Questions regarding overclocking

[Groda]

Hello everyone, hope you all are doing well. I have some questions when it comes to overclocking my i9 9900k using my ASUS ROG Strix z390-f motherboard.

My build:

GPU: Zotac AMP! RTX 2080ti
CPU: i9 9900k

PSU: Corsair HX1000i

CPU cooler: Corsair H150i Pro

Motherboard: ASUS ROG Strix z390-f

RAM: Corsair Vengeance LPX 3200MHz (CMK16GX4M2B3200C16)

 

These are the settings I have changed so far in the BIOS:

AI Overclock Tuner: XMP1

ASUS MCE: Disabled
Long Duration Package Power Limit: Maxed out (downclocked during stress testing otherwise)

Short Duration Package Power Limit: Maxed out (downclocked during stress testing otherwise)

CPU Core Ratio: Sync all cores to 50

CPU Core/Cache voltage: Manual, Core Voltage Override: 1.3

 

Question 1: When I try to run a stress test to see if I'm stable and check the temperatures it clocks down the CPU to about 4.4GHz (non avx load) with the following performance limits being flagged in HWinfo:
RING: Max VR voltage, ICCmax, PL4

IA: Electrical Design Point/Other (ICCmax, PL4,...)

Which limiter in the BIOS have I missed to turn off/increase to max? If I turn MCE on this throttling doesn't seem to happen but instead I bluescreen during cinebench, which I don't with MCE off, the throttle doesn't trigger in cinebench but does trigger in prime95 and OCCT large package. So MCE seems to make the system more unstable.

 

Question 2: I've read that the ASUS motherboards tend to overshoot with VCCIO and SA voltages if left on auto, with XMP turned on my motherboard sets VCCIO to 1.328v and SA voltage to 1.216v. Should I try to decrease them since from what I read that VCCIO is quite high?

 

Question 3: I've noticed that when I run stress test with a voltage set to 1.3v I seem to get up to 1.32v vcore reads from HWinfo. I've read that I should use the LLC options to try to account for voltage droop, but I seem to be getting the opposite of voltage droop? Or does the motherboard automatically set it at some level and I might try to lower it further because of the overshoot?

 

Question 4: Back when I overclocked my last CPU the agreement seemed to be that running Prime95 to test stability was the way to go. When I've been reading up about overclocking the i9 9900k however I can't seem to find a general consensus on what programs to test with? Some say realbench, others say prime95 but running a custom test that doesn't hammer AVX loads as hard as the regular test. While others still repeat the mantra of regular prime95 with AVX on is still the way to go.

 

But from what I understand the i9 9900k runs so hot that running it in regular prime95 with AVX on is almost possible if you wanna clock it around the 5.0GHz mark (obviously depends on cooling)? I'm mostly going to be using the system for games. But since I've read that Battlefield V and now Anthem has started to use AVX loads (not to the level of prime95 small FFTs obviously), I want my system to be stable for that type of AVX loads at least. If I try to run prime95 with avx on at 1.3v my temperatures start of at pretty much 90°C. What testing programs do you recommend for stability testing?

 

Thanks in advance for any help you can give me. This PC has been a pain so far with two faulty pairs of RAM and I'd really like to get it into working shape during this weekend so any help is very much appreciated. The fact that pretty much every thread and youtube video I find on the subject says different things on what programs to use and what's considered stable has been very frustrating.

[Jurrunio]

Q1: From my experience, manual overclock makes the board ignore power limits and any throttling will be thermal based. Maybe it's not actually stable that that voltage (or not enough loadline calibration (LLC)? Especially since MCE forces the CPU to run 4.7GHz and it does crash in that case. You can try set AVX offset to 0 too, this settings means "reduce max multiplier by this value when AVX instruction set is in use".

 

Q2: I myself keep both VCCSA and VCCIO below 1.25V, but 1.35V is still deemed acceptable for high memory overclocks. What is weird is that SA is going high, but IO isn't. Anyway, sometimes memory overclock is better with lower VCCSA and VCCIO so I'd drop them back down. To avoid memory and CPU instability from happening at the same time when you're dealing with Q1, I'd turn off XMP, memory back to 2133MHz and make SA 1.05V, IO 0.95V which are their stock values.

