Overclocking Advise

[ek5932]

Hello All,

This weekend I impulse bought a new 9900k and Aurous master, cooled by a Corsair H100i. I'm pretty new to overclocking but I have everything up and running - clocked the CPU to 5GHZ (4.7 uncore) @ 1.3 vcore. I'm happy with everything but wanted some advise on a few of things:

I think my overclock is pretty decent; ~mid 30's (C) idle temps and generally under 60 when running Prime95 blend. Do you think it's safe to it a bit push harder? The OC guide I followed stated I shouldn't go over 1.35 vcore; am I safe to get the fastest clock speed at that, as long as I'm getting decent temps?

Secondly, when running blend [in prime] and gaming my CPU temps are OK but the Small FTTs test kills my PC instantly. Is this because it runs AVX commands? There's an offset in my Bios, should I configure that to keep the test stable?

Lastly, in my bios settings for XMP I only have a single profile option which I selected, with the correct speed for my memory (3200) and everything else is set to auto. Is this enough? I have read something about tight/loose timings which is more important then speed. Is this something I need to worry about? It's a gskill 16gb kit.

Thanks in advance! It's fun to start pushing my system.

 

[JCLDJB]

Can you post full specs please? Like PSU as exactly which type of RAM?

Hello All,

This weekend I impulse bought a new 9900k and Aurous master, cooled by a Corsair H100i. I'm pretty new to overclocking but I have everything up and running - clocked the CPU to 5GHZ (4.7 uncore) @ 1.3 vcore. I'm happy with everything but wanted some advise on a few of things:

I think my overclock is pretty decent; ~mid 30's (C) idle temps and generally under 60 when running Prime95 blend. Do you think it's safe to it a bit push harder? The OC guide I followed stated I shouldn't go over 1.35 vcore; am I safe to get the fastest clock speed at that, as long as I'm getting decent temps?

Secondly, when running blend [in prime] and gaming my CPU temps are OK but the Small FTTs test kills my PC instantly. Is this because it runs AVX commands? There's an offset in my Bios, should I configure that to keep the test stable?

Lastly, in my bios settings for XMP I only have a single profile option which I selected, with the correct speed for my memory (3200) and everything else is set to auto. Is this enough? I have read something about tight/loose timings which is more important then speed. Is this something I need to worry about? It's a gskill 16gb kit.

Thanks in advance! It's fun to start pushing my system.

I agree, it is fun to push a system.

First of all, if you run Small FTTs and it 'kills' your PC instantly - do you mean it freezes/stops working? That means your system isn't actually 100% stable and you may need to increase the voltage a bit so that it becomes stable.

Yes, you can try changing the AVX mode, that may help.

And, when you say 60C under full load running Prime95 blend, how long did you let it run that way? Because it can take a good 20-40mins for the liquid in the AIO to fully heat up, only then you get your real temps.

Keep your voltage under 1.4v max (but for daily use, the lower the better...so ~1.35v is probably a good baseline).

I personally don't like Prime95 as it just feels too much of an 'extreme' case; but saying that, if you can hold Small FTTs for longer than 20-30mins then you know for sure that anything else you throw at it (like gaming/3d rendering/video editing) will run just fine).


As for Memory, before you start messing with memory timings, I strongly suggest you read up about:

- How RAM actually works (GamersNexus comes to mind as they explain it really well.) Just google: <<gamersnexus memory timings>> and it should come up.
- Learn in advance, what you need to do if you accidentally choose a memory timing that your system cannot recover from (most motherboards come with a failsafe, meaning after 1-3 failed boots, it reverts back to stock Bios settings) - BUT I have had a systeme not revert back because of a failed memory timing overclock and I had to remove 1 stick of RAM, or even both, and put a new stick in, just to get the system booted..
- Frequency VS Memory Timings, ideally you want to get to your desired frequency first, then reduce timings afterwards, getting them as low as possible while keeping a system stable (set up Memtest on a USB boot drive in advance so you can stability test your RAM overclocks).

Enjoy the overclocking and post back for any questions.

If any others would care to share their thoughts, by all means.


[EDIT] Oh and also get Cinebench along with the FPS results of a few games you play (using their in-game benchmarks) for a before/after comparison...so you can see how much of an improvement your overclock actually is. Aways fun to see.

 

[ek5932]

Hey, thanks for taking the time to reply.

