Ryzen 5900X overclocking help

I managed to get my IF to 2000 with memory @ 4000 pretty easily (though I'm running on a DH, not sure if that helps or not)
 
It seems people on dark hero have better luck with IF @ 2000. Mine doesn’t go above 1867 but I haven’t tried new bios properly yet. Will do it tonight.

Did you see significant improvement in benches and are you using curve optimizer?
 
It seems people on dark hero have better luck with IF @ 2000. Mine doesn’t go above 1867 but I haven’t tried new bios properly yet. Will do it tonight.

Did you see significant improvement in benches and are you using curve optimizer?

I have only tested CB R20 at the moment, 0 improvment.
I did run a bandwidth test,

1800/3600 @ 14-14-14-14 : 51.xgb read / 49.x gb write
vs
2000/4000 @ 17-18-18-18 : 55gb read / 51.x gb write

I do notice the temps are 1-2 degrees higher with IF @ 2000 as well (maybe due to more SOC voltage)
 
I have only tested CB R20 at the moment, 0 improvment.
I did run a bandwidth test,

1800/3600 @ 14-14-14-14 : 51.xgb read / 49.x gb write
vs
2000/4000 @ 17-18-18-18 : 55gb read / 51.x gb write

I do notice the temps are 1-2 degrees higher with IF @ 2000 as well (maybe due to more SOC voltage)
what about RAM latency? thanks in advance :)

What I found suprising with the limited time, testing and knowledge, I thought the RAM latency will improve when going from 3600CL18 to 3800CL16. Both read/write speeds did improve, however if I'm not mistaken it only improved 1ns (63ns vs 62ns).
 
My personal opinion after wasting a lot of time with this setup is to just leave it alone. It will show some points here and there in benches but in actual games it doesn’t make squat worth of difference.

Unlike Intel where going from 4.4 GHz to 5 GHz was a real difference in performance at 1440P gaming which is what I do.
 
what about RAM latency? thanks in advance :)

What I found suprising with the limited time, testing and knowledge, I thought the RAM latency will improve when going from 3600CL18 to 3800CL16. Both read/write speeds did improve, however if I'm not mistaken it only improved 1ns (63ns vs 62ns).

What test do you want me to run?
 
I literally give up now. I have no way to explain anymore wtf is going on with my computer.

This is after putting rad on top and again playing 30 minutes of Immortals Fenyx Rising. This computer has a complete mind of its own.

View attachment 315480
View attachment 315481

As you can see CPU is now running at 71 C. However, everything else has cooled down including RAM, GPU and all components. Also the Auxfan4 didn't turn on. I am actually doing nothing different except changing rad position plus playing Immortals Fenyx Rising in exact same area (area 2 for 30 minutes each time).

I am sticking with what I have now and calling it a day. It is ridiculous. Will play some more when new bios is out.
Stop using cpu monitor? Only accurate monitoring software is hwinfo.
 

There ya go, It would not let me set the CL to 17, which is what this memory is setup for, possibly a bug in the bios:

cachemem.png
 
There ya go, It would not let me set the CL to 17, which is what this memory is setup for, possibly a bug in the bios:

Check your BIOS for "Gear Down Mode" setting. If that is enabled, that is why your CL would run at 18.

And no, the sensors are the same but the software is not always accurate. HWInfo is the correct software to use for temps...or Ryzen Master.
 
Check your BIOS for "Gear Down Mode" setting. If that is enabled, that is why your CL would run at 18.

And no, the sensors are the same but the software is not always accurate. HWInfo is the correct software to use for temps...or Ryzen Master.

I have both, and both report the same temps FYI
 
Pretty sure they all read the same sensors :)
It's not about the sensors but reading and compatibility with the Zen design. Most apps are doing it wrong and are creating observer effects by the very nature of how and often they are polling for that data.
 
very correct.



>>It seems people on dark hero have better luck with IF @ 2000. Mine doesn’t go above 1867 but I haven’t tried new bios properly yet. Will do it tonight.

KickAsscop, how did it fare ? It's couple of days already and you didn't mention anything.
When you tried IF overclocking, what were your CCD and IO Voltages ? Set them to 1.15V, don't leave at Auto. Configure these voltages, save, reboot, only then try to reach 1900MHz IF.

To try IF 1900MHz = DDR4 3800 frequencies, set VERY RELATED timings first before attempting to set 3800MHz memory frequency (something like 20-20-20-46) because that way, you will be absolutely sure what causes your problems - either InfinityFabric itself, or memory modules. People tend to blame IF too often when actually it's their RAM unable to keep up with timings at set frequencies.

