Ryzen 5900X overclocking help

[greyboxer]

My suggestion tune memory. Ignore the rest.

After my experience on 3700x this is absolutely how I’ll be doing with my 5900x.
Get the highest 1:1 fclk to dram setting that is stable, tighten timings, and move on. These babies are fully optimized out of the box anymore.

 

[soulesschild]

For anyone with the dark hero where is pbo2 vs pbo in the bios? I've been using the pbo tab in amd overclocking section and not the one in extreme tweaker. Is that correct? Pbo2 has the curve optimizer right?

Also, I'm curious why people are putting a higher negative value on their bad cores vs the good cores. Wouldn't you want more voltage to the bad cores to keep them stable (ie a smaller negative value)? So instead of negative 10 on good core and negative 20 on bad, vice versa.

I'm currently running an all core -10 with CO and would like to dial it in more if possible.

It's under advanced. Don't use the ASUS menu one as that's PBO1.

 

[Endgame]

The updated bios (asus hero WiFi 3204) seems to have added another menu for memory settings and voltage. Some of the settings in the main extreme tweaker section don’t seem to work anymore.

after playing with it a bit, it seems Agesa 1.2.0 is more stable that 1.1.X - I can run at IF 2000 for about 9 minutes of cinebench before it crashes (it used to crash just launching). Still trying to find out where its actually stable though

Ended up going all the way back to IF at 1866 to find stability... exactly where I was before 1.2. I could run at 1966 for an hour or two before crashing, and 1933 for 4ish hours, but always end up coming back to the host rebooted. The frustrating point is that it’s so close to stable that it seems like just 1 setting away from stability... but I have no idea what that setting is. I’ll bet all I need is one voltage tweak.

Overall I would say 1.2.0 has promise, but doesn’t quite seem to be there yet.

 

[KickAssCop]

1.2 messes up the previously stable memory overclocks for many people on both MSI and ASUS boards. I am staying away. My PC is running fine for now. Don't need any more trouble. I also have not really seen real benefit (as in real world applications) for trying to overclock IF/mem by another 34-134 MHz. I think 1867 seems like a nice balance.

On the positive side the good news is that my CO clocks of -10 for one bad core and -25 for all others has been rock solid. Also, right now I am at PBO +200 MHz but there might be opportunity to boost that to 250 or 300 MHz. Will try it this weekend.

 

[Light1984]

On the positive side the good news is that my CO clocks of -10 for one bad core and -25 for all others has been rock solid. Also, right now I am at PBO +200 MHz but there might be opportunity to boost that to 250 or 300 MHz. Will try it this weekend.

Is that a function of your MB? Mine only allows up to 200.

 

[KickAssCop]

Is that a function of your MB? Mine only allows up to 200.

If you use the OC menu/feature on MSI board you can go up to +500 MHz on the PBO boost. I have it set at 200 for now.

 

[Endgame]

If you use the OC menu/feature on MSI board you can go up to +500 MHz on the PBO boost. I have it set at 200 for now.

I’ve found more consistent clocks running +75 instead of +200 on my asus board. Perhaps MSI is better in this regard?

 

[KickAssCop]

How are you checking clock consistency?

 

[Endgame]

How are you checking clock consistency?

Nothing scientific - just watching Ryzen master during various loads.

All core load tested with 32 threads of world community grid, +75 clock was 4.65 pretty consistently. At +200 I would see 4.55 consistent with very brief jumps up to 4.7 (1 sec at 4.7 every 1-2 minutes).

At +200 on single core load, which I tested much less (one run of single core cinebench) I would see very brief (1 sec) reports of 5100mhz on my 5950, but it would spend most of the time at 4925. When I dropped to +75, it would spend almost all of its time at 4975. This was with the older Asus bios though - I haven't been watching clocks with AGESA 1.2

 

[KickAssCop]

Hmm. Interesting. I see much higher clocks w/ 200. I think the scores should be guiding principle on higher or lower. I have seen higher scores with 200 than lower clocks.

Will do some CBR20 runs to see if I also see a similar performance delta.

 

[soulesschild]

This is a great video,


I did what he did in this video and went to optimize my curve first. I seem to be stable with -25 on all cores and -15 on my 4 best cores and 50Mhz boost. It's actually on average boosting "better", went from seeing 5050 as my max clock on CCD0 to 5100 on CCD0. CCD1 seems to have only a 25Mhz improvement 4875->4900 where when I had +200 and -10 on best cores I was only seeing 4850-4875 on CCD1.

