overclocking

[short blade]

getting near 50k will probably require exotic cooling

[red terrace]

are you scoring better with FT off and higher clocks?

[short blade]

I got higher benchmark scores with FT but it has an incomprehensibly high impact on stability

fast timings: best I could daily was -55 mV 2638MHz

[red terrace]

interesting, in my experience it's borderline impossible to crash the RDNA 4's with memory OC. they just lose a ton of performance

[short blade]

default timings: daily -90mV 2806MHz

[short blade]

yeah for some reason it's hugely affecting how much the core can UV though

[red terrace]

oh, that's wacky

[short blade]

I did see other people saying similar things so I guess it's not that surprising

still weird though

this was my vram testing last night

steel nomad scores

got average of multiple runs for the most interesting range

landed on 2806

had the balcony door open at night, fans at 100%, and waited some time between each test to minimize the impact of heatsink saturation

debloated OS

damn, core 1 is "stable" at -50 but the clock stretching is instantly obvious lol

the effective clock actually looks fine but the performance loss is massive

[short blade]

hmm... my entire approach to tuning CO on zen 5 may have been flawed

it seems that dialing cores in one at a time is not very useful

need to make changes in groups

core 1 at -19, -20, -24, -30, -40, -50 didn't change effective clock or voltage at all on any core

but trying -30 on cores 1, 2, 3, 4 all at the same time results in a huge boost in effective clock and lower voltage across all 8 cores

24/50/19/19/19/30/35/35:
5404mhz effective clock: 1.28v

24/30/30/30/30/30/35/35:
5415mhz effective clock, 1.21v

if this is actually stable the extra headroom might tempt me to play with bclk...

you know what? I'm already tempted

time to try it lol

man, zentimings always looks so silly with bclk

this is so weird

cores 1-7 got faster but core 0 got slower

with 100 bclk all cores had very similar effective clocks

[short blade]

looks like static OC is definitely gonna be the way to go if i want to push this chip further than what PBO can do

[short blade]

alright I got humbled real quick by small ffts lol

[short blade]

welp, back to PBO, it's not clock stretching with 24/50/50/50/50/30/35/35 lol

i used to be able to tell if the test errored out with screen off by listening to the fans but now it's too quiet for that

[steady lance]

Top 6-11 are using Direct Die, so I might have a shot once I am in a situation to do so.

[short blade]

you can see one of them has a 1080 rad and one of them is using phase change

then there's some ln2 subs

[steady lance]

The Mo-Ra external radiators yes?

[short blade]

yeah

[steady lance]

Are those things able to cool the fluid back to ambient?

[short blade]

for a short benchmark the coolant probably barely heats up yeah

this chip is kinda scaring me now

its just... taking 30/50/50/50/50/50/50/50 without complaining

no clock stretching

lower overall heat from lowering multiple cores at once improving stability compared to doing each core 1 at a time?

cause I swear core 0 used to be unstable at -25 and core 4 used to be unstable at -20

[short blade]

alright it's just taking -50 all core lol

something's not right

alright OCCT did not like that lol

[steady lance]

What tests?

Just OCCT?

Core 15 is horrible on my CPU.

Barely passes the criteria for 5.75ghz stock.

[proven canopy]

https://en.wikipedia.org/wiki/Newton's_law_of_cooling

Mora are convenient if you want to put one on the wall, or stand with qdc, for most, high end rads in a case are just fine.

[steady lance]

What do you mean by QDC?

Quick Disconnect?

[proven canopy]

quick disconnect fitting

[steady lance]

Gotcha.

I had an idea in #old-cooling a while back.

[proven canopy]

Lol that that got separated out

[steady lance]

I wanted to create a portable cooling tower with 4x420mm radiators setup vertically in push/pull with a powerful fan in the bottom and top setup as Intakes to feed into the middle where the 4 radiators will get air from then expel outwards.

[proven canopy]

I don't know why you think you'd need that many

Buy a single 420 and a used window ac...

