Ryzen 5900X overclocking help

My effective clock shows up as 1900 MHz even when CPU is boosting to 4.7 Gs all core.

Also are 5900X owners considering Intel 11900K upgrade?
 
My effective clock shows up as 1900 MHz even when CPU is boosting to 4.7 Gs all core.

Also are 5900X owners considering Intel 11900K upgrade?
Something doesn't seem right. If I were you, I would try clearing the CMOS. Then loading optimized defaults. Then checking things out. And then slowly try your settings again. CPU overclock first. Then RAM overclock.

Personally, I am not considering 11900K. It sucks you are having so much trouble with your system.
 
That seems likely a bug in HWinfo64. Although looking at it now, my top effective clock is correct at 4.9Ghz. Anyway, trust Ryzen Master.
 
There is nothing wrong with my system. It is operating as advertised. Unless you count the extreme overclocking I am doing on it.
 
Last edited:
Bumped my RAM OC up to 3733 c14-13-13-13-27-40-T1 (GDM Enabled) at 1.5v. Scores went up. Latency is down to 55.5ns in AIDA64. Makes a big difference in how snappy everything feels. Windows seem to open before my mouse button comes back up. :D
 
I have a 5950x running my ram 64gb 2x32gb 3800mhz 1900fclk on a x570-i you need to use the ryzen calculator and properly set your voltages. you need to set the soc voltage , VDDG CCD , VDDG IOD , cLDO Voltage properly. for me it was soc 1.1 the others 1.05v. my board got flaky with anything less than 1.05v on the VDDG voltages. When you are running 1900mhz fclk you are overclocking infinity fabric.
 
I tried higher SOC voltage but didn't touch the others. Can you share what voltages did you try?
 
MaxT ?
your reddit post shows PBO enabled. What's your L3 cache bandwidth? PBO kills my L3 cache speed
 
MaxT ?
your reddit post shows PBO enabled. What's your L3 cache bandwidth? PBO kills my L3 cache speed
I have a note in there about L3 cache speeds. We are supposed to be seeing around 1200MB on 5900x, since they have 2 CCDs. However, even at stock everyone with PB sees between 600 to 700. Once you start using PBO, in my case specifically, the higher I went on TDC limit and the lower I went on EDC the worse my L3 cache scores got in AIDA64. If you read about this on overclock.net forums, most people say AIDA64 doesn't calculate this correctly and they say to ignore it, since it doesn't affect ANY of the other benchmark scores. If you enable PBO Fmax enhancer (maybe an Asus exclusive feature), you will get scores over 1100, but that kills the all core boost quite a bit, so everything else suffers. Also, they say people doing manual all core overclock are not affected by this.

So the bottom line is - ignore the AIDA64 L3 scores. They don't seem to be accurate (for PB/PBO) and nothing else is affected. You can also try SiSoft Sandra, as that measures L3 cache as well as other bandwidth and latency related stuff.
 
Just to complete this thread. After experimenting a shit metric ton, I decided to go back to default PBO without any voltage controls. Reason for that is that it gave me the most stable outcomes in terms of performance. I ended up w/ +200 MHz on PBO (5.15 Gs SC boost), default volts, default platform thermal limits etc., IF 1867/DDR4 3733, CL16-19-19-19-39-65. It scores highest in CBR20 (with exception of 4.7 Gs all core clock that scored 9100). I tweaked the memory also. Ended up w/ these results.

CPU-Z ~650/9700
Geekbench 1676/14969
CBR20 ~630/8630
AIDA ~56.7 Gs/62 ns

I also replaced a couple of Corsair ML120 fans (got free replacement from Corsair) and synced the RGB. Did some cable management and now the computer looks as below. Max temps I see are 81 C in CBR20 but normally 70-75 C in most games. In COD BOCW it does hit around 85 C (during menus not sure why :rolleyes: ). However, in game it is usually around the 78-80 C mark. As far as I understand this is normal operating behavior.

D8-B0-FE8-E-4224-4-BFF-A1-E4-448-A03-E32164.jpg
DD33501-F-0729-4-DE1-8-B9-A-2920675-D0-A43.jpg
53-F178-FE-DB6-C-4-F08-9-B07-6-E4-ED6-F48527.jpg

Time to enjoy this beast. Btw all stability tests passed and no WHEA errors or random reboots after 3-5 days of computer being idle. I wanted to try custom water cooling as well but it was about 800 bucks to cool CPU/GPU w/ two radiators. I really don't think it is worth the time, effort and maintenance. As you can see I top mounted my rad again allowing for all component temps to go back to normal especially my graphics card that runs at 62-63 C max load w/ overclock to 2 Gs/20.5 Gs at only 1800 RPM (inaudible).

