5900x oc help?

[funsoul]

Hi All!

A bit stuck getting my 5900x overclocked so figured I'd ask here as some of you have already figured this chip/platform out a bit.

Goals:
- Obtain maximum overclock for ambient benching so I can nail out a few boints, then,
- Obtain best overall performance for gaming

Basically, want to see what I can get out of the setup and have some hwbot fun then, after that, set the system up for maximum overall gaming performance.


Build:
cpu: AMD 5900x
mobo: Asus TUF GAMING X570-PLUS (WI-FI) (bios 4403, the most recent available)
ram: Team T-CREATE EXPERT 32GB (2 x 16GB) DDR4 3600 TTCED432G3600HC14CDC01 (B-die)
vga: EVGA GeForce RTX 3090 Ti FTW3 ULTRA GAMING, 24G-P5-4985-KR
drive: Seagate FireCuda 530 M.2 2280 1TB PCIe Gen4 x4 NVMe 1.4 ZP1000GM3A013
AIO: NZXT Kraken X73 360mm
psu: EVGA SuperNOVA P3 220-P3-1200-X1 1200W PLATINUM
case: Fractal Design Define 7 Gray Brushed Aluminum / Steel Mid Tower
fans: be quiet! SILENTWINGS 3 120mm
ram cooling: clip-on fan bracket from an old corsair kit
monitor: Samsung CRG9

Some of the links I found most useful for setup and testing:

- https://www.overclock.net/threads/m...guide-for-amd-5000-series-processors.1777081/
-

https://www.reddit.com/r/Amd/comments/fdl7cf

-

https://www.reddit.com/r/Amd/comments/lmtpk5

- https://skatterbencher.com/2021/08/18/skatterbencher-26-amd-ryzen-9-5900x-overclocked-to-5223-mhz/
-

- https://intensewebs.com/index.php/amd/amd-cpu/key-points-ryzen-oc
-

- https://www.overclockers.com/forums/threads/tweaking-my-new-5900x.801166

What's gone on and where am I atm:
- spent a couple weeks running CTR (and Project Hydra). OC was decent enough but results were sub-optimal (r20's in the 8300-8500 range). Heat was a real issue as it applied pretty hefty positive offsets everywhere and deltas between clocks and effective clocks was dramatic (so likely clock stretching).
- spent the next few weeks playing with Core Cycler and found my offset curves for both AVX and non-AVX loads. They work nicely for sure (r23 scores 23,000 - 23,150, r20 8850 ish, average effective core speed just over 4.5GHz).
- Saw an article where the user set 2 profiles (avx/non-avx) in a different tool (another brand's version of ryzen master) and the applied profile is auto-applied when loads are avx or non-avx. Any tips or links for how to make this work would be appreciated, ofc! I still have to google around for this).
- Found most efficient PPT, EDC, TDC values for max boost clocks, lowest deltas between clocks and effective clocks deltas, keeping max temp to 90C or less) for AVX/non-AVX loads.

PPT 178
EDC 132
TDC 117

Curve values (all offsets negative):
Core non-AVX AVX
0 15 25
1 15 25
2 5 7
3 16 25
4 23 30
5 5 15
6 25 30
7 20 30
8 25 30
9 30 30
10 30 30
11 27 30

Max boost override doesn't work at all for me...not even 25MHz. Also, my avx curve is weird. Did testing to find max effective clocks but the deltas between core and effective are enormous. Hoping to get both curves set properly so deltas are much, much smaller (@freeagent seems to have his running much more smoothly). What am I missing?

Any help, guidance, etc to help me find my 2 target overclocks would be hugely appreciated. Thank you!!!!

 

[freeagent]

What am I missing?

I am running my limits waay higher, I am also using AGESA 1203c because these newer ones for X3D broke PBO for the rest of us.

