overclocking

[sudden torrent]

I'd be more worried about the "graphics driver not approved" warning

They're probably related

[solar wigeon]

https://support.benchmarks.ul.com/support/solutions/articles/44001788174-it-says-my-graphics-driver-is-not-approved-what-does-that-mean-

[sudden torrent]

Ok, and?

Are you saying you're using a driver that launched a few hours ago?

Because that info isn't anywhere in the image you sent

[proven canopy]

Lol, nobody gonna care

If you want a decent-ish OS just use https://www.dropbox.com/scl/fi/s4ekfedb3j2lfjr5nz9pl/m15a-21h2-3.0-1.0.mrimg?rlkey=fh4u3zemc4rbvt07v99x3udrw&st=5gytefwe&dl=0

If you want to eek out the last few single digit percents per bench/platform, test wegg 4/5/similar.

[steady lance]

is this a windows install?

[proven canopy]

It's a macrium image

If you want the base iso that the above .mrimg was cooked from, you can diy from here https://www.dropbox.com/scl/fi/e78onlz1jkjbv7a3oay0w/21h2forks1.0.iso?rlkey=7ovle3vkv53kdz0hilqjq0din&st=moll2i1i&dl=0

[steady lance]

would this be better than having run the script in #windows-ios?

I already used the script is why I am asking.

on a side note I am quite excited to try overclocking on this new motherboard since I am already seeing better results like with being able to max our TREFI wheras my previous board for whatever reason it couldn't.

[abstract wedge]

lmao ive had this iso for sooo many years

mega convenient for testing coz of how fast it installs

[sudden torrent]

It's mostly beneficial for benchmarking, not for gaming

[idle oar]

@steady lance any thoughts on my mem timings btw?

[steady lance]

Here are my timings, I have 2x32 Hynix sticks.

[steady lance]

what die do you have?

[steady lance]

I am assuming the only change you have made is increasing the Frequency?

[idle oar]

Should be M die. I havent touched the frequency. I changed the twr, trfc1, twtrl, trrd, twtrs, trrds, trcdwr, trp, tras, tfaw, trtp

[steady lance]

Your perfectly safe up to 1.45v. Fal will correct me if I am wrong but I believe 1.45v VDD/VDDQ is within spec and you can go higher but you will run into a board/cpu memory controller limitation before that is worth doing.

VDDIO is safe at 1.45v just disable the iGPU to avoid any potential risk.

My suggestion is, start at 1.45v tighten every thing as best as you can then crank the voltage down till you find out what your lowest stable voltage is.

[steady lance]

your ram should overclock better than mine since it puts less strain on the memory controller I believe.

[idle oar]

the VDDIO cannot be changed, and the max voltage is 1.43

[steady lance]

really?

[idle oar]

yes

[steady lance]

Did you enable High Voltage?

I have a MSI X670E Gaming and MSI X670 Pro Wifi board both allow for changing the VDDIO and have the option to enable high voltage for the ram.

@idle oar it should be called something like DRAM HIGH VOLTAGE MODE

If you can't find it then start at 1.43v use the same process

Personally I prefer starting by maxing out TREFI then moving onto tertiary timings

Ping me if you have questions.

[idle oar]

I spent two days testing and tuning all the timings i listed, there were very clear failure points for all of them. i havent decided if i want to go through the whole process again to raise the frequency. i also found that some of the timings heavily impacted the temps of the CPU during anta777 extreme

Even though it's only for a few seconds I really dont enjoy seeing the CPU spike so hard during anta

and there were a few timings where only moving them a few points would change the spikes by 7-8 degrees

i feel like if i was going to go hard with this kind of stuff I probably should've bought a top tier cooler

im still waiting for my gpu undervolt to prove itself too much, i know Control is a very demanding game but with everything set to ultra i was struggling to hit 60 fps. i dialed the RT settings back to high and it looks like i can sustain 70-90 fps. but i had a cutscene drop down to 4fps and havent figured out why. an initial google search said it might be my monitor refresh rate is too high, i havent had much chance to try it out yet to see if its improved

[steady lance]

perfectly normal.

Just to make sure

Before you start each test are you restarting the pc and making sure no apps start up?

Only do 1 kind of tuning at a time

otherwise you are just asking for something to fail.

[idle oar]

I change one timing in bios, reboot, open zen timings to confirm the timing is shown, then open hwinfo and then run anta777

The default EXPO settings were very loose, I reduced a few of the timings quite heavily, some of the others were less willing to change and some raise the CPU temps too high for my liking

[steady lance]

what cooler?

[idle oar]

There's a full VDDIO section in the BIOS, but everything I tried wouldn't allow any of the settings to be changed. The research I did indicates the motherboard is hardlocking the VDDIO stuff

Thermalright frozen infinity

[steady lance]

X3D chips will get hot no matter what you do.

[steady lance]

360mm?

[idle oar]

Yeah that's what I've read, but spiking to 96-97 just feels off to me, even if its for 2-3 seconds only. YEs 360mm

[steady lance]

Is it spiking or sustaining those temps?

[idle oar]

It spikes for 2-3 seconds then goes back to normal, theres a few spots in the 777extreme test that force it then its fine

[steady lance]

this is consistent?

[idle oar]

It happens every time I run the test, it happens in the first 15 seconds then goes back to normal temps

[steady lance]

also change tREFI to max, it makes for a easy yet nice boost to ram performance.

[idle oar]

only OCCT and 777extreme have pushed it anywhere near those temps

[steady lance]

Basically extends how long it takes before the ram refreshes itself

[idle oar]

I played a couple games last night on max settings and never saw GPU or CPU above 60, my understanding is the OCCT and anta tests push things to unrealistic limits that arent likely to be seen in real world scenarios

it's 'normal' to see spikes, but still feels yucky to me

[steady lance]

Emphasis on unrealistic.

[idle oar]

and it's all tied to the timings, my spikes when i first started tuning were in the low 80s, the tighter the timings got the higher the spikes got. which i understand is due to the strain on the memory controller

[steady lance]

also disable the iGPU as that can help regarding that.

No voltage will be used for the iGPU = less heat.

[idle oar]

yeah i did, i also disabled the memory protection thing in windows

[steady lance]

Personally I think if you have gone this far you should dive in and do the process right from the start.

Will yield better results and you will come out knowing how to do this step by step.

[idle oar]

i probably can, it's just so boring/exhausting doing a change then running anta for 20 minutes, and repeating that all day long

[steady lance]

I got a tip for you.

Start with the most unstable timing that you can get to post into windows.

I start by intentionally using a optimistic timing then making it more realistic till I eventually find the stable number.

It will save you hundreds of hours in the long run testing that way.

[idle oar]

I know that when I tried moving the frequency from 6400 to 6600 it instantly failed to post, but i didnt change anything else so it was likely a variety of factors

[steady lance]

6600 Synced would make your cpu a golden child.

[idle oar]

ill have to check the bios again when i get home at lunch and see if there's a high voltage option. i can probably do tuning while im working and let the tests run while i work. i have a saved profile that seems stable, so i can always revert back or slowly chip away at a new profile with more tuning

[steady lance]

I am able to post and do stuff at 6600 but it's not stable, granted if I mess with my settings enough I may be able to stablize it

[idle oar]

since it's baseline 6400, id like to think it can be tuned beyond that, but maybe not

[steady lance]

Most Zen4 cap out around 6000

to my knowledge Zen5 is better but not by much

[idle oar]

but it's a non-binary capacity, so that's working against it as well

[steady lance]

2x24GB overclocks better than 2x32GB if that is of any comfort to you.

[idle oar]

when i first did the build it looked like the ram was a major issue, but it turned out it was my own stupidity. i ordered a set of gkill 2x32 6000 but then found out the ram was fine, so im shipping the gskill back today

[steady lance]

You should be able to overclock better than me since you have a better CPU, Board, and RAM

[idle oar]

but virtually zero knowledge 😉

and i hope my aio can keep up, it feels like i probably shouldve gotten the better thermalright aio even though it doesnt match my fans. it was like $5 more and gamers nexus considers it the best value aio currently

i thought about just buying it and eating the $50 for the one i got

the x3d temps stress me out sometimes

[steady lance]

decent AIO.

[steady lance]

You best get used to it.

X3D chips run hot period.

Any headroom you give it will be used to boost higher resulting in the same high temps

[steady lance]

I can vouch for my LT720.