 

Q3: Yes the board's auto LLC is being a tad aggressive. Asus boards give 5-9 levels of LLC depending on the model (not sure what Z390-F has), but I recommend using the level that's 1 step higher than the middle one in the first try. I also forgot if Asus LLC gets more aggressive at bigger numbers (level 9 so to say) or smaller numbers (level 1), my Asrock board is the latter but who knows.

 

Q4. Do non-AVX test like Prime95 version 26.6 and find the right voltage (might want to add 0.05v for extra stability) first, then do AVX test with new Prime95 and increase AVX offset values until it gets stable.

[Groda]

4 hours ago, Jurrunio said:

Q1: From my experience, manual overclock makes the board ignore power limits and any throttling will be thermal based. Maybe it's not actually stable that that voltage (or not enough loadline calibration (LLC)? Especially since MCE forces the CPU to run 4.7GHz and it does crash in that case. You can try set AVX offset to 0 too, this settings means "reduce max multiplier by this value when AVX instruction set is in use".

 

Q2: I myself keep both VCCSA and VCCIO below 1.25V, but 1.35V is still deemed acceptable for high memory overclocks. What is weird is that SA is going high, but IO isn't. Anyway, sometimes memory overclock is better with lower VCCSA and VCCIO so I'd drop them back down. To avoid memory and CPU instability from happening at the same time when you're dealing with Q1, I'd turn off XMP, memory back to 2133MHz and make SA 1.05V, IO 0.95V which are their stock values.

 

Q3: Yes the board's auto LLC is being a tad aggressive. Asus boards give 5-9 levels of LLC depending on the model (not sure what Z390-F has), but I recommend using the level that's 1 step higher than the middle one in the first try. I also forgot if Asus LLC gets more aggressive at bigger numbers (level 9 so to say) or smaller numbers (level 1), my Asrock board is the latter but who knows.

 

Q4. Do non-AVX test like Prime95 version 26.6 and find the right voltage (might want to add 0.05v for extra stability) first, then do AVX test with new Prime95 and increase AVX offset values until it gets stable.

Q1: From what I heard it's normal behaviour on the newer ASUS boards that you have to increase all the power limits manually if you don't enable MCE. Found this in an article about ASUS ROG Maximus and i9 9900k: 

"Here’s the weird thing: If you leave the ASUS board to just auto, without any form of XMP enabled, then the Multicore Enhancement feature is left to “Auto” as well. In this version of Multicore Enhancement, instead of boosting to the maximum single-core frequency, which is very obvious when done and violates spec, ASUS is extending the turbo boost duration. That’s the new form of MCE. It’s tricky and well-hidden."

From what I can tell I've just missed one BIOS setting for whatever it is that makes it slow down when I don't have MCE on but googling my issue doesn't bring up anything on what setting I might want to change to get rid of it.

 

Q2: I have tested the XMP profile 1 in memtest86 for 8 or 12 passes, 6000% coverage in RAMtest and done 5 hours of prime95 with avx on so I don't think I have RAM instability. (Might change once I start lowering VCCSA and VCCIO obviously). Might want to try to lower them before I continue with the CPU overclock though because I've heard it might lower temperatures somewhat? Or get rid of XMP while I do the CPU overclock, but then I might have to reconfigure it again when i turn XMP back on due to the increased RAM speed affecting how the CPU operates?

 

Q3: From what I could read from the BIOS I think it defaults to level 6? Which I think is the middle of the levels +1. (1-9) Might try to reduce it further. Though that doesn't bode too well for how fast I'll be able to get this overclock running since I know 4.8GHz avx offset isn't stable at 1.3v on the current settings I have.

 

Q4: Alright, that means I'm going to have to go down to at least 4.7GHz or even less most likely due to the heat generated by prime95 with AVX load. I've had people at other communities tell me to just run realbench for 8 or so hours and avoid running prime95 with avx on (unless you do a custom test to make sure it doesn't hammer the avx instructions as hard, since no real life application uses the AVX functions like the small FFT test does). But I was kind of leaning on just finding some setting that's stable during AVX load in prime95 and go with that.