[steady lance]

I intentionally wanted it a bit excessive for the aesthetics.

It's by all means a overkill idea.

[proven canopy]

Spoiler on a honda civic vibes

[steady lance]

?

[steady lance]

Have you ever watched Jurassic Park? If so do you remember the scene where Dennis Nerdy (Fat Computer Nerd) goes into the Cold Storage and pulls open the cylindrical DNA freezer?

Basically what I wanted to do was create a radiator tower that was reminiscent of that DNA freezer with the shape and how it pulls out.

[proven canopy]

I read both books 20 something years ago, and yes, I have unfortunately seen the movies past JP1

[proven canopy]

hah https://www.reddit.com/r/watercooling/comments/gzdck0/meet_the_zalman_reserator_found_this_in_my_closet/

I remember wanting one of those in 2008 for my e8400

[steady lance]

That's cool!

[steady lance]

.
Can the increased voltage from Scalar be offset by a negative CO to keep the max voltage within the default of 1.4v?

Wondering because I set Scalar to 10x yet my max Voltage in Single Core is still within defaults.

[steady lance]

Seems so.

As I am lowing my CO values, my max core voltage has also gone down.

Using SVI3 to monitor Voltages.

[short blade]

scalar does nothing on my 9800x3d, dunno about 9950x3d

had to drop core 5 from -30 to -25 to be stable in aida64 avx512

should be stable in just about everything now

[steady lance]

Not enough runs for a definitive answer yet, but for me it looked like it was holding clocks longer by about 20-50% and possibly slightly higher clocks.

But more importantly, I found that by utilizing Core Voltage Offset I can make Negative CO's that weren't previously stable, now far more stable. So I can use a strong negative CO to get high frequencies into the V/F curve ceiling then use Core Voltage offset to make sure the CPU gets enough power.

[short blade]

uhhh

CO is just negative core voltage

so if you combine CO with positive core voltage offset you're defeating the purpose of CO

don't do that

[steady lance]

@short blade how much variance would be considered acceptable before it becomes Clock Stretching?

[short blade]

just check if you gained or lost performance

[steady lance]

Rechecking my CO values for my sanity

[steady lance]

5.88Ghz on Core 1!

and seems fairly stable, only time will tell

Will work on the other cores later

[steady lance]

@short blade, @frigid snow, @modern walrus so after having spent a week figuring out what each core is capable of, I started doing per core tuning last night and Core 1 (CCD0) is doing 5.85Ghz effective without problem! If I lower the CO by 1 I should hit my Fmax limit of 5.865Ghz.

[steady lance]

Seems stable enough in CC.

[modern walrus]

anyone else getting these? not sure if riser bad or another reason to dislike my motherboard rn lol

Also, PSA: I set LCLK the same as FCLK (2133) & got a bunch of WHEA errors. I have no idea wtf this is supposed to be set to; if faster or slower makes any meaningful difference to anything; or how LCLK is supposed to help with USB issues on AM5. Keeping it on Auto until I get educated ig

[steady lance]

This is very likely due to the LCLK.

[modern walrus]

Ehh not sure tbh. I just crashed after CMOS clear. Strange

[steady lance]

you may have corrupted your os.

[modern walrus]

well

I reloaded the BIOS profile that passed VT3 208x

iirc the only major difference is that I disabled virtualization

but then again, PBO isn't enabled rn

but looking much better

[steady lance]

@modern walrus @frigid snow

Ran the test for 12h30m, no errors.

Kargu (Cache+FPU)

So I have a good baseline for my CO, gonna work on detailing the cores later.

[modern walrus]

noice

[frigid snow]

Who’s pinging me

Oh

[steady lance]

MEEEE!

[frigid snow]

My test failed my breaker tripped

[modern walrus]

I just updated BIOS & now am redoing memory OC/torturing myself

[frigid snow]

Ima be here tommrow Undervolting my laptop

[steady lance]

https://tenor.com/view/nerd-alert-gif-19367699

[frigid snow]

🙂

[steady lance]

Good.