Also, your score of 8300 or 8500 Cinebench R20 mentioned on first page is very good - default configuration nets around 8000-8100 points whereas 4.8GHz overclocked provide around 8700-9000 points depending on several factors such as cooling which includes case ventilation [!] so it's not just "I have H150i Pro" factor.

2000MHz IF is not guaranteed by any means and only will be achievable with newest AGESA. For some boards, that might be end of January only. Still, don't have too much hopes for 2000MHz IF, that will be premium league exactly as 1933MHz is premium with Zen2.


Your CPU boosting to 5.15GHz single-core should be one of THE BEST cpus. Seriously golden. I haven't seen anyone mentioning 5.15GHz in any review etc.
 
Thanks for the inputs. I did try the 1900 MHz with default ram timings of 16-19-19-39 and it didn’t boot. I did give it 1.1 volts on soc and ram 1.45 volts but no dice.

I tried curve optimizer since it works now and got about -15 on best and better cores and -10 on everything else. I was able to score about 8600-8700 CBR20.
Got my Geekbench from 1500/14K to 1605/14.3K. https://browser.geekbench.com/v5/cpu/5791246
Got my CPUZ from 600/9200 -> 672/9594. https://valid.x86.fr/p4d4sm
Got my Timespy up from 17.87K -> 17.909K https://www.3dmark.com/spy/17236849.

I think my ram is just garbage and that is why I have not had any luck with clocking this processor further. Since I have to take a screw driver every time I mess up the clocks I give up actually.

5.15 boost is only SC. My multi boost is typically between 4.6-4.9 GHz as observed in Ryzen Master. In CB it is usually 4.5 and backs down to 4.45 towards the end.

I haven’t tried full manual OC of 4.7 GHz which is what most CPUs seem to hit.
 
Last edited:
Generally speaking you want more voltage going to your best cores than your worst ones, because the bad cores will never boost high-- the Windows scheduler will only clock (and volt) up the best CPUs in lightly threaded workloads. That said if you tested it out and are happy with your single-threaded scores, you're golden there. I would test -15 (or even lower, I use -20 myself) on your bad cores though, that would likely help your multi-core performance a bit.

Have you tried DRAM calculator for your RAM timings?
 
Thanks for the inputs. I did try the 1900 MHz with default ram timings of 16-19-19-39 and it didn’t boot. I did give it 1.1 volts on soc and ram 1.45 volts but no dice.

I tried curve optimizer since it works now and got about -15 on best and better cores and -10 on everything else. I was able to score about 8600-8700 CBR20.
Got my Geekbench from 1500/14K to 1605/14.3K. https://browser.geekbench.com/v5/cpu/5791246
Got my CPUZ from 600/9200 -> 672/9594. https://valid.x86.fr/p4d4sm
Got my Timespy up from 17.87K -> 17.909K https://www.3dmark.com/spy/17236849.

I think my ram is just garbage and that is why I have not had any luck with clocking this processor further. Since I have to take a screw driver every time I mess up the clocks I give up actually.

5.15 boost is only SC. My multi boost is typically between 4.6-4.9 GHz as observed in Ryzen Master. In CB it is usually 4.5 and backs down to 4.45 towards the end.

I haven’t tried full manual OC of 4.7 GHz which is what most CPUs seem to hit.
Is curve optimizer better using BIOS v153 over v151? I didn't have much luck with v151 and curve optimizer.
 
It depends on AGESA version more than BIOS version. AGESA 1.1.9.0 seems a lot better than previous ones, particularly with dual CCD Zen 3. I had a ton of black screen WHEA reboots before upgrading to that beta BIOS.
 
I couldn’t get it to work previously. Now it does work. So I would say yes it is worth upgrading to 153 bios.
 
My 5900X came in a few days ago, popped it in my Gigabyte X570 Pro WiFi and just enabled PBO. I also tweaked my B-Die ram to 3600 at CL16. Not sure if these scores are good lol. I don’t think I want to play around with the curve optimizer and all that stuff, it’s hitting almost 5GHz anyways.

1610344707935.png
 
Last edited:
My 5900X came in a few days ago, popped it in my Gigabyte X570 Pro WiFi and just enabled PBO. I also tweaked my B-Die ram to 3600 at CL16. Not sure if these scores are good lol. I don’t think I want to play around with the curve optimizer and all that stuff, it’s hitting almost 5GHz anyways.