I am running mine at 235PPT, 160 TDC, 190 EDC, and then all core -25

 

[funsoul]

Thanks freeagent! Yeah...saw those in another thread and was stunned. How in the world do you keep temps sub-90 (or sub-100)?? I saw you have mad air cooling but on my side, the x73 aio can't seem to handle those EDC/TDC levels Already tried re-mounting the heatsink....that sucker was on there real good...no real change in temps after re-paste/re-mount.

I also see you've got very smooth running cores....your delta between core and effective is gorgeous! Blasting!

I'm wondering if I've got some settings out of whack that are spanking me? How did you get your curve so smooth (those low deltas)? Your boosts are higher, too. I started the process with boost override set to disabled, attempted to optimize curve/boosts then tried to apply boost override...would I be better off starting with an override of, say, 100 then working curve?

Feeling kinda stuck and a little frustrated but happy to start from default again and redo everything if it'll help me get better results. Atm....r20 is stuck in the low-8900's

Rando bench issue....wprime just refuses to start. Launching it results in an "undefined error". The wprime that comes with benchmate runs without any issues (but times stink). Anyone had this issue and got it fixed?

 

[freeagent]

I did the opposite of what everyone else does I think, because it didn't make sense to me, and GFlops were always low. I just found the max the CPU could pull from the socket, and set my curve around that. But.. there is one caveat. My setup will not pass core cycler. It will fall flat on its face. But! Luckily windows does not use the CPU the way that program does. Everything else is a pass, no problems, no whea.

 

[funsoul]

Core Cycler was a great learning experience but would be totally fine ditching it altogether.

hmmm....regarding Agesa...should I revert to an older bios?

How do my setting look? Anything glaring that I've missed or where I use a bad value?

Did you follow a guide or can you line out how I should approach the setup? Think one of my issues is too many guides lol Any tips, guidance or pointers would be hugely appreciated. Also happy to play with higher PPT, EDC, TDC.

One other bit of weirdness that would be great to overcome is frickin' Ryzen Master. I can't seem to set the correct settings so I can make changes in RM (for example, to ppt, edc, tdc voltage) and click apply and have them take without having to reboot?? Anything I've tried, RM always replies with "need to reboot due to auto-overclock settings). Would make ppt/edc/tdc voltage testing a heckuva lot less painful.

 

[freeagent]

My 5600X is corecycler stable, but when running it on my 5900X it might as well be left at stock.. and that is no fun

As for AGESA, I would.. but you don’t have to.. mine suffers on the newer ones. The drop from 1.5v max to 1.425v is good for those who can’t cool 5900X but terrible for top of the clocks.

I did not follow a guide, because when I tried performance always suffered as a result. I gave up and tried my way. Works awesome

Ryzen Master is a complete turd, I ditched it with a quickness even before I got my first Zen 3 CPU.

 

[funsoul]

Hmmm...looking here I can't find a bios with AGESA 1203c ....they go from AGESA V2 PI 1.2.0.3 Patch A (bios 4002) to AMD AM4 AGESA V2 PI 1.2.0.6b (4204) to AGESA V2 PI 1.2.0.7 (4403...installed). Definitely would like to go for the gusto lol....should I use bios 4002?

I see you went for all core -25....I'll definitely try all core but in prior tuning rounds (I've only tried per core so far), cores 2 and 5 couldn't handle big negative offsets. Earlier testing focused on getting max boost clocks (which is around edc 130-135, most cores boost to 4950 at that edc) but the deltas between core and effective are in the 200-500MHz range

Clearly I made a wrong turn somewhere so starting over but at least I know a bit more now.

Any tips or guidance would, of course, be appreciated. Again...hoping to get max oc for some hwbot subs before working on settings more appropriate to just gaming.
TIA!!

 

[Nebulous]

I could never get Ryzen Master to work properly and had nothing but issues using it. I ditched it and went a different route.

With the 1080Ti, my negative PBO was -25. Now with the 3080 I got it set to -20 on all cores. Few of my cores reach 5050 while the others top out at 4950 and 4900.

I'm still learning how to clock these Ryzen chips.