I've tried getting Small FTTs stable but my PC still crashes pretty much instantly - even with my vCore set to 1.4. If a set a large AVX offset of 10 then it appears OK but looking in CPUz I get odd fluctuations down to 4ghz (from 5) even when I'm not performing AVX load. I'd rather not have these fluctuation - do I really need the test to pass? When I run blend for 20 minutes my temps are at ~85 with no crashes so the build seems stable enough. And if Prime isn't a true representation of a proper use case and my PC isn't freezing under normal load does it really matter?

Also, sounds like I have some research to do for my memory clocking. One thing I'm glad you mentioned is the Bios settings reverting settings after a number of fails. I thought this was some weird behaviour with my motherboard but now this makes sense.
The whole process sounds a bit scarier than I thought it would be, but I'll read up and see how I go!

 

[JCLDJB]

Hey, thanks for taking the time to reply.

....and if Prime isn't a true representation of a proper use case and my PC isn't freezing under normal load does it really matter?

It matters only if you want to claim that you have a stable 5Ghz overclock on a 9900k

And that was just my personal opinion about Prime95. =)

There is no 'Prime95 law' that says you must reach full stability, but let me put it to you this way, if you didn't overclock your CPU and ran Prime95, it wouldn't crash.

Running P95 and getting a crash means there is 'some kind of instability' with the CPU, and at the end of the day, Prime95 is one of the most basic programs to test your CPU's computing capability (which is why it gets so hot compared to other programs).

And also, because there really is no bar to determine (how unstable a CPU Overclock is...20% unstable?....50% unstable?..etc..) it is either stable, or unstable.

Saying that, it's good that you were able to boot and even game at a 5Ghz Overclock (some CPUs don't even boot at 5Ghz, no matter the voltage).

But if you're still looking for some kind of reassurance of CPU stability this is what I recommend:
Run a game's benchmark (the newer the game, the better), then while that's running in a Window in the backgroud, run Cinebench and see if it crashes.

For me, this is enough to tell me if my system is good enough to use on a day-to-day basis.
It is, howeever, not enough for me to claim online, that my overclock is stable.

For assured stability I do a lot more thorough testing as I am, really, trying to PROVE stability.
My 'stability confirmation suite' consists of:
Cinebench, Unigine's Superposition, 3-4 passes of Memtest (or about 8-12 hours of runs), and a combination of CPU heavy applications all running at once, like a 3d Render with Blender, with Cinebench in background while gaming at 4k...and yes Prime95 from time to time, but on rare occasions.

If you are looking for absolute stability:
Can you get a stable run of Prime95 with 4.8-4.9Ghz wit x.xx voltage? If you don't want to use Prime95, just try Cinebench with a Game running in background (or run Cinebench then ALT+TAB into a game and 'play' it for the time it would take for the Cinebench run to complete).

[edit]
I just remembered, if you don't want to use Prime95, you can try RealBench ROG by ASUS.

 

[ek5932]

RealBench yields better results than Prime. My temps go upto 100 though, but the system doesn't crash which is a good thing ^^

Granted my system may not be 100% stable but I think I'm content with its current state. Sure, some synthetic test doesn't pass but as long as the system is stable in real world scenarios I'm happy. I think I'd rather have the 5hz for what I do use, then to lose a few hz to be 100% stable.

I also understand that I shouldn't go around saying my system is stable now as well

Thanks for taking the time to help - I really appreciate it.

 

[JCLDJB]

RealBench yields better results than Prime. My temps go upto 100 though, but the system doesn't crash which is a good thing ^^

Granted my system may not be 100% stable but I think I'm content with its current state. Sure, some synthetic test doesn't pass but as long as the system is stable in real world scenarios I'm happy. I think I'd rather have the 5hz for what I do use, then to lose a few hz to be 100% stable.

I also understand that I shouldn't go around saying my system is stable now as well

Thanks for taking the time to help - I really appreciate it.

Happy to help.

Damn dude, 100C sounds toasty..but Realbench is a beast.
Curious to know why your system didn't shutdown (thermal protection) at that temp.

Now the following few points are things to keep in mind when having a system OC'ed 24/7:

Wondering what the real temps are like on your VRM's...

This is something to watch as many just assume the motherboard can take sustained loads...but I have had a few go on me because I pushed things too far and for too long.. (consider getting a small fan for your VRMs as it will help with cooling cuz the last thing anyone wants for their system is to have a board fail in few months time cuz of too much load on mobo's VRMs/Mosfets etc..)

Hwinfo can help you guage what kind of power is going through them (voltage/amps).

Just bear that in mind and have fun overclocking!