All Cores, 50
CPU Ram Cache Ratio: 44
DRAM Frequency DDR4-3200Mhz
Xtreme Tweaking: Enabled
CPU SVID Support: Disabled

CPU Core/Cache Current Limit Max: 255.75 (MAX)
BCLK Aware Adaptive Voltage: Enabled
CPU Core/Cache Voltage: Manual Mode
CPU Core Voltage Override: 1.335V

-External Digi+ Power Control
CPU Load-Line Calibration: Level 3
CPU Current Capability: 140%

-Internal CPU Power Management: Turbo Mode Parameters-
Long Duration Package Power Limit: 4095 (MAX)
Package Power Time Windows: 127
Short Duration Package Power Limit: 4095 (MAX)

*All other settings not listed are set to their default settings, being it Auto or any other assigned by XMP II.

[Malfeasance]

Really appreciate that, thank you. I'm currently investigating this boot failure problem at the moment. Refuses to start with settings that I know worked previously now. No chance I've caused any damage to anything unless something that wasn't good to begin with has failed, because voltages and temps were all just fine. Loads just fine on stock settings so I'm assuming that's a good sign.

I ultimately went with the Gigabyte board because of the excellent VRM cooling, but at this point I wish I'd gone with an Asus. Their bios is incredible compared to this mess. I still haven't been able to find any comparable internal CPU power management settings or CPU current capability.

Will post back if I'm able to get it started on any custom settings again.

Edit - getting somewhere. I was able to successfully boot with a saved OC profile at 4.9Ghz, 1.35v. AFAIK the only things I changed before were adding a slight boost to DRAM and VCCIO voltages as well as lowering cache ratio from 45 to 43. Will keep tinkering.

[Jelly Donut]

Ursprünglich geschrieben von Malfeasance:

Really appreciate that, thank you. I'm currently investigating this boot failure problem at the moment. Refuses to start with settings that I know worked previously now. No chance I've caused any damage to anything unless something that wasn't good to begin with has failed, because voltages and temps were all just fine. Loads just fine on stock settings so I'm assuming that's a good sign.

I ultimately went with the Gigabyte board because of the excellent VRM cooling, but at this point I wish I'd gone with an Asus. Their bios is incredible compared to this mess. I still haven't been able to find any comparable internal CPU power management settings or CPU current capability.

Will post back if I'm able to get it started on any custom settings again.

Edit - getting somewhere. I was able to successfully boot with a saved OC profile at 4.9Ghz, 1.35v. AFAIK the only things I changed before were adding a slight boost to DRAM and VCCIO voltages as well as lowering cache ratio from 45 to 43. Will keep tinkering.

When doing a OC, if you cannot pass a 35 min Prime 95 run and a 35 min Furmark CPU Stress (non AVX CPU Stress), there is a very high chance the system will crash regardless of activity.

[Malfeasance]

I'll give them a shot then. For an update, I put the CPU back at 4.9Ghz, 1.375v, turbo LLC (1.38v max observed) basically restarting my progress, and ran it through OCCT for brief 10 minute runs, not ideal, but a starting point for more intense testing.

Large data set with no AVX went through fine, temps great. Large data set with auto (AVX2) hard locked within 8 seconds.

I bit the bullet and applied a -1 AVX offset and tried again. Both passed this time. Small data sets on both passed as well, but small with AVX flirted big time with 100c, though never hit it, so 5.0Ghz is definitely off the table without at least an AIO.

This leads me to conclude that Monster Hunter is likely using AVX, which would make sense considering it's DX12, hence the sudden lockup when DX11 games have been perfectly happy. Strange considering I'd played it in a couple sessions without trouble, as well as DX12 Hitman 2. I'm curious as to whether MH being a grotesque CPU hog made it worse.

I'll have to verify this at a later time by keeping an eye on clocks in MH, but for now I'm going to hit it with more stresses to make sure everything is good at this voltage. Thanks again for your input. Unfortunate that the chip doesn't seem to be willing to go much higher without pouring on the juice, but it is what it is. My 2600K was a lame duck as well, only ever managed a mediocre OC on that as well, so I'm used to it

Edit - Furmark passed 35 mins with no sweat. What Prime95 tests do you suggest? Small FFT pushes towards thermal throttle really quickly, basically what OCCT was doing during AVX testing, but ran. Locked nearly immediately on Blend.

Edit 2 - Just read where a guy bought an Aorus Pro and said he had to crank voltages to get a worse OC than his Asus board. Fantastic, definitely seems to be the case here as well.

[Malfeasance]

Update post:

Good news and bad news I guess. Decided to run more OCCT in place of P95, same type of tests anyway. Despite passing a 10 minute test previously, small batch AVX ran for 8 minutes or so at -1 offset before spitting out errors. Great.

Down to -2 offset. Small AVX ran for 20 minutes this time, looking good. Sudden hard lock, not good.

At this point I'm down to stock boost with -3 offset...