[idle oar]

Probably a good thing that PBO is basically unusable or else it would likely be even hotter 😄

[idle oar]

You probably didn't see it, but I got very lucky to not have fried the CPU. For the first few hours of use the AIO was only partially mounted, somehow the latch that holds the CPU in place didn't lock in and so the AIO was partially offset, but it's been mounted correctly now with fresh paste

[steady lance]

🤦‍♂️

[idle oar]

lol yeah, it was not a good day for me

between that and the benchmark issues i was having, i was in full meltdown mode, but thankfully it all seems to be working well now

But I did cyberpunk benchmarks on full ultra settings with ray tracing and managed 120fps average. Returnal ran on full max settings at 160+ fps. Control has been a bit less happy, but it's a notoriously difficult game to run well with high settings. I don't have many other games to truly push the graphics to the limit

[steady lance]

Minecraft and Ark Survival Ascended

[idle oar]

I would never download Ark, and I imagine the only way to push Minecraft is with mods and that's more work than I am interested in

A lot of places I've seen say to use Hogwarts Legacy as a test for graphics stuff. I bought that in the last Steam sale, so I will try that soon too.

[idle oar]

I took it to halfway to max and it caused major stuttering in games

[steady lance]

set tREFI first before other settings and see if it still stutters. This reminds me of when I had done my 2nd and 3rd timings before doing tREFI and I experienced some really weird things

Oh I forgot to mention.

tREFI is very limited by thermals and in my case required some extra voltage to stablize.

[idle oar]

Im honestly starting to get overwhelmed by all of this again, I cannot get any amount of ray tracing to work in Control, either with any of my tunings/undervolt or with everything at stock. But Cyberpunk is able to average 110fps with no drops below 96. Control even with medium RT settings crashes, has major stutters and drops to single digit fps for several seconds. I am reading it may be a driver issue but it seems odd how badly it runs.

i've only played a handful of games but they all seem to run ok except control, and control works fine until I turn on RT then it just has problems all over the place

[idle oar]

It's looking like it might be DX12 and how it interacts with Control + the 5X card

[wind willow]

Clock stretching or nah?

[steady lance]

watch for current clocks not average.

So long as current clocks are within around 25mhz give or take you aren't experiencing clock stretching

[steady lance]

Sounds like a problem with the game.

[wind willow]

Was watching current effective clocks

[steady lance]

Your fine then.

Well within margin of error.

Also, congratulations on the 5090 Astral #build-pics.

[idle oar]

Yeah, it's looking like a combination of it being one of the first games to really lean into RT and it not playing well with current hardware like my 5X card + the newest drivers and DX12 interactions.

[steady lance]

What gpu do you have?

[wind willow]

Ty

[steady lance]

As a rule of thumb if it's within 25mhz you don't need to worry about clock stretching. granted you might experience very subtle stretching on occasion but not enough to justify reducing your average performance in my opinion.

@wind willow

[wind willow]

Should I just go from -30 to -20 and see what happens? I feel like I get those same mhz variations on stock.

Part of me feels like overclocking and undervolting is pointless with this cpu because of how strong it is

[idle oar]

5070ti

I've got ethernet now, I am considering rolling back to Windows 10 to try older drivers to see if the game runs any better

[steady lance]

9800x3D overclocks well in comparison to 7000x3D.

[steady lance]

Worth a try.

Are you doing per core undervolt or?

[wind willow]

All core

Not gonna go through separatly

[steady lance]

Ahh alright.

[wind willow]

I wonder if its even worth overclocking the 9800x3d if the effective clocks don’t go anywhere near even the base clock during games.

[steady lance]

How are you going about overclocking?

The way I do it is, do PBO, raise PPT/TDC/EDC high enough so that isn't a restraint, apply the clock override and then do a per core undervolt (yields much better results).

[wind willow]

200+mhz, -30CO 1x scalar

[steady lance]

did you make sure your cores can hit +200mhz?

[wind willow]

They are. Easily

[wind willow]

Gonna make one more change… tmax of 85 instead of stock 95

[steady lance]

Keep it at 95, you will be limiting your boost clock

[wind willow]

You don’t think 95 can lower lifespan? I get that its stock and only really hits it for a few seconds during loading shaders

[steady lance]

No.

[wind willow]

Kk

[steady lance]

Perfectly within spec.

[wind willow]

I guess i am being way overly cautious

[steady lance]

Your cpu will die of other causes long before tjmax is a issue

Yep.

Think of it this way, if it's within stock spec no need for caution.

Caution is only warranted when things go beyond specifications for example +100c.

[steady lance]

@wind willow, @steel grove @idle oar any progress with your tunings?

[idle oar]

nah, ive been trying to just enjoy some games instead of smashing my head against the wall

[steady lance]

https://tenor.com/view/viktor-arcane-viktor-arcane-gif-24485287

[steel grove]

Yeah just left it at -20 undervolting all core and it was running perfectly well and stable last night

Didn't do anything else after that initial undervolt

[wind willow]

200+mhz, -25 CO, and 85c max temp. Works perfect

[steady lance]

Excellent.

[normal stratus]

confused on how this happened, my 11400h (with a slight undervolt) scored better than an m1 in multi-core? how is this possible

[acoustic crane]

No hyperthreading on the m1

[normal stratus]

fair enough

updated scores:

[balmy stream]

https://pcpartpicker.com/list/ytFPv4

who can help me optimize this pc to its max potential?

[proven canopy]

Buy some S8B used, $105 for a B tier 850 is mid

$210 for a 240 1080p VA in 2025 is eh

Could be worse

[acoustic crane]

Worse than eh

$210 gets you 1440p

[balmy stream]

i got it mad long ago for like 150 2 years ago now

[safe estuary]

spend $5 on the 14600k with an igpu

For $10 more you can buy the 14600k from newegg and get

• 1TB T-FORCE G50 Gen 4 NVMe (which is TLC!)
• Intel Bundle (Civ VII & Assassins Creed Shadows)

[safe estuary]

https://pcpartpicker.com/list/8T2dfd

Mobo SSD and CPU are newegg bundle

unless ur not asking for a new list...

[red terrace]

modded my 9070

[red terrace]

not bad for not having touched voltage yet

sorted for graphics score only since i'm on 7800x3d so my overall score will always be worse than like 14900ks and shiz

[red terrace]

okay that's more like it

BZ has some unsubmitted scores considerably above mine but i have heaps headroom left so we'll see how far it goes ig

[dull ginkgo]

Throw it on hwbot

[red terrace]

plan to once i finish dialing everything in

i think this is as far as i push it today until i can set up my multimeter for actually accurate voltage reading lol

kinda sketch to just keep upping the offset without knowing what my voltage value is lmao

[red terrace]

hot

[steady lance]

What mod is this?

[red terrace]

unlimited power limit via some shorted pins and voltage control via elmorlabs evc2se

[steady lance]

Gotcha. How does the GPU perform currently compared to stock outside of benchmarks?

[red terrace]

tbh i haven't really done a lot of gaming with crazy overclocks yet. my little brother wanted to play fortnite earlier so i had him stability test for me, and it was hitting 3.4ghz stable once i dialed in the voltages a bit (WAY less aggressive than my bench settings lol) vs around 3ghz before mods

i've been insanely busy with work and university lately, so it's sorta been benchmark or games, not enough time for both haha

[steady lance]

Fair fair.

I still need to fix my booting problems with my PC so I haven't been trying to do any new overclocks but I applied my previous OC on my old board to the new one for the ram.

Will s and zen timings once I am at the PC.

[acoustic crane]

iirc the 9070 with a 9070xt vbios was overclocked to 9070xt perf

Power limit mod should have similar results I'd imagine

[red terrace]

i'm performing a good ways above XT performance atm

[acoustic crane]

oh dang

[red terrace]

.

an overclocked 9070 xt would kill me, but i'm 7% above average/stock XT graphics scores or something like that

[dull ginkgo]

think my stock (power limited) 9070xt sits at around 30k on normal TS

every run where I mess withed core/vram was on TSE though

[red terrace]

that's what 3dmark pulls up yeah

i would really like to push 34k but idk if can. i need to be able to read voltage before i figure that out 😭

[red terrace]

regrettably, BZ actually has a score higher than mine i am benching more now

[abstract wedge]

tune ur os more

[red terrace]

yah, i kinda just gate screwing around with OS stuff

i managed to tie scores, but it wasn't validated bc systeminfo moment. i have tomorrow off work so ig i'll try again

[short blade]

i have acquired a 9800x3d

it is once again time for the ram overclocking rabbit hole

[short blade]

rip, seems to be incapable of 6400 1:1

[steady lance]

R.I.P

[short blade]

false alarm, it might be but i have to decide if it's worth dailying 1.3 vsoc for it

6200 works at 1.18

6400 passed an hour of occt large at 1.3 but errored out after 30 mins at 1.29

still have to hit it with more stability testing to decide if it's actually stable at 1.3

[steady lance]

Keep me posted. I am curious how the overclocking goes.