 

Thank you for taking your time to try and help me.

 

If anyones around that has one of the newer ASUS ROG motherboards have you had similar experiences compared to me in regards to my first question?

[Jurrunio]

Q1: It is possible that MCE still did more things than just making all core turbo sustained for infinite amount of time, it's just not as critical as the turbo duration thing.

 

Q2: The temperature decrease, if any, will be because the CPU is doing less work and gives you less performance. Memory bottleneck, that is. You might have to reconfigure the overclock after re-enabling XMP, but usually a slight increase in core voltage will do.

 

ROG users? @Princess Cadence is also using 9900k and Z390 (but a Maximus 11 Gene, so not a half-assed ROG like the Strix is) and @Morgan MLGman on older Z170 Maximus 8 but I don't think Asus changed their BIOS much. Not the basics anyway.

[Groda]

Thanks again for taking the time to help me out.

 

At this point I'd even settle for getting it stable at 4.8 GHz without it throttling down due to power limits to be honest. Though I was not able to find a good setting for vcore that's had me stable at 4.8 GHz back when I had MCE on, unless I overshot it and the heat made it unstable that is.

[Jurrunio]

How hot is it running at 4.8GHz (until it crashes, that is)?

[Groda]

Can't remember what the last voltage I tried for 4.8GHz was but I think it was either 1.28 or 1.29v and for prime95 with AVX on it runs at around 90-95°C the time before it crashes (think it takes about 5 or so minutes to crash). Voltage doesn't include the overshoot I get from the default LLC.

[HoneyBadger84]

Are you sure you got a good mount on your CPU block? Those Temps for that voltage/speed seem ridiculous. For reference I have a Thermaltake 280mm AIO and I'm only seeing 90C max on one core out of the 8 (the rest are in the 78-86C range) during heavy stress testing at 5GHz 1.29V - seems to me there is something amiss. Airflow, cpu block mounting/thermal paste, or maybe a bend on the tubing that's cutting flow? 

[Groda]

I followed a guide on how to mount the cooler on the CPU etc so I hope so. I guess I could try to reattach it but I have to check if my friend has some thermal paste left since I currently don't have any. The picture is how I have it set up atm. There's three fans on the radiator, two fans that pulls air out at the top and one fan that pull air out at the back. (Might have negative pressure? I'm not too good with the air flow...)

 

Don't think the tubing is bent? At least not more than most of the guides I've seen have them curved? Are your temperatures during avx load in prime95 or in other stress tests? Since OCCT and realbench seems to run cooler for me.

 

If you don't mind me asking what's your VCCSA and VCCIO voltages at? Since I've heard they contribute to more heat and mine are pretty high as they are atm set to auto. I might just have gotten a shitty CPU but I'm kind of disappointed it doesn't look like I'll get to overclock it any and I'm not even sure I can get 4.8 stable during avx.

 

Thanks a lot for taking your time to reply.

[Jurrunio]

On 2/17/2019 at 2:12 PM, Groda said:

Can't remember what the last voltage I tried for 4.8GHz was but I think it was either 1.28 or 1.29v and for prime95 with AVX on it runs at around 90-95°C the time before it crashes (think it takes about 5 or so minutes to crash). Voltage doesn't include the overshoot I get from the default LLC.

that's too hot.

 

Try tighten whatever's holding the CPU cooler onto the CPU tighter, higher mounting pressure can sometimes help. Otherwise maybe the thermal paste application isnt good enough, or the pump speed and fan speeds arent that high.

 

20 minutes ago, Groda said:

If you don't mind me asking what's your VCCSA and VCCIO voltages at? Since I've heard they contribute to more heat and mine are pretty high as they are atm set to auto. I might just have gotten a shitty CPU but I'm kind of disappointed it doesn't look like I'll get to overclock it any and I'm not even sure I can get 4.8 stable during avx.

VCCSA and VCCIO dont affect heat output much, though better keep them below 1.35v at all costs and below 1.3v if you arent being aggressive for the sake of the memory controller.