Ran small non-AVX for 35 minutes at 4.9Ghz. Finished with no errors, max temps just over 80c. Am I just looking at the world's worst 9700K, Aorus Pro, or is something faulty here? Vcore is still 1.375v, turbo LLC. At this point my only option is to lower LLC and raise Vcore more, which will definitely push small AVX temps past 100c. I honestly can't believe this chip is being so stubborn in this one department. To be fair it also won't run 5.0Ghz with non AVX at 1.375v either.

Last time I tried to apply slightly higher voltage to DRAM and VCCIO is when the boot loops and bios resets happened. I'm seriously tempted to send this Gigabyte board back and pick up an Asus alternative, but realistically I'll probably just run with what I got at the moment, as poor as it is.

[Jelly Donut]

Ursprünglich geschrieben von Malfeasance:

Update post:

Good news and bad news I guess. Decided to run more OCCT in place of P95, same type of tests anyway. Despite passing a 10 minute test previously, small batch AVX ran for 8 minutes or so at -1 offset before spitting out errors. Great.

Down to -2 offset. Small AVX ran for 20 minutes this time, looking good. Sudden hard lock, not good.

At this point I'm down to stock boost with -3 offset...

Ran small non-AVX for 35 minutes at 4.9Ghz. Finished with no errors, max temps just over 80c. Am I just looking at the world's worst 9700K, Aorus Pro, or is something faulty here? Vcore is still 1.375v, turbo LLC. At this point my only option is to lower LLC and raise Vcore more, which will definitely push small AVX temps past 100c. I honestly can't believe this chip is being so stubborn in this one department. To be fair it also won't run 5.0Ghz with non AVX at 1.375v either.

Last time I tried to apply slightly higher voltage to DRAM and VCCIO is when the boot loops and bios resets happened. I'm seriously tempted to send this Gigabyte board back and pick up an Asus alternative, but realistically I'll probably just run with what I got at the moment, as poor as it is.

This sounds pretty awful. what motherboard are you using? Master/Ultra/xTreme?

[Malfeasance]

Aorus Pro non-WiFi, bios F11. F12 introduces "Fix CPU Vcore and power behavior" whether this is worth it I'm not sure, would rather not risk the brick if I can help it, but I may have to.

Edit and delete/new reply to avoid double posting but bump thread. I'm throwing in the towel on the project at this point. My only question now is to either:

Leave the AVX offset and run 4.9Ghz. I stepped down the Vcore to 1.35v and it passed OCCT small data. I planned on continuing to step it down to reduce the now unneeded heat and run Realbench for an hour or two with HWInfo to verify no errors.

Or I drop the OC entirely. My thinking is that with the OC, single core programs that happen to use AVX, should they exist (I've read CS:GO trips offsets and uses something like 2 cores), will only run at 4.6Ghz instead of boosting. But at the same time 4.7Ghz with AVX wasn't stable at high voltage anyway, let alone stock auto.

Is it worth the tiny improvement to worry about dropping it? Alternatively, are there a huge amount of games that don't use AVX anymore to worry about keeping the OC?

[xSOSxHawkens]

So, first thing is you are going about it from the wrong end. You are starting high and working down. You need to start at stock and work up.

Second thing is to use a batch of tests *AND* keep an accurate log. This will allow you to track with great consistency the voltage requirements of the chip you have, as all are different, and doing it this way will allow you to learn the weakness and strengths of your particular chip. An (old by todays standards) example from my personal past was my first gen core2quad which could clock its FSB to 1866 but could only hold a 3.6ghz core. Every chip will have limits, but you can push them in more ways than just one.

For you, Start by locking the chip to its stock non-turbo core speed on all cores and make sure turbo boost is off, then reduce the voltage in minor increments until you reach stability issues. Bump it a notch or two back up and record that as your minimum stock voltage.

Now start upping the all core locked frequency at 100Mhz per jump till you hit stability issues, record tempetures and scores from the test suite as you go.

Once you hit the stability up the voltage one tick at a time till you regain stability, then start the process again.

Eventually you will hit the upper middle speeds you can hold, and by that time you will have a noticable pattern in voltage per 100Mhz required, which scales higher as you go fairly consistently. You will also be able to track tempetures, and predicted benchmark increases, both of which can be used to spot minor stability issues that are hard to notice (such as the system running stable, but taking a bit longer in just one bench that it should b/c of a *super* minor instability). This will allow you to have a good idea of what the chip will actual want or need when you hit walls on the upper end, and also allow you enough info to know when you have hit the wall period.

Its hard to get to it at the moment, but I have a fairly detailed log on my 4790k build I can post for you as an example so you can see what I mean. I will try to post it in the next few hrs.

For testing programs I recomend

OCCT
POV Ray
Intel CPU Test
Intel Burn test (OCCT)
Cinnebench
there are a couple others. The test suite I use is about 15 minutes per run set of all programs for basic tests, with extended testing being daily use. I start with the shortest or most likely to fail tests first and then move on, that way I am most likely to detect failure in the earlier parts of testing to save time.

[76561199051690163]

Don't