[short blade]

stopping for the night but current status: 6400 getting sudden bursts of errors after 2+ hours error-free

I can't go higher on vsoc, probably going to take some time trying out other voltages, RTTs, drive strengths etc. to see if 6400 can be stabilized with 1.3 vsoc

I have my current build to use while dialing in this tune so not in a rush at all, want to get it right

[restive cargo]

Nice

👏

[red terrace]

ty

[short blade]

hmm managed to get 6400 through 1 hour of VT3 at 1.29 vsoc

i think it's still too early to draw conclusions but seems like the bursts of errors might have had something to do with cldo vddp?

kinda confusing cause i see other people run that really low

but it hasn't happened since i raised that to the expo value

[steady lance]

nice, that's progress.

[short blade]

i guess it really was just cldo vddp

very interesting

1.27 vsoc did this as well as 1hr vt3 and 1hr occt large

1.25 vsoc errored out after 40 mins of occt

I figure 1.27 is low enough to daily anyway, so good to know that I can in fact daily 6400

should be good to start working on timings I figure

[steady lance]

what software is that for the ram test?

[short blade]

karhu

[abstract wedge]

https://drive.google.com/file/d/1kfl7Mfc1-t8r_N14fK_mQO2hTBB5ZnPl/view?usp=sharing

has my (licensed) karhu inside and a load of other useful oc stuff

i usually have a drive with all of this on it for convenience

no benchmate anymore coz they kept updating it and i was lazy

[short blade]

i see a lot of people are able to run SCLs @ 4 at 6400 1:1

my tWRWRSCL boots as low as 1 (seen some speculation that this timing doesn't actually do anything, didn't stability test) but tRDRDSCL doesn't POST at 4

is there anything i can mess with to attempt to tighten tRDRDSCL to 4 or is it just imc/board/ram/whatever limit?

i have heard that GDM off makes it harder to run SCLs @ 4, unsure if true

[steady lance]

Both SCL should be able to do 1 if you have a good ram kit and decent imc.

[short blade]

[steady lance]

Wait oh my bad.

[short blade]

i don't think i've ever seen anyone running SCLs <4 on AM5 but i could be wrong...

[steady lance]

I was thinking SC.

[short blade]

2 was the lowest i could do on AM4

ah

yeah i shotgunned all of that straight to 1 and it worked fine lol

[steady lance]

SCL below 4 is useless on 7000 series.

I do not know about 9000.

[short blade]

i'm stuck at 5

[steady lance]

Felix and Fal drilled that into me.

[short blade]

though upon looking more closely it appears that the vast majority of people running SCLs @ 4 have GDM on

[steady lance]

What IC do you have and voltages?

[short blade]

so i guess i can test if SCLs @ 4 works with GDM on, and see which performs better between that and SCLs @ 5 with GDM off

[short blade]

16 Gb SR hynix A, 1.27V VSOC, 1.15V VDDP, 0.92V VDDGs, 1.4V VDD=VDDQ=VDDIO, 1.8V VPP, 1.1V VDD MISC

[steady lance]

I have Hynix likely A die. SCL for me scales with VDD/VDDQ.

You can up VDDIO but above 1.4 I recommend disabling iGPU just incase.

[short blade]

I was planning to use the igpu to drive wallpaper engine on secondary screens

well, SCLs happily booted at 4 with GDM on

[steady lance]

Ahh okay. While not confirmed, but according to some research Fal helped me dig up a while ago VDDIO affects the iGPU so turning that to high may cause it to degrade much faster.

1.4 is safe territory as far as I am aware. But +1.45 iGPU should be disabled.

[short blade]

1.4 is what expo set it to, so i would hope that that's not high enough to cook the igpu

[steady lance]

You have SC set to 1 already?

Yep, looks like it.

Off topic but it looks like AMD is going to warranty my 7950x3D.

So that's neat.

My 7950x should be here tomorrow so I can use and mess around with over locking that bad boy.

[short blade]

just tested, honestly seems like both are within margin of error

[steady lance]

https://www.reddit.com/r/overclocking/comments/1hwau12/what_am5_settings_will_help_stabilize_gdm_off/

this reddit may have some helpful info in the comments regarding GDM behavior.

"There is also a massive trap in that many timings appear stable with GDM on because they're not actually used, effectively rounded up behind the scenes. Most notable on RRDS/RRDL/FAW and SCL's, but could be in more places. For example with uclk=memclk/2 and GDM, the RRDL setting will always behave as at least 8 so if you set it to 4 it can work because it's effectively still 8. Turning GDM off, it can actually behave like 4 yet require 6 or 8, so it becomes unstable when GDM is toggled when there is no issue with GDM itself."

[short blade]

well good thing GDM off is the first thing i did after freq

"That’s tricky as unless you have a godly board gdm 1T is impossible."

hmm...

interesting take

[steady lance]

@short blade do you recall if the 7950x also had issues with high soc voltage?

I am going to receive it tomorrow and I am thinking about having some fun tuning it this weekend.

[short blade]

i think it was all am5? but agesa hard caps to 1.3v now

i didn't know about the nitro settings until now

nitro 1/2/0 unstable at 1.27 vsoc 😦

seems ok at 1.3 so far

[steady lance]

Nitro settings?

What are those?

[short blade]

nitro rx data, nitro tx data, nitro control line

they're buried in the amd overclocking section

HUGE performance impact dropping from the default 2/3/1 to 1/2/0 but requires higher vsoc to stabilize

[short blade]

man, ram oc really is just so much easier on ryzen

it's coming together fast

got like 2 timings left to mess with

and it's only been 4 days?

meanwhile the b-die on raptor lake...

[short blade]

interesting... 2200 FCLK has a huge benefit to test speed in y-cruncher VT3 but a huge regression in karhu

I know that just means it's unstable but it's interesting nonetheless

[steady lance]

Awesome!

[steady lance]

Interesting. Is there a formula for each of those or just keep reducing till the last stable digit?

[short blade]

I honestly just shotgunned it straight to 1/2/0 and then raised vsoc until it worked

supposedly it doesn't go any lower than 1/2/0

[steady lance]

gotcha

How big of a performance difference did it make? @short blade

I will be receiving my 7950x tomorrow so I am going to have some fun tuning it.

[short blade]

it made triple the difference of every timing combined in terms of aida latency

negligible difference for bandwidth though

think it also made karhu and y-cruncher run around 1% faster

[short blade]

progress update

also passed 3 hours of VT3

pretty much just got tCL and tRCDRD left to mess with

tRCDRD won't go below 38 stable at this VDD but haven't really checked if it scales yet

[alpine mantle]

Also go to sleep lol

[steady lance]

awesome work!

[steady lance]

don't forget tRFC.

[short blade]

yeah that's done

tRFC2 and tRFCsb are unused timings so i set them to 420 69 for fun

[short blade]

okay i think this is done

might try to drop voltages a bit but timings should be all done

6400c26 seems doable but doesn't seem worth raising VDD that high just for tCL

I tested it for fun and it was able to bench at 1.6 but not stable

[steady lance]

6400c28 is still excellent.

Good job!

My 7950X arrived today so I am gonna have some fun with it this weekend.

[short blade]

[short blade]

alright well update

OOCT really does not like tRCDRD 37

kinda wild it can pass karhu and VT3 with no errors or performance regression but get 3200 errors in 3 mins of OCCT

but OCCT is happy again with tRCDRD 38

time to move on to CO...

[short blade]

so I just tossed on +200 fmax with -30 all cores overnight thinking it wouldn't be stable but I could weed out some unstable cores faster

...no errors after 10 hours of corecycler

[steady lance]

run VST3 in Y-Cruncher.

but so far that's good news.

[short blade]

Curve Optimizer: 44/44/44/39/29/49/49/49

Curve Shaper:
High Freq Low Temp: +15
Max Freq Low Temp: +15
High Freq High Temp: +10
Max Freq High Temp: +10

[steady lance]

Holy, those are some good offsets. Negative I hope xD.

"Curve Optimizer: 44/44/44/39/29/49/49/49"

[steady lance]

it looks like your performance dropped?

[short blade]

yeah, I wonder if it's clock stretching? effective clocks looked good though

gonna try bumping some cores upward and see if the performance improves

[short blade]

definitely negative 💀

[steady lance]

what are your PBO settings?

[steady lance]

try disabling the CO on one of the cores and re-test. On both my 7950x3D I noticed a behavior where when using Per Core Negative CO, if the entire CCD has a negative offset it causes the entire CCD to try boosting much higher and for longer but causes a lot of instability unless perfectly tuned in.

[short blade]

I bumped the 3 49s to 48s and it's at 9.42 test speed now

[steady lance]

I wonder if this might be a similar situation.

[short blade]

so I guess those 3 cores might've been clock stretching a bit

[steady lance]

you are testing with only the benchmark and HWinfo running I hope?

[short blade]

yeah

pretty fresh debloated OS

[steady lance]

if you haven't already, try running the VST3 test for Y-Cruncher.