Hope this helps someone :).
 
Very nice. I am purposefully not pushing it that hard. Also I have crappy G Skill ram so my benches will always be behind unless I push it a bit harder. I did try curve optimizer a bit with higher numbers but it led to WHEA errors.
 
Very nice. I am purposefully not pushing it that hard. Also I have crappy G Skill ram so my benches will always be behind unless I push it a bit harder. I did try curve optimizer a bit with higher numbers but it led to WHEA errors.

Makes sense. I am only pushing the RAM really. The CPU is actually undervolted and the 3080 is at stock volts but boosted curve. The RAM at 1.49v is the only scary part, but everyone says b-die can take it long-term. We'll see.
 
Last edited:
I'm sure it's technically fine, I just wouldn't be comfortable with it, much like pumping up my CPU to run at 90C all full load. Yes it's within spec, yes AMD says it's perfectly fine, but it gives me the heebie-jeebies and I'm willing to forego a small bit of performance for peace of mind. And a cooler room in the summer.
 
Agree that is a pretty crazy ram overclock. But given the temps it should be fine.
It's the norm for b-die. Check out the community spreadsheets for Zen2 and Zen3. 9 out of 10 people run that. :) It's a $200 kit in a $3k PC. If it dies, it dies. Probably won't happen for years.
 
Link to purchase the kit. I may order it.
It's the norm for b-die. Check out the community spreadsheets for Zen2 and Zen3. 9 out of 10 people run that. :) It's a $200 kit in a $3k PC. If it dies, it dies. Probably won't happen for years.
 
Last edited:
Is there a way to figure out which core is causing the WHEA error. I can run my computer at -30 CO but after 1-2 days it gives a WHEA error.

I even ran -15 CO for 3-4 days and then suddenly a WHEA error.

Right now I have it on -15 on best and better cores but -10 on all others. I think this setup lasted me days without any errors. Going to keep it here unless someone tells me how to locate the problematic core so I can just put that on -10 and everything else on -20 or something.
 
Is there a way to figure out which core is causing the WHEA error. I can run my computer at -30 CO but after 1-2 days it gives a WHEA error.

I even ran -15 CO for 3-4 days and then suddenly a WHEA error.

Right now I have it on -15 on best and better cores but -10 on all others. I think this setup lasted me days without any errors. Going to keep it here unless someone tells me how to locate the problematic core so I can just put that on -10 and everything else on -20 or something.
Yes, sort of.

In the event viewer, you will see the APIC ID number. That ties into a physical core of the CPU. The physical cores start at 0 and so do the APIC IDs. The formula is FLOOR(APICID/2). So you are looking at something like:

Core 0 - APIC IDs 0 and 1,
Core 1 - APIC IDs 2 and 3,
Core 2 - APIC IDs 4 and 5,
and so on....

However, there is catch here. I think some of those APIC IDs are reserved, so eventually some are skipped. So you might see something like APIC ID 26 or 27, which are more than what 5900x has if the APIC IDs were truly sequential. I am not sure how to properly identify those other than some trial and error.
For me, my preferred cores were Core 4 and Core 5, and true enough the APIC IDs I had in the event viewer were 8 and 10. Once I adjusted the CO for those two cores, I haven't had a single idle reboot.

This is as long as we are talking idle reboots, when the PC is idle or doing something light like windows updates or zip file extraction. I think these show up in the event viewer is cache hierarchy WHEA errors.
If you are getting "interconnect failure" WHEA errors, those are related to RAM/IMC and VDDP/SOC voltages or IF stability in general.
 
Hmm, that works. I am getting APC ID 0 as the error. APC ID 0 = Core 0 and that btw is my preferred core as well (not the best core).
I have given it -15 CO. Do you think I need to increase it or decrease it? Seems like an increase will fix it (as in putting it back to -10)?

This is the error 18 I get:

A fatal hardware error has occurred.

Reported by component: Processor Core
Error Source: Machine Check Exception
Error Type: Cache Hierarchy Error
Processor APIC ID: 0

The details view of this entry contains further information.
 
Hmm, that works. I am getting APC ID 0 as the error. APC ID 0 = Core 0 and that btw is my preferred core as well (not the best core).
I have given it -15 CO. Do you think I need to increase it or decrease it? Seems like an increase will fix it (as in putting it back to -10)?

