Manually Overclocking Ryzen 2 3900X

[noko]

While I don't normally like to OC using Windows software, this is working out relatively well. Since some call the 3900X having one Chiplet and one Shitlet, OCing those differently may allow better performance overall. Anyways this is my first run using Ryzen Master to OC CCD0 to 4.5ghz and CCD1 (Shitlet) to 4.3ghz. CPU voltage adjusted to 1.35v and in the bios using LLC 4. I will shoot for higher clocks later depending upon how well I can cool the CPU. Here is the settings in Ryzen Master:

​

Video in how the cores respond when rendering - note when no load is on the core they do down clock and sleep as well:



This method may prove very fruitful.

 

[tangoseal]

I'm not sure manually overclocking is best anymore. Your generating heat and using excess energy for a cpu that sits idle 95% of the time. Unless your firing it up and committing CPU all night long every day to intensive tasking.

However the enthusiast in me appreciates your work and hope you enjoy it.

 

[Xenozx]

your scores are awesome. I score 7400 in r20 on a x470 board but I didnt manually overclock. I simply changed the BCLK to 104 and left everything else auto for the most part. The cool thing about this, is it will drop down to like 1.1v when not being used much and 3.4ghz. as soon as you load up a game, different cores will hit different clocks, but usulaly anywhere between 4.3ghz and 4.7ghz. If i do a stress test of say r20 that puts every thread to use, it will pretty much do an all core of 4.225 ghz. sure for benchmarking purposes 4.5 and 4.3 beats 4.225 all day, but how often when your using your computer do you really max out 24 threads? I ran a test on 12 threads and I get 4.40ghz under load. 8 thread 4.5ghz.

 

[noko]

I'm not sure manually overclocking is best anymore. Your generating heat and using excess energy for a cpu that sits idle 95% of the time. Unless your firing it up and committing CPU all night long every day to intensive tasking.

However the enthusiast in me appreciates your work and hope you enjoy it.

I agree, just exploring now. I was close to breaking 8000 in CB20 but had to do other stuff. Planning on finding overall ceiling for each core. Since Ryzen Master you can OC per core.

For my mother board any front side bus changes turns PB2 off and will be stuck at 3.8 ghz unfortunately.

 

[Xenozx]

I agree, just exploring now. I was close to breaking 8000 in CB20 but had to do other stuff. Planning on finding overall ceiling for each core. Since Ryzen Master you can OC per core.

For my mother board any front side bus changes turns PB2 off and will be stuck at 3.8 ghz unfortunately.

Mine does too until I load ryzen master and activate a profile then the 3.8ghz unlocks!

 

[noko]

Mine does too until I load ryzen master and activate a profile then the 3.8ghz unlocks!

Very cool! Did not know that, might be able to push the memory 1:1 higher than 3800mhz is my first thought. Thanks!

There is hell allot of new OCing methods which seems like very few are exploring with Ryzen2. So much could be written about and explained but is not by most sites. Now the Enthusiasts in the various forums are exploring and sharing results which is cool.

 

[Darth Ender]

staying at a static frequency vs scaling is not as bad as it sounds, especially when the scaling algorithm being used by the operating system just rockets the frequency up to max whenever the cpu gets even a little bit of activity since when it's sleeping, it's pretty much off in either situation. idle cpu temps and power usages are barely different.

All core 4.3 manual overclock ends up having me hit 80C ....and most other benchmarks / stress testing seems to max at around there ...so i'm fairly happy with that. Compared to PBO ....which tended to only allow such speeds under particularly light actual loading on all cores. As for not having 4.6Ghz available to 1 core (2 threads), meh. I do not tend to have single thread high cpu tasks as often as others I guess. Compiling on all cores is a far more common task I do so this is more ideal.

when booted to windows, my r20 scores are over 7400 ... with my old 3200mhz ram.

 

[Digital Viper-X-]

I'm not sure manually overclocking is best anymore. Your generating heat and using excess energy for a cpu that sits idle 95% of the time. Unless your firing it up and committing CPU all night long every day to intensive tasking.

However the enthusiast in me appreciates your work and hope you enjoy it.

you mean the cpu doesn't throttle down on low workloads or idle when manually ocing?

 

[Jamie Marsala]

you mean the cpu doesn't throttle down on low workloads or idle when manually ocing?

No the CPU is locked at what you set it at. When I first got my 3900x I set it to 4.3 all core in the BIOS, I tired 4.4 but it would not boot. I did not fiddle with voltages. It sat at 4.3 all day long without ever moving. I set it back to normal and it would clock to 4.25 at boot but raise and lower as needed. On a single thread I could get 4.3 out of it during testing and about 3.9-4.1 all core depending on the test. I then upgraded from a B450 board to an X570 board with the latest BIOS, just yesterday, and I can get 4.55 single core and 4.2 all core.

 

[Jamie Marsala]

According to HWInfo64 it seems somewhere there is a setting that once the chip hits 80C that it is limited to 4400. Next reboot I will check to see if it can be set higher temp wise. I have hit 4.55 on several cores running AIDA64 benches and when running Cinebench R20 Currently I am only set to run to 4650 Boost so I am going to see what happens if I bump it to 4850 and see if it does anything or if I just need better cooling than the Noctua.

 

[kamikazi]

I tried messing around with all core OC today. With CPU LLC at 3 and voltage at 1.35, 4.3 worked pretty well. I tried one chipset at 4.5 and the other at 4.3 and that was a no go, reboot. I raised voltage to 1.35675 and it wouldn't run Cinebench due to errors.

