• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

[OC Beginner] MSI Z490 10900K OC help?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

NEW2OC

Registered
Joined
Jun 18, 2022
Hi everyone!

Total OC beginner here (been building PCs a long time but late to the OC game) and I have a few questions...

I just recently received a MSI Z490 Godlike + 10900K, slightly used, as a gift! I'm trying to decide what direction to go in with overclocking for what I do. I will mainly be using this PC for design (Photoshop, Character Creator 3/4, Blender, Unreal Engine, ArcGIS, Wonderdraft, ETC) with a focus on large scale map creation and some rendering, but I will also do some gaming here and there. Should I keep everything on auto except voltages to get my temps down(If so adaptive/fixed)? Or should I use turbo ratio for 5.2, 5.1, 5.0 ghz clocks based on number of cores in use or something like that? Will 5.1 or 5.0 all core be better performance-wise? I've been reading up but I feel like I have more questions now than when I first started researching!

Specs:
Custom case
EVGA 1000 P2
Arctic Freezer II 360(Push/Pull exhausting)
MSI Z490 Godlike (BIOS E7C70IMS.1A0)
i9-10900K
ASUS 2080 TI STRIX
G SKILL TRIDENT Z DDR4 3600 14-15-15-35
2 x 180mm Silverstone Air Pens

I'm attaching some screenshots to show what I'm at currently...

Annotation 2022-06-17 202810.png Annotation 2022-06-17 202829.png AUTO VOLT - 5.1.png CBR23 52-51-50 Ratio.png Cinebench mid run.png MSI_SnapShot_02.jpg MSI_SnapShot_03.jpg MSI_SnapShot_04.jpg MSI_SnapShot_05.jpg
 
Last edited:

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
The 10900K/Z490 has an option called TVB(turbo velocity boost) try enabling that and see where your temperatures sit. Under heavy loads, these are hot CPUs and the temps will be your limitation. You might want to try an actual stress test aside from CBR23 such as AIDA64 FPU test
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
The 10900K/Z490 has an option called TVB(turbo velocity boost) try enabling that and see where your temperatures sit. Under heavy loads, these are hot CPUs and the temps will be your limitation. You might want to try an actual stress test aside from CBR23 such as AIDA64 FPU test
Hello Johan,

I tried running Prime95 small with AVX on and off, AVX it downclocked for some reason to like 4.3 ghz... I assume that means TVB is on? I posted my bios settings but it is kind of a headache to find some of the more obscure settings. Should hyperthreading be turned off or will I lose performance in that case? I would be fine with 5.1 in non-AVX cases and 5.0 or 4.9 with AVX? Is that possible without an offset?

Also, is it even reasonable to test with AVX on in Prime95/OCCT? OCCT also throws errors in most cases that I have anything off auto...

Do I need to change ring to be closer to my highest ratio/or all core clock?

Sorry... I have a lot of questions.

P.S. I've read a lot of guides and most are based on ASUS boards... The bios looks so different between the two. MSI has some terrible guides, most say "set all core to 51 and leave everything else auto!" which is terrible for temps. I am aiming for performance and longevity.
 
Last edited:

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
These CPUs will run to their temperature threshold when left on auto. I'm not familiar with all of the software you listed but Blender will use AVX. AVX can cram a lot more work into each cycle so I wouldn't disable it since it can be beneficial. If you're going to be using this for large projects then I would assume you would prefer stability over performance. When left on auto with TVB enabled you're going to get 99% of the performance your system has to offer plus up to 5.3 GHz in light loads such as gaming. For your usage, I would leave it on auto. It will also automatically downclock to keep the temperatures in line.
As for the ring, it can add a small bit of performance in certain benchmarks but for the most part, you won't see any real gains by increasing the speed.
As you have already noted above setting an all-core OC raises the power usage and temperatures dramatically and also opens the door for instability, in my opinion on the 10900K it really isn't beneficial as that CPU is already a powerhouse and running very near its limits right out of the box.
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
These CPUs will run to their temperature threshold when left on auto. I'm not familiar with all of the software you listed but Blender will use AVX. AVX can cram a lot more work into each cycle so I wouldn't disable it since it can be beneficial. If you're going to be using this for large projects then I would assume you would prefer stability over performance. When left on auto with TVB enabled you're going to get 99% of the performance your system has to offer plus up to 5.3 GHz in light loads such as gaming. For your usage, I would leave it on auto. It will also automatically downclock to keep the temperatures in line.
As for the ring, it can add a small bit of performance in certain benchmarks but for the most part, you won't see any real gains by increasing the speed.
As you have already noted above setting an all-core OC raises the power usage and temperatures dramatically and also opens the door for instability, in my opinion on the 10900K it really isn't beneficial as that CPU is already a powerhouse and running very near its limits right out of the box.
Thank you so much, reading this from you gives me more confidence to stick with mostly auto. Do you recommend just leaving it entirely default or should I go with the 52/2 core, 51 3+? I'm curious as to why my score was under the thread ripper score when I've seen others hit 17k on R23 with 10900k/10850k? If I left my ratios on auto, would I also have to leave voltage on auto or could I set an offset and what would be a good offset to start with?
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
I think I may have found my problem... My Trident Z kit is 3600 cl14 4 x 16gb... Could XMP be causing my instability? Would it be better to down clock to 3200? When I disable XMP I pass OCCT with no errors.
 
Last edited:

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
I think I may have found my problem... My Trident Z kit is 3600 cl14 4 x 16gb... Could XMP be causing my instability? Would it be better to down clock to 3200? When I disable XMP I pass OCCT with no errors.
Maybe. You have four sticks which puts more stress on the IMC than two. Try raising System Agent voltage 0.1V from where it is and see if that helps.

An easy way to see if it's the RAM..........disable XMP and see if the instability still happens.
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
Maybe. You have four sticks which puts more stress on the IMC than two. Try raising System Agent voltage 0.1V from where it is and see if that helps.

An easy way to see if it's the RAM..........disable XMP and see if the instability still happens.

Hello EarthDog!

I'm still having some serious issues even with XMP off. I can pass R23 on loop for 2+ hours and hit scores close to 17k with an all core 5.1 fixed ratio at 1.32 vcore LLC2 and 3. But I'm not sure which is best yet. Hottest core never exceeds 81c. But I've noticed something strange in OCCT. I was pulling 300+ (320ish I think?) in AVX/AVX2 tests and creeping up to the mid 90s... So I capped long/short to 300. I'm really confused now, though. I'm all of a sudden getting really low downclocks (sub 3.7 in some cases) but I'm passing OCCT AVX with no errors. HWINFO shows my highest core clocks hitting 5.1 but when I run a bench in OCCT my highest clock is 4.3 in AVX and standard multi core workloads and my score is well below average. Average is like 300-400 points higher in OCCT and 5.1 max / 3.7 min with vcore min at 1.32 and max 1.42. I am not sure why this is happening, I'm obviously missing a setting in the bios.
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
Let's tackle one thing at a time... the memory.

So, even with XMP off, it's not stable... all four sticks?
Are we talking stable in general stab testing of the CPU? Or memory focused stability testing? If CPU, yes. With XMP off I was still getting errors.

I guess I should just go ahead and run memtest?
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
Again, four sticks add more stress to your IMC, try to raise System Agent voltage (VccSA) 0.1V and then test to see if you are stable.
Ok, how long should I be testing for with AVX on? And is turbo ratio 52 3 core 51 all core ok? My temps are better than auto ratios that way and my score is better than baseline? Or should I go 51 or 50 all core fixed? I immediately thermal throttle in that situation with AVX on.
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
Second time... we need to focus on ONE THING. ;)

If you have a CPU overclock, go back to default on everything. Enable XMP and see if it's stable. If not, add the 0.1V of SA voltage and try again. Once the memory is stable, we'll work on the CPU.

To answer your question about all cores the same or some higher/lower, that's up to you and how you use your PC. Logic dictates that if your use scenarios benefit from all cores being higher, then do that. If you don't need them all at X speed and want faster 'turbo' or a couple of cores faster, go that route. That said, the difference between them isn't much (depending on settings of course). But again, let's focus on the memory and circle back what clocks for the processor. I can't do the ADHD thing, lol.....