 

[funsoul]

Thanks Neb!
Yeah....I've essentially ditched it but was hoping to use it for more ppt/edc/tdc testing so I can avoid all the rebooting after every setting change lol

Did you start with enabling a boost override value (like 25, 50, 100) before curve tuning or after? I started with it disabled, wondering if that's why my max boosts are all 4950 and, after tuning, even a +25 boost override doesn't work.

 

[Nebulous]

Oh yeah I enabled boost override to positive and max of 200. Lemme go into my bios and chart down my settings for ya Funz.

*Edit*

Alrighty, here are my settings:

PBO= Advanced
PBO Limits = Disabled or Mobo
PBO Scaler = Manual> 10x
Curve Optimizer = All Cores
Optimizer Sign = Negative
Magnitude = 20
Boost Clock override = Enabled + Positive
Boost Clock Override = 200
Thermal Limits = 80

I manually set my vcore to override and 1.36. I tried using offset method, but could never get that to work right either. I also set ram setting to manual and entered timings and voltages. I dislike using XMP

 

[funsoul]

aha! Thanks Neb!! Wondering if freeagent started the same? That'd definitely explain (I think) why my max boost is 4950 versus 5000-5150.

Getting excited for another round of lots of tests. Hoping I get enough ideas/tips so I can finally get this thing humming along.

update: Did a little playing around....set 235PPT, 160 TDC, 190 EDC and tested boost overrides 25-200. Scores in r20 and gb3 show improvement up to +100-125, above that the scores go down a little. Temps are much better than expected (90.5C). Tried all core negative offsets of 5, 10, 15, 20 and 25. At -20, started getting some failures (1 core doesn't seem to like -10 or less, another core starts hating at -20) so set offsets to +5 from the AVX offsets I found in initial testing. Still haven't getting to where I got using my initial approach but seem to be moving the right direction.

Again...any thoughts or guidance would be a big help.

 

[freeagent]

Right now I have boost over ride off.. not sure why.. must have been smoking something. But I run with the taps wide open all the time. She likes it like that. Using near stock limits is no fun.. You can see all of my settings here.





I also changed my paste lol.. was using SYY-157 now I am using Thermalright TF7.. it came with my coolers so I thought I should try it out.. pretty good I would say..

Also.. it was not quiet

 

[Woomack]

You have lower temps in idle on most sensors than I have my room temps.

 

[funsoul]

Hiya Woomack!
Been doing some testing, will post an update when I get through the last of this round. Started with freeagent's power settings then tested EDC's from 190 down to 110 to find best then TDC testing from 160 to 110. Some interesting results! Optimal edc is in the 130-140 range (so far).

Ppt usage never seems to get above around 180 :/

 

[Woomack]

I have problems with memory, so I can't help you much, and I don't have AMD rig for tests right now to check anything. I'm waiting for the new series, hopefully in a month, but who knows if there won't be any delays.

 

[funsoul]

Did some EDC testing from 120-190. Still want to test 110 as there's something interesting there (you'll see when I finish/post TDC test results) as well as a couple re-tests (like those 2 edc140 results, if those laggy cores in the custom curve run can be fixed, that'd be a solid set) and more in the 120-155 range. Below's what I found so far (all test use 10x scalar).

Optimal EDC is somewhere in the 125-135ish range so far but there's something funny the 2 edc140 results with highlighting. Will post screenshots showing hwinfo for most of the tests tomorrow if anyone thinks they'll find anything I'm missing (other than the efficiency regarding ppt, tdc, edc where all should run in the 99.6-99.9. Very odd that ppt never gets above 80-83% regardless of all tested tdc/ppt combos?

 

[freeagent]

Its a little cooler in the house this morning.. I changed my TIM the other day, and enabled boost over ride




Not core cycler stable, but everything else stable..