View attachment 317945
Seems fine to me. My bandwidth is about 56.7 Gs but latency is about 64 ns due to loose timings. In bigger scheme of things it is worth no difference except synthetic benches.
 
So been trying to play around a bit more with the Ryzen build.

I finally got the Ryzen master to work without reboots. The way to do it was to install it and then change to manual OC and then forget about it. If you keep on changing between default, auto OC and manual OC it will go into a spiral where it will request a reboot for all new settings that it wants to apply.

Given this I tested some manual OC on this CPU. It is not a great CPU but I did try some new stuff whilst keeping IF at 1867 and DDR4 at 3733 as anything higher has not worked for me thus far.

I got the CPU stable up to 4.7 Gs all core w/ 1.375 volts. This resulted in a 9100 CBR20 score. At 4.625 Gs all core I could go as low as 1.3 volts and this resulted in a 8900 CBR20 score.

Similarly on Timespy CPU test I was able to push the CPU score to about 14842 which is the highest I have seen.

Another thing I learnt is that CPU HWMonitor does not report voltages and temps correctly. I tried a few things and could see that even when I set the voltage to 1.25 in the bios CPU HWMonitor showed it as always at 1.25 w/ some minor changes but Ryzen Master was reporting it bouncing between 0.3-1.35 volts still when running benches. So basically the override mode does not work on this CPU OR just turning on Ryzen Master is overriding the information about volts and letting it run defaults. However, I believe it is the first case that Override mode in bios is bugged. On temps CPU HWMonitor had delta temps of 1 C - 5 C higher than what Ryzen Master was reporting at a point in time. Same for volts.

temp-delta.jpg

Armed with this knowledge I have gone back to AMD Overclocking and enabled Curve Optimizer yet again with override mode of -100 mV. This keeps my CPU under 80 C for heavy loads. Not sure if this helps someone but I am dumping the HWMonitor program completely for this build. Btw HWInfo also suffers similar problems. If you really want to monitor temps and volts the only place is Ryzen Master out of the 3 programs. I also managed to improve some of my secondary memory timings and brought down my latency from 64 ns to 61.7 ns.

I will mess around w/ some additional voltage (maybe 1.4 volts) and see if I can get 4.75 Gs out of this puppy.
 
Last edited:
I figured out my problem with PBO on my asus board.

There is PBO and PBO2. Pick one or the other. Whenever I tried AI PBO, it was basically stock or any tweaks lost performance or was just not there. PBO2 on the other page is separate and with recent BIOS updates has started working as advertised.

PBO1 seems to have any manual boost available, realistically I couldn't get even 225mhz overboost to work with it.

PBO2 is where the power limits are kept and curve optimizer. This is the one that works now.
Curve optimizer seems to be inverted (negative adds voltage and positive removes)

The key to getting any performance increase out of this damn thing is go play with allocation; you are still using the same wattage of power, but you are taking away from the stronger cores to provide more power to the lesser cores.

In my case, core 3 and 4 were the best cores. They got a positive 5 setting, while the weaker cores got a negative 10 setting.

Scalar is useless and counter intuitive; 10x adds more voltage and will cause it to hit limits and downclock further, I left mine at auto after testing all the settings

Used to boost to 4300-4400, now it will hold 4650 most of the time, and in some cases hit 4700 multi core. Cinebench seems to be hard on it at 4523 max. Vermintide 2, to get best bench score, I had to shut off virus scanner, icue, logitech sw, evga precision and best was 202. Now with all those background programs running, I can still hit 199 fps. All apps running on stock pbo, score was 184.

Stock PBO
CPU Z 620
Multicore 4600

PBO2
642
5000

3DMark
PBO
CPU score 7800
PBO2
CPU 8800
 
Curve optimizer seems to be inverted (negative adds voltage and positive removes)
Why do you think this is the case? I've seen similar comments on reddit and people said this is not true. Also, the description clearly states this is not the case. I'd be careful assuming this without some really good proof.
 
Why do you think this is the case? I've seen similar comments on reddit and people said this is not true. Also, the description clearly states this is not the case. I'd be careful assuming this without some really good proof.
From my experience (asus crosshair hero viii, 3102 beta bios), the negative offset seems to reduce voltage, which opens up available TDP for additional clockspeed. Setting -20 with PBO set to motherboard, my 5950 hits 4650 all core load (cinebench) or higher with other loads (ex >4700 in World Community Grid Open Pandemics). In this case, EDC appears to be the limiting factor (my temp is anywhere between 77 to 79c).