[Groda]

1 hour ago, Jurrunio said:

that's too hot.

 

Try tighten whatever's holding the CPU cooler onto the CPU tighter, higher mounting pressure can sometimes help. Otherwise maybe the thermal paste application isnt good enough, or the pump speed and fan speeds arent that high.

 

VCCSA and VCCIO dont affect heat output much, though better keep them below 1.35v at all costs and below 1.3v if you arent being aggressive for the sake of the memory controller.

Going to attempt to reseat the CPU and cooler this weekend. I set the corsair controller for the cooler to balanced (defaults is quiet) so pump and fan speeds should be good? Did a quick benchmark in assassin's creed odyssey and the fans hit 1360rpm and the pump is at 2100, dunno what's "normal" but the PC is getting quite loud during that time so it feels like it's at least having the fans go up in speed properly.

 

The reason I'm wondering about VCCSA and VCCIO is because I saw in a thread on reddit that somebody said they lowered their temps by like 10°C by reducing them... They might have been mistaken though I'm just desperately trying to figure out if I might have settings in the BIOS that contribute to the heat as well.

[HoneyBadger84]

2 hours ago, Groda said:

I followed a guide on how to mount the cooler on the CPU etc so I hope so. I guess I could try to reattach it but I have to check if my friend has some thermal paste left since I currently don't have any. The picture is how I have it set up atm. There's three fans on the radiator, two fans that pulls air out at the top and one fan that pull air out at the back. (Might have negative pressure? I'm not too good with the air flow...)

 

Don't think the tubing is bent? At least not more than most of the guides I've seen have them curved? Are your temperatures during avx load in prime95 or in other stress tests? Since OCCT and realbench seems to run cooler for me.

 

If you don't mind me asking what's your VCCSA and VCCIO voltages at? Since I've heard they contribute to more heat and mine are pretty high as they are atm set to auto. I might just have gotten a shitty CPU but I'm kind of disappointed it doesn't look like I'll get to overclock it any and I'm not even sure I can get 4.8 stable during avx.

 

Thanks a lot for taking your time to reply.

Vccsa and vccio are something you shouldn't leave on auto. According to Buildzoid (extreme overclocker guy) you should never leave those on auto as motherboards can set them waaaaaaaaay too high. His guidelines we're 1.2V is "normal", anything over that is getting in to what's required for pushing the memory heavily, and anything over 1.35V is not good for system longevity. 

 

I have mine set at 1.2V on both because I'm running my RAM at 4000MHz CL17 with fairly tight timings at the moment for testing. If I were you I'd start at like 1.1 and see how they are. My board defaults to something silly low like 0.95 and 1.05.

 

Edit: added screenshot of what my board does "stock" except the 1.29V, that's for 5GHz.

[HoneyBadger84]

I just noticed something about what you said. Are all of your fans configured as exhaust/pulling air out of the case? If so that is not ideal. Are the top fans on a radiator or just case fans? If they're just case fans, try putting them as intake. I have a top intake fan blowing on my VRM and RAM, it helps quite a bit with thermals. 

[Groda]

I have three fans on the radiator that pull air into the case from the front, the two fans at the top are regular case fans that pull air out of the case and then the fan in the back is also a regular case fan that pulls air out of the case. Had a friend tell me that's a good configuration, but I noticed that the video I was watching where somebody made a build in the same kind of case that I have turned the two fans at the top to be pulling in air as well.

 

Thanks for sharing the image. I'm going to have a look at the adjusting VCCSA and VCCIO during the weekend as well when I'm reseating the CPU and cooler as well as maybe turning the fans around on the top.

[Groda]

So I remounted my cooler and it doesn't seem to have had any effect on the CPU temperatures, it might have even gotten 1-2°C hotter. Though that's hard to judge, could just be ambient temperature or something that makes the difference. After remounting it I got a bluescreen on the regular settings that were previously stable, but reseting bios to default and changing them back again looks to have solved that. I am currently running a bunch of stress tests just to reconfirm that it appears to be stable again before I try changing anything else.