 

[funsoul]

That's stunning freeagent....blasting!
Have you done much heavy multicore benches like r20, geekbench3 or x265? Would love to see something comparable to what I've been testing. Slowly (finally) starting to figure things out a bit better. Best scores so far are around 9010 in r20, 69,800 in gb3 and 30.2 in x265 4K. What I'm seeing so far is best, general power settings are in the EDC 120-140 range with TDC values a bit below EDC. Of the 3 benches I used for testing, each requires it's own edc/tdc values so will work on each benchmark individually.

I've not played with linpack....is that what you use to test all those cores in your screenshot? Would like to do some apples-to-apples compares to see roughly where I'm at with the cpu.

Trying to wrap up multicore testing soon and check single core performance. Best app for that? Does linpack do it? Been taking forever, itching to move on to ram lol. What's the best focus for a general, daily gaming setup? Don't think there are any games that'll use all cores but most use 2 or more at this point. For gaming, should I disable SMT (intel world, HT off) and then find ideal curve and power settings?

Did testing at EDC 110, may still do more in the 125-145 range but will probably focus on single benches from here forward. Also looking like I should revisit my pbo curve...seems there's something not quite right there (core 2 is funky for sure).

TDC test results added. Also adding a little x265 fine tuning tests.

 

[freeagent]

That's stunning freeagent....blasting!
Have you done much heavy multicore benches like r20, geekbench3 or x265? Would love to see something comparable to what I've been testing. Slowly (finally) starting to figure things out a bit better. Best scores so far are around 9010 in r20, 69,800 in gb3 and 30.2 in x265 4K. What I'm seeing so far is best, general power settings are in the EDC 120-140 range with TDC values a bit below EDC. Of the 3 benches I used for testing, each requires it's own edc/tdc values so will work on each benchmark individually.

I've not played with linpack....is that what you use to test all those cores in your screenshot? Would like to do some apples-to-apples compares to see roughly where I'm at with the cpu.

Trying to wrap up multicore testing soon and check single core performance. Best app for that? Does linpack do it? Been taking forever, itching to move on to ram lol. What's the best focus for a general, daily gaming setup? Don't think there are any games that'll use all cores but most use 2 or more at this point. For gaming, should I disable SMT (intel world, HT off) and then find ideal curve and power settings?

Did testing at EDC 110, may still do more in the 125-145 range but will probably focus on single benches from here forward. Also looking like I should revisit my pbo curve...seems there's something not quite right there (core 2 is funky for sure).

TDC test results added. Also adding a little x265 fine tuning tests.

I use Pi32M for the top clocks, and Linpack to check the bottom end, I like GFlops, and I find if I lower my limits, I lower my GFlops, as well as make things a tad more unstable at the top end. If I use newer AGESA 1206 or 1207 they lop off a bunch of vcore, so I cant hit the high clocks on all core, just one or two. Clocks all around suffer, but GFlop output remains the same, maybe a tad better, and the CPU runs a bit cooler. Kinda neat how they did that on the bottom end. Kinda wish this board had DOCS.

I can run R23 for you, with PBO and stuff I am usually between 234xx-236xx with Win 11. This is not core cycler stable, it will usually crash on test 4.. I think. Everything else I run is fine, no WHEA's, not reboots, crashes, nothing.. just core cycler And windows doesn't even use the CPU the way that program does.. I think its just a big waste of time personally.

For gaming, I just leave SMT enabled, no need to turn anything off.

 

[funsoul]

Yeah...thanks...I can't help but feel I'm missing something important Seems we get similar r23's. I stopped testing using that bench as there are no globals for it lol

Starting to look at the ram. Already identified a problem. Anyone have AIDA screenshots they can share so I've got decent comparisons? Ideally, 3600, 3666 and 3733 (no matter what I've tried so far, system refuses to post at 3800/1900).

I'll add my AIDA's here shortly.
update: AIDA's added. Something is wrong Iirc...these sticks should be doing much better at these speeds. Also something seems funny with my oc profile as some scores are actually worse than the default profile. Any ideas or comparison AIDA's?

adding my 3733 timings from the oc profile. Anyone spot anything?