I'm on the newest BIOS and still haven't seen 4.6 yet. I've seen the 46 multiplier on one core, but FSB is 99.8, so it will never get there. I guess I could tweek the FSB for the e-peen, but it's not worth the trouble.

 

[noko]

Had more time to do more testing:

• CCD 0 (6 cores), every core separately could do 4.6ghz and pass AMD Ryzen Master stress test at 1.4v. None tested could do 4.65ghz with 1.4v, even tried a few at 1.45v without success
• CCD 1 (6cores), only one core could do 4.5ghz at 1.4v and all could do 4.4ghz
• To cut it short or end result for all cores, which would add more heat, drop voltage more, I went with 1.425v for testing sakes and did not do much better then before
  • Ended up with 4.525ghz CCD 0, 4.375ghz CCD 1
  • Anything higher was not particularly stable and even this was not tested fully
  • What surprised me was stability was not increased at all when decreasing memory speed to 3600mhz in attempts to get the CCD's to OC higher
    
  • Tried to get to over 8000 pts in CB20 but after many attempts that proved very elusive
    • 
• I also tried different core clocks in the same CCD, as in 4.6ghz on one or more cores, example the ones that PB2 selects for boosting. Any core speed different from the others, even if only a minor bump would cause any restart to hang multiple times
  • I just stopped testing that avenue due to how much time it was taking getting the system up each time it crashed
  • Other motherboards may have better success there, bios update or AGESA updates may improve that
  • Since each CCD cores in my 3900 X sample actually are relatively close in performance to each other I just concluded in my case it was more wasted effort then potential benefit
• I then went with base clock over clocking using PB2, reduced memory speeds and just using 102 mhz for fsb I was getting blue screen after blue screen. This was something on first gen Ryzen that worked particularly well for me
  • I will probably pursue this more in the future

In conclusion:

• For specific cases or work loads you can potentially get more performance by clocking each CCD to their respective highest stable Over Clock depending upon your CPU (this would be the 3900x and 3950x which has two cpu chiplets)
• In my case that improvement for workload like CB20 went from 7500 to 7976 a whole 6% improvement at best, not fully tested as stable
• I went back to PB2, I think AMD did a great job in extracting as much performance automatically from Zen 2 while maintaining stability, lower heat and power costs
• As the platform matures, bios, software etc. and maybe if I get another motherboard - I will most likely retest in the future

 

[kamikazi]

Had more time to do more testing:

• CCD 0 (6 cores), every core separately could do 4.6ghz and pass AMD Ryzen Master stress test at 1.4v. None tested could do 4.65ghz with 1.4v, even tried a few at 1.45v without success
• CCD 1 (6cores), only one core could do 4.5ghz at 1.4v and all could do 4.4ghz
• To cut it short or end result for all cores, which would add more heat, drop voltage more, I went with 1.425v for testing sakes and did not do much better then before
  • Ended up with 4.525ghz CCD 0, 4.375ghz CCD 1
  • Anything higher was not particularly stable and even this was not tested fully
  • What surprised me was stability was not increased at all when decreasing memory speed to 3600mhz in attempts to get the CCD's to OC higher
    
  • Tried to get to over 8000 pts in CB20 but after many attempts that proved very elusive
    • View attachment 190322
• I also tried different core clocks in the same CCD, as in 4.6ghz on one or more cores, example the ones that PB2 selects for boosting. Any core speed different from the others, even if only a minor bump would cause any restart to hang multiple times
  • I just stopped testing that avenue due to how much time it was taking getting the system up each time it crashed
  • Other motherboards may have better success there, bios update or AGESA updates may improve that
  • Since each CCD cores in my 3900 X sample actually are relatively close in performance to each other I just concluded in my case it was more wasted effort then potential benefit
• I then went with base clock over clocking using PB2, reduced memory speeds and just using 102 mhz for fsb I was getting blue screen after blue screen. This was something on first gen Ryzen that worked particularly well for me
  • I will probably pursue this more in the future

In conclusion:

• For specific cases or work loads you can potentially get more performance by clocking each CCD to their respective highest stable Over Clock depending upon your CPU (this would be the 3900x and 3950x which has two cpu chiplets)
• In my case that improvement for workload like CB20 went from 7500 to 7976 a whole 6% improvement at best, not fully tested as stable
• I went back to PB2, I think AMD did a great job in extracting as much performance automatically from Zen 2 while maintaining stability, lower heat and power costs
• As the platform matures, bios, software etc. and maybe if I get another motherboard - I will most likely retest in the future

So you were getting 7500 with all cores at 4.3, then got close to 8000 when you pushed CCD 0 to 4.525 and CCD 1 to 4.375? What is your stock score with no PB2, no tweaks? I'm looking at 7263 there.

 

[noko]

So you were getting 7500 with all cores at 4.3, then got close to 8000 when you pushed CCD 0 to 4.525 and CCD 1 to 4.375? What is your stock score with no PB2, no tweaks? I'm looking at 7263 there.

When using PB2, I get around 7500+ with a modified bios, manually I can do better but I don't think the benefits of the performance increase is worth it.

Newest bios has all core boost somewhat lower so more in the mid 74xx range. Never tested all cores doing 4.3ghz. Here is what a modified previous bios did just using PB2:

 

[Neapolitan6th]

Thanks for sharing your testing. I was always curious on the per core OC headroom Ryzen has. I definitely seems like 1 bad ccx is what holds back most all core OCs.