[short blade]

VT3 is VST3

[steady lance]

I wonder why mine shows up as VST3?

[short blade]

might just be the name of your batch file or whatever

you'll see it if you look at command lines.txt in the y-cruncher folder

[steady lance]

Hmm alright.

I usually run VST3 for at least 200 iterations before calling it 100% stable.

followed by Cinebench 24 and 4 iterations of Core Cycler

[short blade]

I just capped it at 6 hours, happened to be 178 iterations doubt there's a significant difference

[steady lance]

gotcha.

[short blade]

test speed is pretty stable at 9.4-9.43 after bumping those cores up to 48

[steady lance]

good. that's within margin of error difference.

[short blade]

oh yeah I'm also at +200 fmax, just used motherboard limits cause as far as I'm aware the chip is too limited by other things to matter

[steady lance]

Neat.

Have you confirmed all cores can boost by +200?

[short blade]

yeah corecycler runs all day with each core reaching >5440mhz effective clock

+200 is 5450mhz on paper

[steady lance]

Excellent.

Seems you got yourself quite the golden boy.

[short blade]

chip seems to need quite a bit more voltage at higher temps (all core small ffts avx2 90C) so this curve shaper thing is really helping

[steady lance]

I am at the finishing stages of getting AMD to warranty my 7950x3D and so far while I haven't gotten a direct Yes, it appears they might give me a 9950x3D as a replacement.

[short blade]

can use curve shaper to maintain much higher light load boost without losing stability at max load

and no worries about idle stability which was always the killer with ryzen 5000

[short blade]

👀

that would be really cool I hope you get that

155W seems to be the most my cooler can handle, running at 92C

5040mhz effective clock @ 1.08V

in small ffts

offsets should effectively be 34/34/34/29/19/38/38/38 under all core load with curve shaper

[steady lance]

LT720 in Push/Pull with PTM7950 is the way young padowan.

Max temp of about 76c under high loads (+250W).

and games around 50-60.

[steady lance]

Thank you. I am really pressing them to give me a upgrade after AMD having sent me two faulty 7950x3D chips.

One with random crashes and this one which has booting problems.

[short blade]

meh, I'm just using a cheap $44 240mm

with its included paste

gets the job done

9800x3d has way higher heat density, so it's not like the cooler itself is saturated at 155W

the air coming off the rad is barely even warm after an hour at max load

either way, being able to sustain 5040mhz all core indefinitely in a load like all core small ffts is plenty

[steady lance]

fair fair

I am in no way wealthy enough, but as work ramps up I hope to eventually be able to create a Super Radiator tower that guarentees ambient temp fluid always goes to the CPU and GPU and there is a company that makes a Synthetic Diamond waterblock meant for transfering heat so I'd like to use that on the cpu and maybe gpu.

[red terrace]

atp just rig up a chiller

[steady lance]

By chiller you mean like an AC unit?

[red terrace]

yah

[steady lance]

That would cost a fortune in the long run.

[red terrace]

i didn't realize you meant for this to be a daily driver system lol

[steady lance]

Works for sure but very expensive.

[red terrace]

yeah fairs in that case

too bad peltiers suck a ton

[steady lance]

I tune for daily I don't really bother with benchmark rankings.

[red terrace]

understandable

i bench for scores then dial it back for daily

[steady lance]

Speaking of which I am going to plug in my 7950X and see how that baby runs.

Waiting on the warranty confirmation for my 7950x3D currently. I should get it come Monday.

[steady lance]

Just installed the 7950X, @short blade

Any bets on how good the IMC is?

Well, well, well.

The culprit was indeed the 7950x3D all along.

this install, 100% just confirmed it.

[steady lance]

I just realized you are dailying 1.55v VDD. Is that safe on Hynix A/M?

[steady lance]

6400 synced boots fine and has shown no errors.

[short blade]

I went back down to 1.4 cause I feel like just pushing tCL is not worth needing a ram fan for daily

but with cooling you can daily much higher than 1.55

also, interestingly karhu seems to be much harder on the CO stability than anything else I've tried

I've had to raise some offsets by as much as 10 from what was stable in corecycler, vt3, and small ffts

still getting errors almost instantly

🤨

[steady lance]

Gotcha, luckily for me my system has tons of airflow + my ambient is 10c roughly.

I like my room cold.

To my surprise, @short blade the ram tune I had for my 7950x3D applied just fine to this 7950X. I did increase the VDD voltage and reduce VDDQ and VDDIO as I plan to retune it specifically for this chip. Also, the CPU managed to boot 6600 synced and stable enough to browse/open apps but not game or bench stable.

[short blade]

having to drop offsets so far for karhu that it's probably worth not running 200 fmax anymore

wait you got a 7950x back instead of a 7950x3d?

oh wait nvm

[steady lance]

no no, I ordered this separately.

1. to confirm the CPU is the issue
2. I plan to reuse it as a server chip later

[short blade]

I'm really surprised at how hard karhu is to pass with CO

I guess there's no point in running anything else

pass 12 hours of VT3 just to error in 8 seconds in karhu lol

[steady lance]

reminds me of my AIDA64 SHA3 experience.

[short blade]

it's so difficult to pass that I'm turning off pbo to sanity check that my ram oc didn't somehow stop being stable

[steady lance]

Would pass every thing but that one damned test.

[short blade]

cause I was passing karhu perfectly fine with just the ram oc

[steady lance]

I applied a +200 boost override to this 7950x, gonna play some rivals in a moment so I will find out soon enough how it goes.

@short blade do you per chance know how to adjust resistances for ram? I feel like if done right, this 7950X could run 6600 synced.

[short blade]

lol

I wish I had a better response but afaik it's literally just trial and error

you gotta try a lot of permutations...

expect it to take a few days in all honesty

[steady lance]

What does permutations mean? I do not recall/remember.

Just keep lowering the resistances or do I need to increase them as well?

[short blade]

pretty much just trying whatever

[steady lance]

any risks?

[short blade]

not that I'm aware of

so uh...

I reverted back to pbo disabled and got errors instantly

turned it off, cleared cmos, reloaded the exact same settings and now it's running fine

I wonder if something got screwed up when I tried seeing if memory context restore would work

I saw lots of people mentioning that memory context restore was extremely unstable on am5

[short blade]

alright, well 200 fmax simply wants too much voltage in a load like karhu that's precisely heavy enough to push the cpu near limits but light enough that it can still hit max boost

small ffts was heavy enough that the cpu wasn't getting anywhere near max boost

gonna see if I can get 75 fmax (for 5.3 all core) stable instead

kinda sucks to give up the stable 5.45ghz 1T boost in lighter loads though, wonder if it's worth trying lower PPT to try to maintain that while throttling more aggressively in all-core loads?

unfortunately curve shaper alone didn't seem to be enough to resolve this, as the chip simply doesn't want to run 5.45 all core

[steady lance]

What does the 1T stand for?

Single thread?

[proven canopy]

[short blade]

single thread yeah

[short blade]

interesting, 200 fmax now seems to be working with still high offsets but less aggressive curve shaper

maybe I'm not understanding curve shaper

anyway, it's holding 5425mhz all core at 1.29v for now, I believe AMD explicitly said up to 1.31 vcore is fine for the 9800x3d

[short blade]

um

so I added an SSD and now my ram oc went from stable in multiple tests for multiple days straight to bsod within 10 seconds of karhu, even with CPU at stock

alright

that's pretty cool

I kinda just feel like giving up now

any suggestions? start over?

[acoustic crane]

Remove ssd first

Can't suggest much more pretty pho-cked rn sorry

[short blade]

welp I might have figured it out

too early to call it stable yet but at least it's not hitting bsod at 7% anymore

was thinking about what adding an SSD could have possibly affected

VDDG IOD 0.92 > 0.94 seems to have calmed it down a lot

more load on the IO die means more voltage needed I guess

gonna be adding a third ssd in 2 days when it arrives...

[dull ginkgo]

Can prob check if it has the same effects on a Mobo slot using CPU lanes directly versus a Mobo slot using Mobo lanes

[short blade]

welp back to 10 hours stable by bumping VDDG IOD

[short blade]

I'm gonna be filling all slots anyway

got used 2tb sn770 as part of a bundle so I want that to be my secondary drive, got sk hynix platinum p41 for my boot drive and upgrading my linux drive to 512gb team mp44l so I can finally fully ditch sata

been running linux off of an ancient 240gb sata ssd

this pc is way smaller than my current pc, could fit sata ssd in there still but don't really want to

[short blade]

something about curve shaper is not working the way I expect it to

not quite sure how it works

[steady lance]

@short blade I think I know what may have happened to this 7950X. I think it may have been a victim to a dead CCD. It came as refurbished from AMD sold by Newegg and I can see it visibly that someone had opened the IHS on it and retraced some things.

