Switching from 2 sticks of b-die to 4 sticks was/is an adventure. I never knew how easy I had it with 2 sticks until I tried 4. I won't go back, but the 2 stick easy tweaking will be missed.
4 sticks defaults to gear down mode. Trying to run with CR1 and GDM disabled is unstable. So far it's looking like CR2 is the lesser of 2 evils when it comes to effecting latency. I'm still playing around with the memory, but I'll settle for 3600 CL14 if I have to.
View attachment 35797
Welcome to my world
I'm just happy ive found my CL14 memtest 1000% stable settings with 4x8gigs @ 1900mhz IF 1:1
If some of you other guys want to start playing around with the curve optimizer, i can recommend this as a starting point if you are using a Asus motherboard:
Make sure you have
Global C-state Control disabled in bios before you start.
In Extreme Tweaker:
- PBO Fmax Enhancer: disabled (dont work with 5000 series)
- PBO: manual: 280/235/245
- Overdrive: +50 mhz
- CPU SoC voltage: Manual 1.08 -1.10v
- CPU Core Voltage: Offset 0.01 -0.06v (You want the least amount of offset as possible. Most will be @ +0.05v)
In Digi+
- Max all the current limits
- Disable all the spread spectrums
- SOC LLC: Level 3
- CPU LLC: Level 1 or Level 2
Note: You want the all core voltage to be as low as possible to get the highest all core boost that you can. My logic here is that I am using core voltage offset to maintain enough voltage for a stable single core boost, but the lowest LLC so that on all core loads the core voltage will droop and allow a higher all core boost clock.
In Advanced> AMD Overclocking > PBO
- Set to advanced
- Set limits to "Motherboard"
- Curve optimizer: All core, Negative, 15-25 (can start at 15 and work your way down to 25 if system is stable)*
If you have a 56/57/58/5900X You use the same settings but increase the overdrive setting from 50mhz to as high as your single core test will allow. Most CPU's will max out at a sustained 5050mhz and 5125mhz single core boost.
To test single core boost clock:
- Open HWinfo64, make sure you have the "Effective Core clock" registers exposed.
- Make a note of the top 4 ranked cores.
- Open task manager and to to the "Details" tab.
- Open CB R20, start a single core run
- In Task manager's details, right click Cinebench.exe, click "Select Affinity", uncheck the top box to de-select all cores, put a check box next to core 0, then click ok. (You will have to re-select the affinity every run unless you have an application such as Process lasso that will automatically apply core affinity.)
- In HWinfo, Look at the clock speed for core zero. It should be at your max boost + overdrive (for a 5950X using the settings above, that would be 5050mhz max boost + 50 mhz overdrive = 5100mhz).
- In HWinfo, monitor the "Effective core speed" It should be very near reported core clock. For example, the core clock is at 5100mhz, the effective core clock should be 5085mhz or higher. Generally effective clock will be reported a bit lower than core clock because of how it is calculated. (which is why the effective core clock really needs a sustained load to be close to accurate).
- Monitor Core voltage and effective core VID. Your core voltage should be 1.5v- 1.525v, and effective VID 1.485v - 1.506v. You really don't want core voltage to exceed 1.525v for sustained period (spikes are fine and normal) This is why your core voltage offset shouldn't exceed +0.05 - 0.07v.
- You should be able to complete at least 5-6 back to back runs of CB R20 without crashing.
Once that is done, you move on to the next three of the top rated cores and make sure they are stable. Though it is likely that not all of the top four cores will boost as high at core 0.
Once that is complete move on to all core load stress testing. If an all core workload does not pass stress testing, increase core LLC to achieve stability.
*edit*
* = You can start with -15 on core optimizer with cpu voltage on auto, then work the negative curve offset down until you have all core instability, or until the all core boost clock stops increasing. There is no point in running the neg curve value beyond whatever gives you max all core boost..
Once you have your all core set to highest stable neg offset / highest all core boost clock, Move on to single core.
Open cinebench and open task manager. Start a single core run, then on task manager's details tab, find cinebench, right click and set cpu affinity. Leave only core zero checked.
In hwinfo, click the clock to reset the counters after the single core run starts, and after you set cpu affinity. Then monitor the core 0 effective clock. It should be very close or the same as core frequency.
If the run fails, Start adding positive cpu core voltage offset to get your single core stable.
Once done stability test with IBT linpack on very high/high for 10 passes, Blender Benchmark (all scenes), 1 hour of P95 blend with AVX disabled.