Length of testing depends on the user. I find, after initial dialing in testing (which is much shorter) a few hours of my stress test (AIDA64 with CPU/FPU/Cache/Memory) works fine for how I use the PC. If you ask two different people about MY process, one will think it's too little, the other too much... You need to find what works for YOU... but typically a couple/few hours is enough FOR MOST USERS.
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
Stable under default settings with XMP enabled. OCCT AVX on for 15 mins and 30 mins of R23.
So if you're stable with default CPU and memory settings, sounds like whatever you had going on with the CPU was the issue. Now, you can run some tests longer to make sure it's all good (RAM). When you're done, set the RAM to default and work on the CPU. Once the CPU is stable for a couple hours, re-enable the XMP profile and test both. :)
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
So if you're stable with default CPU and memory settings, sounds like whatever you had going on with the CPU was the issue. Now, you can run some tests longer to make sure it's all good (RAM). When you're done, set the RAM to default and work on the CPU. Once the CPU is stable for a couple hours, re-enable the XMP profile and test both. :)

If I stress with the turbo ratios I mentioned above and everything else on auto, and pass said tests... Can I use the auto vcore as a baseline? Or would that not be a good idea? I guess my logic is if the average and max at 52 on 2 cores 51 all core turbo is stable then that should be what I should aim for with a fixed 51 all core ratio? Then dial is back for temperature control?

Thanks for your help. I apologize if I offended you, I'm not a very patient person. 😅
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
You haven't offended me at all. Just trying to keep things focused.

That voltage will likely be too high already, Try at 1.25V with all core 51x and see... at that point it's just playing with the voltages. But I'd bet good money says you can't plug in w/e the top turbo VID is and be under 100C.
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
Max turbo vid was like 1.466v haha. HWINFO is showing a max vcore of 1.361 but through most of R23 and my AVX tests it was hovering around 1.32. It isn't downclocking anymore with the turbo ratios set. I also didn't disable speed shift... Max temp so far has been 96 at 340w, which it spiked to briefly. Doing the 1 hr stability test and it is currently at 82c. It is not a constant AVX workload, though.

Guess I'll finish this stability test and save the profile in case I want to fall back on this one since it is pretty stable it seems. I'm learning! 😆
 
OP
NEW2OC

NEW2OC

Registered
Joined
Jun 18, 2022
You haven't offended me at all. Just trying to keep things focused.

That voltage will likely be too high already, Try at 1.25V with all core 51x and see... at that point it's just playing with the voltages. But I'd bet good money says you can't plug in w/e the top turbo VID is and be under 100C.
I cannot for the life of me figure out why my scores are so low... That 52 1-2 core, 51 all core turbo is beating a fixed 51!? That makes no sense to me when the 5.2 ghz isn't even being taken advantage of in Cinebench R23? And at lower temps!? Here are some screens of 5.1. It is with a .05 offset 51/46 ring... Also, at anything above 46 ring an offset of any kind gives me a DRAM error on the Godlike's led alongside a 0F code. I've been forced to power cycle and clear CMOS a few times now. I'm currently pretty lost and unsure what direction to go in.

R23 HWINFO.jpg R23.png Results.png

Here is the 52 1-2 core, 51 3-10 core turbo test:


52 2 core, 51 3+ turbo ratio - stab test.png HWINFO 51 6 core - 50 10 core stab test, mid run.png
 
Last edited:

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
It's tough for us to compare as you don't have the same information in the image for each. One shows wattage and volts while the other(s) do not.

Worth noting, VID is not Vcore. VID is the voltage ingrained in the processor for a given clock speed (we like to say, Voltage I Demand). VCore is the actual voltage it gets (and the important one).


Just an FYI, you can put both (multiple) apps in the same screenshot. Reduce the massive amount of whitespace in Hwinfo and it's easy. You can also change where things a viewed in Hwinfo so all the information you may want is on one screen.
aio1.jpg