When running VST3, CR24 etc the All Core Boost between the two CCDs are very different by about 250mhz.

CCD1 performs at the expected 5.1Ghz range whereas CCD0 performs around 5.32-5.35Ghz range.

This is all speculation as to what the repair might have been. No idea if the CCD was actually swapped.

[short blade]

I kinda doubt swapping a CCD is actually a viable repair with how complex the packaging is

but I don't know for sure

[short blade]

well, got 5 hours stable in karhu, rebooted without changing any settings and now it's getting bsod from opening hwinfo 💀

"robust training mode" doesn't seem very robust

maybe tuning ryzen isn't worth the effort

boot to boot variance seems even worse than my old defective 13700K

memory context restore disabled, robust training mode enabled, 8x burst lengths

[steady lance]

@or

@short blade just a thought but could this be a result of you changing those 3 values smthing like 0/2/1?

It appears you have had problems ever since changing those 3.

[steady lance]

this

[short blade]

nah, problems only started once I moved on to PBO/CO

[steady lance]

Oh so the RAM is fine?

[short blade]

but now the ram oc is sometimes unstable even with PBO disabled

so idk

once I get this third SSD installed I'll have to decide whether or not I want to just start over or give up and just daily expo

boot to boot variance seems so ridiculous that I question whether or not it's worth it to tune this platform at all

[steady lance]

Higher LLC might prove helpful for stablizing due to the droops.

and here's a tip regarding AM5 CO boost behavior.

For some reason if all the cores on a CCD have a negative per core offset than the chips will try to boost much higher than it would otherwise.

So just leave 1/8 of the cores without the offset than it should avoid that boost behavior while retaining the benefits of the undervolt thermally speaking.

[short blade]

well my cores are boosting to exactly what I'd expect

there just seems to be ridiculous boot to boot variance on the memory training leaving me unable to test anything properly

[steady lance]

has been consistent with both my 7950x3D and my 7950X.

[short blade]

my cores are boosting to 5425mhz at 200 fmax, freq limit is 5450

don't see any abnormal boost behavior

and atm I'm getting instability even with PBO fully disabled

so yeah

[steady lance]

Hmm.

Motherboard limits for PBO?

[short blade]

fully disabled

not even pbo auto

just disabled

[steady lance]

You are tuning the CPU without pbo?

[short blade]

I disabled PBO while trying to figure out what's going on with the RAM oc

I just passed 5 hours of karhu with RAM oc, PBO active with 200 fmax, and aggressive CO offsets

[steady lance]

just a thought but it might be worth disabling the ram oc and reapplying the CPU oc you think is stable just to see if it's due to a combination of the two.

[short blade]

and now I can't even run 10 seconds of the same test with PBO disabled

that's why I'm saying the boot to boot variance is ridiculous

[steady lance]

This sounds like it's due to the motherboard in my non professional opinion.

Because your saying both the ram and CPU are working but apon rebooting it doesn't.

[short blade]

I think I had negligible boot to boot variance until I added the SSD

which is like

???

[steady lance]

Wait.

[short blade]

I'm not really willing to just not use my SSD slots

[steady lance]

Try pulling the SSD out. I recall FlexibleFelix having always complained that their NVMe ssd shorted and eventually killed their 7950x3D.

[short blade]

and the SSD itself is working fine

like reseating? or just removing

[steady lance]

Start by removing, see if it works then try a different slot, still works, then reset back into the original one.

[short blade]

I'm not willing to run without the SSD, if running 2 SSDs kills the system then I'm just going to return everything and go back to intel lol

[steady lance]

This sounds like either the SSD is shorting something or the slot itself has issues.

[short blade]

both SSDs are functioning as is, there's just annoying boot to boot variance ever since I added it

[steady lance]

Felix's were functional as well on the surface.

[short blade]

I've tried booting OS off of 2 out of the 3 slots, 1 slot is still empty

eventually the plan is to have 2 boot drives (2tb windows 512gb linux) and one secondary storage drive (2tb)

if I have to choose between overclocking and being able to use ssds then I will choose the ssds

[dull ginkgo]

It's probably just pushing io die too much, just don't expect a previous stable to be stable, need to find a new stable

[short blade]

the third SSD arrives tomorrow night

[short blade]

yeah, that's what I thought, pushing VDDG IOD made it stable again

but now there's boot to boot variance

and raising VDDG IOD further seemed to have negative results

[dull ginkgo]

My guess is something else is being partially affected

[steady lance]

Is VDDG IOD separate from VDDIO? Wanted to make sure I am following as I haven't seen that setting in my bios.

[dull ginkgo]

Best idea might be trying to loosen whatever wasn't stable with the lower VDDG IOD

[short blade]

also the second SSD is completely idle, I haven't even initialized it yet

[dull ginkgo]

I'd relate the variance to the VDDG IOD increase if that was just what made it semi stable

[short blade]

I'm probably just going to start over the ram oc once I get the third SSD in

0.94 VDDG IOD is giving me the best results

0.935, 0.945, 0.95 were less stable

[short blade]

yes

[dull ginkgo]

Yea it's probably just the hardware still, it's probably having to poll another pcie device despite not actually transferring anything

[short blade]

yeah but idk it's insane to me that adding an idle ssd is having this much impact on the system stability

just have to wait until tomorrow to start over from scratch

[dull ginkgo]

Prob just headroom that you found before was abusing not using something in the io die

[short blade]

zen 5 io die is complete junk I guess 💀

[dull ginkgo]

Either way mem OC generally tries to push it to its limits lol

[steady lance]

Take it from someone that has been at this on 3 different AM5 chips. Expect to always be confused.

[red terrace]

yeah they reused the zen 4 one. it is literally the exact silicon, iirc even the same stepping

[steady lance]

So far the IMC on this 7950X has proven to be better than the one on my 7950x3D so that's a pleasant surprise.

I will try a proper top-to-bottom overclock for this chip sometime this weekend.

[short blade]

third and final SSD is installed, chip seems to have completely lost the ability to run 6400 1:1 now

it'll boot but extremely unstable

dunno if there's some combination of voltages/resistances that'll get it to work again...

still scratching my head about how much impact adding SSDs has to RAM overclocking on this platform

I'm honestly ready to just give up and drop to 6200

...if that's still stable

[short blade]

update... 6400 1:1 is still doable, however:

before adding SSDs:

• 6400 1:1 GDM off stable at 1.295 vsoc, 0.94 vddg with nitro 1/2/0 8x/8x
• stable at 1.27 vsoc, 0.94 vddg with nitro disabled

after adding SSDs:

• 6400 1:1 GDM on unstable with any nitro enabled (even 2/3/1) at any vsoc, vddg, etc
• 6400 1:1 GDM on stable? at 1.3 vsoc, 1.05 vddg with nitro disabled

too early to say what the minimum stable vsoc/vddg is now, still somewhat early in the testing process

really hoping that once I figure out how to get 6400 1:1 stable again, I can at least shotgun my previously stable timings (and maybe turn GDM back off)

though I am also now wondering if it would be better to run 6000/6200 with nitro 1/2/0 and use the lower vsoc to try cranking fclk (previously stable at 2133, currently at 2000 for testing)

[steady lance]

HOLY SMOKES YES!!!!

AMD is giving me a 9950x3D to replace my 7950x3D!!!!!!

Took a lot of persuasion but they agreed and I just got the FedEx label and warranty approval.

[short blade]

👀

[acoustic crane]

Wow

[calm grail]

Wait what

Absolute w

[steady lance]

I am going to go to FedEx in about an hour to initiate the return

next upgrade is definitely gonna need to be a gpu.

[red terrace]

sweet

they gave me a free 3700x four years ago

[steady lance]

Oh nice. How did that happen?

[red terrace]

some jerk sold me a faulty 3700x on ebay. ebay issued a refund, but the seller didn't want to pay for return shipping, so i got to keep it. i explained that i had purchased my CPU second hand on my RMA slip, hence me not having the cooler/box, and they sent me a NIB 3700x anyway lol

[short blade]

seems like I won't be able to do 2133 FCLK anymore with 3 SSDs, do I just settle for 6400 with low FCLK or try lower freq to try and push FCLK?