 

[Just_A_Person]

Try ProcODT around 34 ohms, and CADBUS 24-20-24-24. You can also set RTTnom to RZQ/7 and see if that helps. MAke sure you crank your DIMM voltage up and set everything manually. One wrong timing can screw up your results. If I go with all AUTO for LLC and core voltage, I get best results. Sometimes I go with vcore offset +0.0250v and LLC5 but not often. I have a 5900x , so it's got 2 CCDs. I run VSOC 1.15, VDDP 0.900, VDDG CCD 1.10 and VDDG IOD 1.10 . The second CCD can handle PBO Curve -30 on all, but on the first CCD it throws errors unless I go with -14 on my preferred core, -16 on my secondary preferred on ccd1, and -24 on the remaining CCD1 cores. It's better to lean slightly less negative on the preferred cores, but CCD2 can run full throttle. See my pic. I have to disable AV and everything else to get a solid reading on my scores. IF your TRFC on your RAM is close but not big enough, you'll see crappy l3 cache scores in Aida64. Locking in an EDC at about 192 and TDC at 133 may produce best results in gaming, but it does impact benchmarks.
Cinebench R23 = 22788, CPU-Z = 679.6 SC & 9912.4 MC, Notice the Aida64 scores which I'm particularly proud to show. And I've tried everything so get an FCLK higher than 1866 but it just throws WHEA bus/interconnect errors. And I mean, I tried EVERYTHING. sigh. Something with the PCI-E RAM tap.

 

[thecold]

I'm still not seeing anything where I want to overclock anything other than memory on the ryzen 5900x. Nothing is wrong with chasing that 10 percent, but I'm chasing more on the stable side nowadays.

From the looks of it, the only real loss is all core, when I run out of stock amperage limits.

4250 all core cinebench
4.35-4.4 all core cpu-z (9600 64.5c 10 minutes of stress)

 

[Supercharged_Z06]

All I did was set memory in BIOS to default/stock settings for my RAM (DDR4 3600/14-15-15-35/1.45V) and set Inifnity Fabric to 1800. Left everything else in BIOS set to auto. Then in Ryzen Master, turned on Auto OC and set Boost Override CPU to 175.

While I could probably tweak lots of things further to include tuning/OC'ing the memory, I'm sort of thinking why even bother... Everything is rock stable and the performance levels are through the roof compared to what I was running last year. (4790K).

 

[NeonFlak]

My "issue" with the 5900x is seeing it hit 90C in things like Realbench 8hr stress test. And I'm on a custom loop on CPU and GPU (6900xt). Though AMD is saying (reddit.com/r/AMD) we have to rethink max temps as the 5900x is perfectly fine to run 90C 24/7 supposedly. With my current PBO settings it does 4606 all core avg and 4985 single core avg with ram running at 3800CL16 and IF 1900. As heat soak starts to occur (1x TX360 and 1x XR5 radiator) during the 8hr Realbench stress test the all core avg starts to drop until it hits around 4475 all core avg.

 

[schizo]

Heh, I do 4.5Ghz all-core at 87C-- with a NF-D15. Power consumption really spikes when you try to get that last 100Mhz or so.

 

[thecold]

My "issue" with the 5900x is seeing it hit 90C in things like Realbench 8hr stress test. And I'm on a custom loop on CPU and GPU (6900xt). Though AMD is saying (reddit.com/r/AMD) we have to rethink max temps as the 5900x is perfectly fine to run 90C 24/7 supposedly. With my current PBO settings it does 4606 all core avg and 4985 single core avg with ram running at 3800CL16 and IF 1900. As heat soak starts to occur (1x TX360 and 1x XR5 radiator) during the 8hr Realbench stress test the all core avg starts to drop until it hits around 4475 all core avg.

What's your ambient and how much power does the cpu use during that time (under all core).

 

[KickAssCop]

Didn’t do any 8 hour tests but my max temps are 87 C w/ a 360 rad. My all core boost in most games is 4650-4750. Single core boost is 5150 but average about 4950.

I don’t understand the reco on ohm and procODT. I don’t even know where in the bios I have to set these things .

I don’t have CBR23 scores. Can run it later but my CBR20 scores are 8900 MC and 630 SC. This is with IF 3733 and DDR4 3733 CL16.

 

[KickAssCop]

Finally got around to running CBR23. This is with Nicehash running in the background and Epic games launcher running (which eats CPU cycles). I forgot to turn it off and have no motivation to run it again at least for now.

 

[KickAssCop]

Slightly better results since nothing was running in the background.