This is the error 18 I get:

A fatal hardware error has occurred.

Reported by component: Processor Core
Error Source: Machine Check Exception
Error Type: Cache Hierarchy Error
Processor APIC ID: 0

The details view of this entry contains further information.
Correct. If you keep getting this error during idle, then you need to keep lowering the CO undervolt for that core. So if -15 wasn't stable, -10 is the next thing to try and so on. Some people can only do -5 on their preferred cores, some couldn't move off of 0, and some people claimed to have defective CPUs where they had to do add positive voltage to get it stable.

My two best cores need -20 and -25 to be fully stable (so pretty good) and that is with a +50 clock boost.
 
Great. I set it to -10 on first core and -20 on all others. So far so good but it takes like 2-3 days to get the WHEA error. Ran OCCT test for 30 mins without issues. All other benches and tests work fine as well. CBR20 actually jumped to 8856 (from previous 8818 or something). So not complaining.
 
  • Like
Reactions: MaxT
like this
Generally you want your best/second-best cores to have more voltage than the others, because they're the only ones that'll actually boost high on lightly-threaded workloads.
 
Been an interesting path trying to mess with PBO2. Part of me just wants to ignore curve optimizer as really it's just chasing bench numbers but it's fun to see how different settings affect certain scores.

On my 5950x with PBO2 on, scalar auto, motherboard limits

SettingsCBR20 Multi scoreCBR20 Single ScoreCPUZ MultiCPUZ Single
+200 boost, -10 on first and second best cores, -20 on rest1084762812804.1667.1
+200 boost, -10 all core1088961212640.7657
+200 boost, -5 on best, -10 on second best, -20 rest1093963112801671.1

Edit: This was on the latest Dark Hero bios, 3204, AGESA 1.2.0.0
 
Last edited:
Not just scores, it also reduces power usage and heat generation so is generally worth turning on, if not necessarily going out of your way to optimize.
 
Wow AMD, agesa 1.1 was trash.

I couldn't even stay stable above 1600 (4x16 sticks). With 1.2 it's now stable at the 1800 mhz which was always stable on my 3900x.

At least that's fixed...
 
Not just scores, it also reduces power usage and heat generation so is generally worth turning on, if not necessarily going out of your way to optimize.
True. Now if only I didn't have to go through 16 freaking cores to optimize it one by one :LOL: Also trying to suss out stability long term seems impossible/fool's errand if you're going per core optimization. I sort of just do a few CB runs then play some games to make sure I'm somewhat stable (nothing mission critical here).
 
Not just scores, it also reduces power usage and heat generation so is generally worth turning on, if not necessarily going out of your way to optimize.
It doesn't do any of that. it just allows all core boost to go higher before hitting limits. Only thing that limits power usage and temps are PPT and to some extent TDC.
 
Assuming you don't have PBO fully unlocked (set to motherboard) curve optimizer will absolutely lower temps, because you'll hit the EDC limit on multi-core workloads. For some reason I always hit EDC rather than TDC or PPT.

Next CTR version is supposed to work with CO, that should automate optimizing it per-core. That will be extremely useful.
 
Assuming you don't have PBO fully unlocked (set to motherboard) curve optimizer will absolutely lower temps, because you'll hit the EDC limit on multi-core workloads. For some reason I always hit EDC rather than TDC or PPT.

Next CTR version is supposed to work with CO, that should automate optimizing it per-core. That will be extremely useful.
EDC runs at 100% no matter what (from my observations). I'm yet to see someone who isn't hitting 100% EDC. It might lower the temps very slightly because at stock settings you will hit PPT or TDC limits quickly.
 
Whichever power limit you hit first, that's where your performance ends.
 
For me if I set to Motherboard I lose performance. On the positive side, I further tweaked some crap and here are my latest benches.
CBR20-4-Feb.jpg
Here is CPU-Z run 673/9952
https://valid.x86.fr/1bwxuf
Here is my Geekbench run 1723/15356
https://browser.geekbench.com/v5/cpu/6247011
AIDA 57.3 Gs / 59.5 ns. Not the best in the world but w/ CL16 it is OK I guess.
cachemem.png
Real bench 341,660
Thu Feb 4 2021
10:51:16
Image Editing: 230034
Time: 23.1618
Encoding: 383545
Time: 13.8914
OpenCL: 415740
KSamples/sec: 76450
Heavy Multitasking: 337324
Time: 22.6251
System Score: 341660
 
  • Like
Reactions: MaxT
like this
Back
Top