I've heard that FCLK stability scales negatively with vsoc, not sure if true haven't investigated yet

6000 is already above the bandwidth bottleneck of FCLK, so I wonder if there's merit to chasing a low latency 6000 tune with FCLK as high as it'll go, or the performance benefit of higher UCLK outweighs the higher bandwidth

[short blade]

seems 2200 FCLK is stable without performance regression at 6200 with nitro 1/2/0

what would yall choose between

• 6200 1:1, 2200 FCLK, nitro 1/2/0, 1.2 vsoc, 1.0 vddg
• 6400 1:1, 2100 FCLK, nitro 2/3/1, 1.285 vsoc, 1.05 vddg

i'm honestly leaning towards 6200 with the lower voltages

[red terrace]

5% higher fclk or 3% better bandwidth. fclk probs makes more sense, especially on an x3d

make those core operations faster, cache is already so fast that memory doesn't scale super well

[short blade]

i'm actually not even sure if 2100 FCLK is stable with 6400, i only tested 2000 and 2133 so far

but yeah even if 2100 FCLK is stable i'm thinking 6200/2200 sounds better

[short blade]

6200 with 2200 FCLK is higher bandwidth

[red terrace]

in real terms yes

[short blade]

since FCLK is the bottleneck on bandwidth anyway

just wondering about the perf benefit of 3200 UCLK vs 3100

6200 with 2200 fclk is hitting much higher karhu test speed

[red terrace]

there isn't anything inherently faster about more IMC clock, it doesn't affect anything on the core. if you're getting more bandwidth with lower UCLK and higher FCLK, then do that

[short blade]

gonna test 8000 1:2 for fun

[red terrace]

enjoy

[steady lance]

Oh nice.

[steady lance]

What's the VST3 score and in-game perf difference?

[short blade]

i'm just looking at karhu for now

but it's ~212 vs 200

in favor of 6200 with 2200 FCLK

[short blade]

doing timings now and 6200 with 2200 FCLK is beating any test speed I ever managed to obtain at 6400

[red terrace]

FCLK on top

[short blade]

alright well it's a start

[steady lance]

@short blade do you have any tips for me?

I have been trying to tune this 7950X with major success and frustration. CCD0 can boost to a impeccable ALL-CORE of 5.55GHz with -30 Negative CO but as a result of CCD0 boosting so high it's causing cores 10 and 11 from CCD1 to throw an occasional error likely as a result of not being able to keep up.

Temps never exceed 87c.

[short blade]

shouldn't CCD0 and CCD1 be boosting separately?

[steady lance]

Boosts separately but in my experience with the 7950x3D(s) if one CCD boosts to high compared to the other it causes the other one to have stutters / very minor errors

I have never found a fix for this problem so I was curious if you knew of one.

[short blade]

no clue, never had a dual ccd chip before

[steady lance]

rip

I really want to get 5.6GHz all core stable.

oh by the way Mushr, guess what?

[short blade]

congrats

wow, VT3 test speed is WAY faster with 6200/2200 FCLK than it ever was with 6400/2133 FCLK

[short blade]

i want to say this is probably stable

[steady lance]

nah definitely not

||I am being sarcastic||

[short blade]

gonna try reducing vsoc now, and enabling power down mode to see if I can get memory context restore working without losing too much performance

[short blade]

alright, well it turns out that a lot of the issues I was having with bsod on boot were just my windows install starting to die

did a fresh install and I'm now able to use MCR without even enabling PDM

[red terrace]

many such cases lol

esp after tuning memory for so long. a lot of the times i reinstall windows mid-oc because i'm paranoid about winblows melting and throwing errors at me not caused by my settings... it's happened before and it sucks

[steady lance]

anyway to restore windows without doing a fresh install?

I don't have the luxury of overclocking on a separate system.

Speaking of paranoia, I decided to check the Replacement Status of my chip to confirm that AMD is going to send me the 9950x3D and it turns out they are. The part number for a 9950x3D is the listed replacement.

[steady lance]

Seems the command sfc /scannow works very well.

@short blade so I decided to start from scratch on my OC starting with CPU rather than ram for a chance of pace.

So far it looks like negative 40 CO works on Core 0.

passed 50 rounds of VST3 and I am running it through core cycler for a few rounds before I try gaming with it.

[short blade]

there are definitely cases where you can blow up windows beyond the point where sfc/dism can save it

I was getting bsod boot loops at jedec, but when I managed to get it stable enough sfc said no integrity violations

don't ram oc on an OS with anything important on it

[short blade]

progress

[short blade]

new bios may have granted me the ability to run 6400 again, with 2200 fclk this time

asrock definitely cooked something nice up with this bios update

can run memory context restore now perfectly stable

[steady lance]

Awesome.

Btw @short blade not concrete but I am like 99% sure that Core 0 is stable at -40 CO and I am debating if I want to disable CCD1 so that I can have 8 super good cores.

[steady lance]

gosh dammit I think I jinxed myself

[red terrace]

can't you set COs independently? especially across CCDs

[steady lance]

Yes. I am already doing that.

The problem is that if CCD0 or CCD1 boosts to fast compared to the other it can cause the other to become unstable on 7000 series AM5.

[red terrace]

strange

[steady lance]

good news, it seems I managed to get both CCD's stablized.

currently sustaining 5.5GHz all-core on CCD0 and 5.2GHz on CCD1, I am going to try and get both to boost higher.

[short blade]

watching this number slowly go up is fun

finally broke 240

[steady lance]

@short blade what test suite and settings do you recommend for CPU stability?

I have just been using Y-Cruncher VST3 and CoreCycler.

[short blade]

I'm literally just using karhu right now lol

karhu is harder to pass than VT3 and corecycler

this is with 20/20/20/20/19/30/30/30

don't think I'm gonna bother perfecting the curve, just nearest 5 should be good enough

check your clocks in VT3

my chip only boosts to around 5.1 all core in VT3, so I don't think it's super useful for checking fmax

karhu has the chip pegged at 5425mhz the entire time at nearly max load

so I've found it harder to pass than anything else

[heady aurora]

Hello

this is a POWERCOLOR Vega 64 Red Devil

i want to know whats the default for this ?

gpu started white artifacting

[steady lance]

gotcha. Mind sending me the official link for this test?

[short blade]

it's paid software

https://www.karhusoftware.com/ramtest/#introduction

I'm using an extension of it called KGuiX that additionally reports test speed

https://github.com/jjgraphix/KGuiX

[steady lance]

Gotcha, thank you. I think I will but it.

[short blade]

very useful tool well worth it if you like to tweak multiple machines imo

especially on AM5 it's been the primary tool I've used for both CPU and RAM oc

I will say the fact that version 2.0 has been pushed back this much is kinda weird

but the current version is useful enough to be worth it on its own

[short blade]

alright, finally done with curve optimizer

160W PPT, 180A TDC, 240A EDC
+200 Fmax
20/25/20/20/19/30/30/30

[steady lance]

@short blade, the replacement 9950x3D is finally on it's way to me!

Took forever, AMD kept saying they are waiting for their next shipment of the Ryzen 9 9950x3D to arrive, so I am guessing it showed up this morning.

[frank ferry]

is there any way i can undervolt my nitro 5?

throttlestop doesnt allow me to do nothing

[alpine mantle]

Specs?

[static oasis]

Intel xtu should do the trick

but you might need to get an older version because they locked some stuff for certain devices or something

as for the gpu you could use afterburner

[steady lance]

Does anyone know the safe limits for the 9950x3D Voltages?

[steady lance]

I think this 9950x3D can do 6600mhz synced with a little work.

It only gave 1 error, 40m into the Anta777 stress test.

@short blade thoughts on how I can stablize it?

[proven canopy]

[steady lance]

Thoughts?

[proven canopy]

2k fclk, 16/8 rrd l/s?

[steady lance]

I am currently using the XMP with increased DRAM/VDDQ/VDDIO to test if 6600 would somehow work.

Haven't touched timings yet.

It managed to last a very long time without erroring in the Anta777 Extreme profile for TM5 at 6600Mhz UCLK=MCLK. Do you have some recommendations on how I can stablize it? @proven canopy I'd really appreciate some advice.

@short blade thoughts?

Looking online at others who have 6600 1:1 stable, I noticed they have a higher CLD0 VDDP voltage. what is the safe limit for that?

[steady lance]

Found the CLD0 VDDP voltage. I've set it 1.1V.

[steady lance]

Stopping the test here for now so I can do stuff on the pc, but so far it looks like bumping the VDDP voltage from 1v to 1.1v made the difference I needed.

I will run a overnight test when I go to sleep.

[steady lance]

@short blade 6,600Ghz 1:1 is stable!!!

[steady lance]

Well this is annoying. It passed TM5 Anta777 Extreme profile but after a while failed the Y-Cruncher VST3 profile.

@short blade do you know what the maximum safe VDD and VDDQ voltage is for Hynix A/M die?

Also I have my VDDP voltage set to 1.15V is there any danger in leaving it that high for daily?

[short blade]

I've heard that 1.15v vddp is the safe limit, dunno for sure

my expo profile set vddp to 1.15v

not sure about VDDQ, but you can send VDD to the moon as long as you can keep the temps in check

some people daily 1.8+

however, not much actual benefit to sending VDD that high as only tCL scales

if you are feeling very determined to get 6600 1:1 working, you might want to play around with resistances

[steady lance]

What are considered safe temps?