The bios is pretty annoying, in that you have to set many settings in 2 places - once in advanced -> amd overclocking and once in the extreme tweaker menu. Changing just one doesn't seem to always work.
 
A quick experiment on my bios showed what the reddit said about my particular board and bios version. I made only the change to curve optimizer and saw negative raising voltage up, while testing the positive was dropping it. It's not normal, I stumbled on a reddit thread for my particular board describing the issue with it and the bios version. I am not sure on other boards and am hoping in future versions this will be corrected. The point being that whatever the value; you are playing a balancing act with the power limits, take some power away from stronger cores and give weaker cores more power.

You will have to verify your bios version and run your own tests to see which is the case, but with nothing except curve optimizer changed and some cinebench runs will confirm how it behaves for you.

With the extra performance boost, my peak cinebench temp hits 83C @ 4523 mhz. Stock PBO had me hitting 81C @ 4400MHZ while dipping to 4325mhz during runs prior.
 
Last edited:
CO on MSI board works correctly with negative resulting in higher boost and lower voltage.
 
So, messing around with manual overclocking, I was able to get this so far:'
CCD0: 4750mhz
CCD1: 4625mhz
voltage 1.275v (Cinebench stable, didn't try prime yet)
CBR20 scores 9150 on multi-core with this setting.

There seems to be a bug with the dynamic OC switching, it likes to set my voltage to 1.1 regardless of what I was setting it.
 
The biggest gripe I had with any attempt at PBO overclocking in the past was my results were getting worse not better despite seeing higher mhz values in cpu-z, ryzen master and core temp. All would say, you are boosting as much as 5000mhz, but the reality was performance got worse. My cinebench 23 scores tanked to barely 10K multicore, r20 was around 1600 yet mhz was peaking 5000?!?

3DMark told the better story; it kept track of the avg mhz over the run and it was terrible. On some runs, it did NOT even look like it was doing any PBO, other times, I could muster 4200mhz. CPU scores varied from 6500-7200 depending on its mood; it was unpredictable.

With my current setup and specific BIOS version, I finally saw results that made sense. Get 200mhz more, see more performance, see avg boost clocks higher. I'm hoping asus fixes their CO for my board in future.

I could not get any manual voltages stable with good oc values; I could find a random cinebench run or would hit prime up and get errors or a crash. CO is the safe bet for getting voltages under control.
 
The biggest gripe I had with any attempt at PBO overclocking in the past was my results were getting worse not better despite seeing higher mhz values in cpu-z, ryzen master and core temp. All would say, you are boosting as much as 5000mhz, but the reality was performance got worse. My cinebench 23 scores tanked to barely 10K multicore, r20 was around 1600 yet mhz was peaking 5000?!?

3DMark told the better story; it kept track of the avg mhz over the run and it was terrible. On some runs, it did NOT even look like it was doing any PBO, other times, I could muster 4200mhz. CPU scores varied from 6500-7200 depending on its mood; it was unpredictable.

With my current setup and specific BIOS version, I finally saw results that made sense. Get 200mhz more, see more performance, see avg boost clocks higher. I'm hoping asus fixes their CO for my board in future.

I could not get any manual voltages stable with good oc values; I could find a random cinebench run or would hit prime up and get errors or a crash. CO is the safe bet for getting voltages under control.

What are you using to validate what your boost clocks are? PBO is a bit tricky, I'll be working a bit more on that.

Also, what is CO?
 
So far I am using a few things to validate the clock speeds; ryzen master, cpu-z, core temp, and even 3dmark. 3D Mark being a better indicator of the game load vs actual core speed. Cinebench tends to be harder on the cpu and drops it to 4523 mhz where games like cyberpunk and vermintide 2 can net higher average clock speeds.
 
Last edited:
So, messing around with manual overclocking, I was able to get this so far:'
CCD0: 4750mhz
CCD1: 4625mhz
voltage 1.275v (Cinebench stable, didn't try prime yet)
CBR20 scores 9150 on multi-core with this setting.

There seems to be a bug with the dynamic OC switching, it likes to set my voltage to 1.1 regardless of what I was setting it.
Those are amazing results of the OC. 4.75 @ 1.275 volts is very low.
 
Last edited:
Back
Top