Need help overclocking my 9800X3D

[CrAsHnBuRnXp]

Is it normal for these chips in prime 95 not even using 100% of the CPU?

I also get a lower score in cinebench??

 

[iSpeakVeryWell]

Here are my 0.02$

I think I'm not too far off if I say that you are a beginner in the OC world (sorry if I'm mistaken). So here some insight I think could be useful in your situation.

- Be careful with those video's you posted. Videos like that are a mixed bag, some are "barely good enough", most are really bad and will make you believe it's easy to OC a 9800X3D to 5.6GHz++. It's not. Keep your expectations low for now. Baby step.

- Never forget that too many people like to brag about their unstable OC while not knowing it's not stable. Do you like when your game freeze and crash? Don't trust anybody OC guide. Not even me. Stess test heavily!!

- Know what kind of OC'er you are and don't lie about it. Are you passionate about it? Does it bring you joy when you thinker in the BIOS? Or you just do it for the sake of going from 200 FPS to 206 FPS and you can't stand the process of doing it?

- Start by stess testing and benchmarking your PC **STOCK** first!! Start by doing a factory default. Make sure it's stable before you start OC'ing. Make sure Expo/XMP are enabled tho.

- Take notes. Take screen shot. Keep a folder organized about your experimentations. What worked, what didn't.

- Don't optimize your whole system at once. Choose between CPU OC, RAM OC and GPU OC. Thinker with the option you selected until proven stable. After that, when you happy with the result you got, you can go to another option. I'd do GPU last. CPU > RAM > GPU is my favorite order.

Now since it's a post about OC'ing a CPU, here what I would do about it:

- I Would leave eCLK alone for now. Tweak PBO first. When PBO proven stable, try to figure out your eCLK headroom.

- I would set PBO limit to disabled (not auto! Auto = Motherboard). IMO (and I really mean IMO, I dont want to fight over it) setting it to motherboard is not going to give you much other than more heat.

- I would leave Scalar to auto. IMO, the CPU know best with this one.

- I'd set +200MHz frequency boost.

- I'd set Curve Optimizer All cores -20

- I would not touch Curve Shaper. It's more for those who really want a heavy/extreme OC.

Depending on your cooling and airflow, there is a good chance that those settings are stable. Compare your stock Cinebench R23 with your new score. If you are happy and want to move on, do at least 2h of Prime95 Small FFT.

If your hungry for more, you can try optimize Curve Optimizer, but this is a process of hours and hours of P95 Small FFT. Their is a chance you can go up to -40, but probably not all cores. For example, I'm at -26 on core 0, and -40 on the 7 other cores, but I've spend many hours stess testing the CPU to find out those values.

If you still want more, now a good time to try eCLK.

Remember: OC is very time consuming and it will either bring you joy or piss you off.

 

[rv8000]

Is it normal for these chips in prime 95 not even using 100% of the CPU?

I also get a lower score in cinebench??

You’re likely clock stretching using eclk + PBO with a frequency offset, I experienced this in CB as well when testing with PBO & eclk. I eventually found PBO offset +100 with no eclk and a negative CO provided the best results without throttling in CB, or eclk 103.5-104 with no PBO frequency offset and negative CO; you might have better results with an AIO (I was definitely thermally limited in CB on an air cooler).

Running your memory at 7000 is the worst of both worlds unfortunately. You’re either dropping or desyncing your IF clock unless you’re forcing 2000 or higher, either way there is a memory latency penalty in this frequency range. If you’re going to run 2:1 for higher memory frequency you need to be running 7800 at a minimum, 8000 preferred to sync back up with the IF. Most times it’s easier and better to shoot for 6000, 6200, or 6400 with your IF at 2000, 2066, or 2133 respectively, at or around c26-c30. Those kind of primary timings should be doable at or with less ~1.5vdd on most a-die kits or m-die kits. Checking stability can be a bit of a rabbit hole if you’re looking to get as tight of timings as possible. I can recommend some timings if you list your specific kit, just remember you’ll need to check for stability without PBO/eclk/or any kind of CPU changes separately.

I haven’t done much testing with PBO + eclk in WoW since finding out 2200 IF was what was causing instability, but eclk may have been negatively impacting performance when I was testing earlier this week; I’d also recommend checking performance with PBO enabled and eclk off and vice versa.

 

[iSpeakVeryWell]

Is it normal for these chips in prime 95 not even using 100% of the CPU?

It is not. Small FFT should max out the CPU.

 

[JustBenching]

I've seen around an 11-12% performance increase with tuned memory on a 9800x 3d, but that's at a forced CPU bottleneck (1080 dlss performance). I've seen up to 35% on my 12900k (ratchet and clank for example)

Is it normal for these chips in prime 95 not even using 100% of the CPU?

I also get a lower score in cinebench??

Don't use prime, I think ycruncher is better for stability.

Also PBO + 200 gives you exactly a 4% fps increase (at extremely low settings) but the power draw goes up by 55%. It turns it from a chill gaming CPU to a flame throwing monster.

I would also steer as far away as possible from skaterbenchers videos, he is getting cherrypicked samples and pushes cpus to the point that they will degrade in a couple of days, but he doesn't care, he is not paying them out of pocket.

 

[Contra]

mem frequency is set to DDR5 7000 which is set by my docp1 AI Overclock tuner option

It's not optimal.
- change to manual. Downclock you ram to 6400MT, set Infinity fabric to 2133MHz, set Memory controller = uncore = 3200MHz, and set lowest timing for your voltage (CL 28-32 etc)