[steady lance]

I am extremely determined. Been trying to do as much research as I can all day and could really use some help.

[steady lance]

I am currently testing 1.16V VDDP and I saw through reddit posts some boards default to 1.15V VDDP.

[short blade]

I believe with tREFI at 65535 you will start to see temp-related errors at around 53C on a-die

[steady lance]

I know that 6600 1:1 as it stands is game stable and idle stable.

[short blade]

again, it's not worth raising VDD for any sort of daily setup

only for benching

[steady lance]

I have VDD at 1.5v right now.

[short blade]

I've heard that more important timings such as tRCD do exhibit some VDD scaling up to ~1.4v so it's probably not worth trying to go lower than that

[steady lance]

VDD = 1.5V
VDDQ = 1.4V
VDDIO = 1.4V
VDDP = 1.16V

[short blade]

but above 1.4v it's pretty much just tCL

[steady lance]

I have 2x32GB Kingston Fury sticks with Hynix does.

Die. Unclear if they are A or M.

Also I just purchased the Kargu Speed test and got the plugin setup

[short blade]

we have different dies so I dunno, ymmv

[steady lance]

Karhu*

[short blade]

I believe you can tell pretty easily by trying to boot 130ns tRFC

a-die should and m-die shouldn't

from personal experience, my a-die is dailying 120ns while my m-die couldn't do lower than 150ns

also, you should test this but you might just be better off running 6400 nitro 1/2/0 than 6600 nitro default/disabled

though 6600 might win if you can run 2200 fclk cause supposedly there's a slight latency benefit from 2:3 "sync"

fclk appears to scale negatively with vsoc and positively with vddg ccd/iod

[short blade]

as far as I'm aware resistances are pretty much just trial and error

[steady lance]

Back

[steady lance]

Any danger to raising / lowering to much?

[short blade]

I don't know

don't think they would let you set terribly unsafe resistances but really couldn't tell you

my intel RPL + b-die preferred 80/0/48 when auto was 80/0/240

[steady lance]

Since raising VDDP from 1.15V to 1.16V so far it is not erroring in the harder tests.

[short blade]

but personally on this AM5 build I didn't need to change resistances from expo

[steady lance]

I am using a XMP profile as my base.

[short blade]

I have 6400 1:1 stable at 1.0 VDDP without performance degradation, seen some people stable at 0.9 but unsure if they tested performance degradation

YMMV

[steady lance]

What does YMMV mean?

[short blade]

your mileage may vary

[steady lance]

Gotcha.

[short blade]

may be possible that you need 1.16v to stabilize 6600

1.16v may or may not be safe

I got no idea

[steady lance]

I know that this RAM kit does about 400 tRFC if I recall correctly

That help indicate A or M?

[short blade]

in any case I suggest you test nitro settings

see if you can boot 6600 2/3/0, 1/2/1, 1/2/0, etc

[steady lance]

What are those?

Nitro Settings I mean.

[short blade]

if you can get 6600 stable-ish you can probably do 6400 1/2/0 easily

6400 1/2/0 may outperform 6600 default

you're gonna have to look for it in your bios but I think it's something related to the memory training

tightening nitro settings has a massive measurable performance difference

bigger than any timing in my case

[steady lance]

Interesting.

[short blade]

tighter nitro settings seems to require higher vsoc at the same frequency

[steady lance]

Also if my RAM can pass 5000% Karhu Coverage is that good to call stable?

[short blade]

so if you're already against the 1.3v wall at 6600 default you might not be able to do 6600 with nitro

[short blade]

no

I've gotten errors much deeper than that

also make sure cache and FPU stress are enabled

[steady lance]

Gimme a sec to check

[short blade]

iirc the furthest I've seen an error in karhu was around 17000%

[steady lance]

Cache was on default

FPU was off

[short blade]

normally I will say it's done with karhu after 20-30k%, but you still want to run multiple tests

cache and FPU enabled should speed up the testing

[steady lance]

So set cache to Enabled instead of default?

[short blade]

your test speed won't be comparable to mine, but you can compare it against yourself to see it go up as you tune

yes

[steady lance]

Whoa

Way faster

[short blade]

over the course of my ram tuning my karhu test speed went from ~180 to 242 so it's pretty cool to see

I think 9950x3d might be able to hit the 300s but not sure

[steady lance]

180 thousand?

[short blade]

180 MB/s

are you using KGuiX?

[steady lance]

Oh my bad. Just saw that it's a . Not a ,

[short blade]

oh yeah

[short blade]

going to bed but good luck hopefully I managed to type something helpful lol

ok yeah you're already way ahead of me after just starting your tune so you can probably see 300s on test speed

[steady lance]

Btw I noticed that the V-Cache CCD can hit 5.7GHz

[short blade]

the test speed tends to steadily increase over the first 15 mins or so and then pretty much stabilize there

[steady lance]

On the 9950x3D.

The 9800x3D iirc you said maxed out at 5.4 right?

[short blade]

I wonder how well that silicon is binned cause my 9800x3d hits very high temps/power just doing 5.4

yeah

[steady lance]

Mine is hitting 5.7 and maxed out at about 80c under heavy load

[short blade]

okay good night

[steady lance]

Good night

[steady lance]

@short blade any idea why ZenTimings reports ProcODT Pu at 160ohms when it's set to 40ohms in bios?

I am messing with the resistances to try and get 6600 1:1 stable.

[steady lance]

My overkill contraption to keep temps down.

I noticed that the errors consistently happened when the RAM got hot in the high 58-62c range so I dug up my last working spare fan to set this monster up.

[steady lance]

58-62c in less than 2m before the fan. So far I am 9m in and peaking at 46c.

Going to take a nap and check back on the test.

[short blade]

i think if you leave tREFI at auto you wouldn't have issues even at 62C

but not sure

[steady lance]

It's been auto

Well this is annoying. It still failed but at least it lasted an hour vs two minutes.

Temps peaked at 46.5c

[steady lance]

Changes:

• removed impedance settings since those weren't changing in ZenTimings

• Bumped VDDIO to 1.45V since that's the recommended safe limit from what I can gather

• Disabled the Integrated Graphics to try and squeeze just a little more headroom.

I will raise VDDP from 1.15V to 1.16V if it's still unstable than try impedance settings again

[steady lance]

48-50c appears to be where 6600 1:1 errors

From other reddit posts it seems Hynix A has a higher temperature tolerance around 60c.

[restive cargo]

Hot spot

[steady lance]

@restive cargo @proven canopy do you two have any knowledge regarding safe VDDP Voltages for Zen5 (9950x3D)? I am having a hard time finding any info regarding it. My motherboard is a MSI X670E Gaming WiFi with latest bios.

Any info regarding safe Voltages is much appreciated.

Also looking for info regarding Impedance values.

[restive cargo]

Not exact numbers for Zen 5, however a good safe starting point is just find out what stock voltages are for the chip that's a good baseline.
The biggest thing is (voltage = temp\time = / safe/not safe)

Also you can check here for other info.
https://www.overclock.net/threads/9800x3d-max-daily-vsoc-voltage.1814148/

[steady lance]

Thank you, I will take a look in a moment.

Has anyone managed to get 6600 1:1 stable with 2x32GB sticks before?

[steady lance]

To check my sanity I ran Karhu at my ram's XMP profile with no PBO just to make sure nothing went wrong or got corrupted.

[steady lance]

Trying 6600 1:1 again but 100% manual starting with auto timings to try and figure out if maybe they were to tight before

Intentionally set tCL to 34 so the rest would be abnormally high

Also apparently I can boot just fine with 1.4v for 6600.

[steady lance]

.

The RAM temps with the glass side panels open are bottoming out / staying at 39.8-40.8 Celsius with an ambient temperature of 17c (Alaska)

Test = Karhu FPU + Cache

Temps under load 30m:

• Ambient air = 17-18c
• Dimm A2 = 40.5-40.8c
• Dimm B2 = 39.8-40.3c
• VRMs = 43.5c

Voltages:
CPU NB/SOC = 1.3v
VDDP = 1.15
VDD/VDDQ/VDDIO = 1.4v

I am testing with the glass panel open to take thermals out of the equation for solving how to stabilize 6600.

[steady lance]

Test failed after 42m.

RAM temps maxed out at 45c.

[steady lance]

Posting this to keep log

[steady lance]

Running another test before I pause it to play some rivals.

Changes:

• VDD/VDDQ 1.4v to 1.5v
• VDDIO 1.4v to 1.45v
• Bank Swap Mode AUTO to APU
• Max RAM temp 41c to 42c (15m saturates RAM temp)

[steady lance]

Temps peaked at 45-46c with the glass panel on.

[steady lance]

Can someone help me understand the different Impedance values like ProcODT Up vs Down?

I am having a hard time finding a source on Google and whenever I do someone in reddit says that the other person was spreading false info

.

Currently testing with ProcODT PU at 30ohms. 80ohms gave me a instant bsod.

[steady lance]

30ohms was able to keep the RAM stable for 7hrs of Karhu then it appears at some point the PC restarted.

[proven canopy]

Watch the skatterbencher vid

[proven canopy]

Newer asrock/asus boards like m16a, am5 apex, z890 ocf have a dimm fit feature that'll let you have the board test a sweep of various values and report back.

[steady lance]

What does the off mean?

[steady lance]

Does MSI have an equivalent?

[proven canopy]

lol

You seem pretty invested in this haha, I'd just get an x870 apex

[steady lance]

Not a financially possible option for me right now.

[proven canopy]

If you got time for this, you got time to make apex money haha

[steady lance]

Damn

Have I been at it for this long now? Lmao

[proven canopy]

TBH, you get similar features through software on the gene, and can probably find one cheap-ish used in the US

Not so much in EU or elsewhere

[steady lance]

I do it because I enjoy learning about my hardware, gaining a better understanding of things and seeing performance/numbers go up

Brb

Back

[steady lance]

I hear a lot about the Gene and Apex. How good are they?

[proven canopy]

Just look up top 9xxx scores on hwbot and see what board they're using

[steady lance]

@proven canopy so here's where I am at now.

ProcODT UP is actually set to 30ohms

Temps never exceeded 45c after 7 hours of Karhu FPU + Cache.

I am struggling to get it 100% stable. As it stands, I can game on it, browse, light apps, etc but it failed Karhu after 7 hours on the last run and restarted as far as I can tell.

[proven canopy]

Most of those timings look auto

[steady lance]

they are

[proven canopy]

Why 16/8 rrds?

[steady lance]

I wanted to minimize get 6600 to work first.

so I set the timings except tCL to auto.

[proven canopy]

Unlikely to work with plug and play, imo daily 6000-6400 and tighten

[steady lance]

Are there any voltages or other settings I can try?

Currently I have been changing Processor ODT Pull Up. Tried 120, 80, 60, 48, 40, and 30 ohms.

30 has had the most success with Karhu.

[proven canopy]

6600 not working is imc/cpu/uclk side, not the sticks

[steady lance]

80 ohms and above failed to post.

[proven canopy]

DR a-die?

A-die can bench 9k+ lol

[steady lance]

Dual Rank, Hynix for certain but not sure if they are A or M.

[proven canopy]

tf is that 53 rcd

[steady lance]

Horrible timings indeed.

[proven canopy]

106 ras etc, looks auto by mb

[steady lance]

All of that is set to auto. I have been trying to get frequency first.

Only things that are not auto are:

• Voltages
• Bank Swap Mode set to APU
• iGPU disabled
• Ram frequency.

[proven canopy]

https://www.youtube.com/watch?v=dlYxmRcdLVw

maybe there's a newer version, idk, but start there

[steady lance]

Found some 9000 series overclock videos from his channels.

[proven canopy]

Also browse this guy's vods https://www.twitch.tv/madness777

oh, I guess he's live

[steady lance]

Actually Hardcore Overclocking, He has a video where he did a 6508 1:1 and said he used BCLK.

I thought BCLK affected CPU, GPU, and SSDs, it affects the ram as well?

Oh, he's now explaining it.

@short blade I am having a hard time finding nitro settings. I am debating trying 6400 with the nitro settings instead of 6600 and default like you suggested.

Nitro, what exactly is it?

Watching these videso currently:

1. https://www.youtube.com/watch?v=wnDEbqpvZvY

2. https://www.youtube.com/watch?v=dlYxmRcdLVw

[proven canopy]

Watch the first and skip the second

[steady lance]

Just finished the first one.

@proven canopy does the speed of the RAM affect what ProcODT UP works?

[proven canopy]

no, rather, I don't think so, for your goal

Again, I think you should try for 6000 tight, learn from there, and optionally try for higher

Whatever you learn will benefit for futher tune

[steady lance]

What order do you recommend most?

I typically do Frequency, Timings (tertiaries to primaries), then fclk.

I know this kit of RAM can do 6400cl28 from my 7950x3D.

.

Does anyone know the difference or know of an article explain ProcODT Pull Up and ProcODT Pull Down for DDR5?

From what I can tell, it's the termination resistance and that the higher the ohms the more current is needed.

[steady lance]

Made some changes after the video.

VDD 1.5 to 1.6
VDDQ 1.5 to 1.45
VDDIO 1.45 to 1.4
Timings:

• XMP 32-38-38-38-80 Enabled
• SC set to 1
• tREFI 65535
• tRFC 500 (I was tighten later)
• Bank Swap Mode to AUTO

Processor ODT Pull Up 30ohms to 32

Dropping the Frequency from 6600 1:1 to 6400 1:1.

6400 1:1 attempt #1:

VDDQ 1.45 to 1.5v
SOC to AUTO
tCL to 30
ProcODTup 32 to 30

Rest are same from previous runs

going to take a break and play with some friends

[steady lance]

Is having GDM off vs on good or bad? GDM off causes settings with GDM on to become unstable.

[dull ginkgo]

Gear down mode should still be rounding timings, but it's weird if 1t CR is stable and then GDM isn't

If your CR is set to 1t, GDM is bad, if your CR is 2t, GDM can be good

[wind willow]

Dang. Cutting is close with that 1.30 vsoc.

[steady lance]

1.3v is safe. PBO actually allows the CPU to hit 1.35V hard cap in the V/F Curve if viewed in Curve Shaper.

For 9000 series.

[wind willow]

1.3vsoc is the capped limit by amd.

Vdd can go higher.

[steady lance]

I plan to reduce the soc to make room for fclk in a moment.

[wind willow]

Is your cpu still the 7950?

[steady lance]

Left it on auto (1.3) so I could minimize failure points.

[steady lance]

9950x3D

Gotta update my username

[wind willow]

Or did you move to the 9950x3d? If so you should be able to do 2200 fclk with 1.205vsoc

Easily

[steady lance]

I am annoyed that I didn't get 6600 1:1 fully stabilized. I think it's due to having dual rank dimms.

[wind willow]

So i am very familiar with overclocking the amd cpus and nvidia gpus

Idk anything about ram

[steady lance]

What does the CR stand for? Are you talking about the RAM being synced 1:1?

[steady lance]

Dual rank is hell

[wind willow]

I have it

64 gigs of 6000 cl30

Idk how to OC it

All the “use buildzoid timings” dont help either

[steady lance]

Debating if I should focus on dropping SOC down then do FCLK, Nitro, or tightening timings next.

[steady lance]

What dies?

Hynix?

[wind willow]

Truth be told idk what die. I think they are hynix

They are g skill z neo or whatever

[steady lance]

have you checked the numbers yet?

[wind willow]

https://a.co/d/jfKfrt0

[steady lance]

On the sticks themselves.

[wind willow]

No

[steady lance]

G.Skill has a numbering system to find exactly what die you have.

[wind willow]

I do not know how to OC ram. I haven’t looked into it

I see

[steady lance]

Brand and die like A or M.

[wind willow]

Well the sticks are in my computer

[steady lance]

If you have Hynix A or M I can help whenever I have some time.

[wind willow]

I can look tonight. I am honestly fed up with OC’ing my 9950x3d. I truthfully do not see a difference in performance when it comes to games

[steady lance]

Anything else then my knowledge isn't so useful.

[wind willow]

And I have good OC/OS curves

[steady lance]

Per core and good temps?

Also I want to help because finding people with dual rank overclocking on 9950x3D is rare

So I want to get a larger sample size as to what to expect

Testing 1.2v SOC

[wind willow]

Nah not per cofe

Cofe

Core***

just a small co on ccs1 and no co on ccd2 and then CS

TJ MAX of 85

Disabled limits, no mobo limits

Meaning default limits

My idle temp is 37.5 with a kryosheet and a 360 arctic aio

But when doing stress tests it hits tj max fast…. Even if its set to 95

Gaming temps are 55-65

[steady lance]

Do per core, if your serious about seeing gains in my experience across a 7950x and two 7950x3D it yields much better results.

[wind willow]

Idk how to configure core cycler lel

And i want core cycler to just give me the answers

[steady lance]

Fairly simple to configure CoreCycler. I can show you later.

[wind willow]

Kk

[steady lance]

But generally speaking.

[wind willow]

How long would it take

To run

[dull ginkgo]

tCR, command rate iirc

[steady lance]

RAM, CPU, then GPU is my preferred way of overclocking.

[steady lance]

Is that Intel only?

I haven't seen it in my settings yet

[dull ginkgo]

pretty sure its this on zentimings