[frigid snow]

I got a 125u core ultra

[modern walrus]

btw what VDDG IOD/CCD & VDDP voltages do you set?

[modern walrus]

how? I thought laptop fun was over with plundervolt microcode like 4 years ago

[frigid snow]

Idk

Ima lower voltage until it crashes

[modern walrus]

supposedly those three voltages can help stabilize FCLK at lower voltage

I'm at 2133 at 1.255 VSOC & it gets into Windows fine but then ycruncher starts laughing at me

[steady lance]

To my knowledge on AM5:
VDDP = safe up to 1.15v (I have seen some run higher but not recommended and likely unnecessary)
VDDG = various people say is say up to about 1.1v some say 1.15 others say 1.05v, generally speaking from what I can tell VDDG-CCD can usually run 0.5v less than VDDG-IOD.

[modern walrus]

is there any reason to run them at different voltages? I think BIOS even tells me they should be the same

[steady lance]

don't need to.

Just for me at least, VDDG-CCD typically uses 0.5v less than IOD on AUTO.

[modern walrus]

I'm at 1.15 VDDP & 1.1 for both VDDG. VDD Misc at 1.15. I might start trying lower

do VDDG voltages even show on hwinfo? I couldn't find them & of course my motherboard doesn't tell me

they also moved VDDIO, & VDD/VDDQ for memory to somewhere stupid for no reason

never again Asrock

[steady lance]

I am not sure.

I know CORE, MISC, and SOC are shown in HWinfo.

[modern walrus]

heck I dropped the voltages & now it looks like it won't POST lol

I find myself oddly missing MSI motherboards & Intel CPUs lately. The grass is always greener ig

[steady lance]

I find that MSI makes good bios as far as User-Friendly UI goes.

[modern walrus]

I like that you can jump around with the arrow keys

[steady lance]

I have had 2 MSI, 1 Gigabyte, and 1 Asus board.

[modern walrus]

hitting Esc for the 98th time is annoying

[steady lance]

MSI has good BIOS layout
Asus has decent bios layout
Gigabyte has okay bios layout

[modern walrus]

I haven't had a gigabyte board in awhile

you know who had a really nice BIOS layout? EVGA

[steady lance]

rip

[modern walrus]

sucks they're just a decent PSU company now

I'm gonna throw caution to the wind & set Auto for all those little bs voltages & see how it goes lol

[steady lance]

When set to auto, after reloading into bios does your board tell you the voltage supplied?

[modern walrus]

not for IOD/CCD but others ya

[steady lance]

Like mine, if I set VDDG voltages to auto, once I reload into bios after booting into windows it will say that VDDG-CCD min-voltage = X and max-voltage = x.

[modern walrus]

oh ya I forgot about that

in AMD Overclocking Settings

I'll check after this trains

[wintry sleet]

@steady lance looking for best performance, just don't ruin the chip lol

Unfervolting apparently helps it

[steady lance]

First, did you tune your ram or just running XMP? Make sure your ram settings are stable before tuning the CPU or you will have a miserable experience finding the issue.

[wintry sleet]

Xpo

Yeah

I don't wanna do that much, just pbo and lower voltage a lil

[steady lance]

Whether you do Per Core or Per CCD it will take about the same amount of time when undervolting and both are long processes.

Because core quality varies so much within each CCD, it makes Per CCD undervolting very inefficient and a overall waste of time.

Expect to spend about 1-2 days per core (assuming hour or so of testing and a overnight test).

.

If you are still interested, let me know and I will give you the next steps.

[wintry sleet]

Holy heck

Is it even worth it

[steady lance]

Depends on your CPU.

If your CPU has good cores it can be extremely rewarding.

If it has bad cores, then well don't expect much.

Typically results yield about 3-5% performance on average with good chips yielding about a 10% increase in performance.

[modern walrus]

meh I don't feel like fighting BIOS for 6400/2133 FCLK anymore. It'll run but with maxed (1.3v) out VSOC. I'll settle for 6200/2066 & see if I can get tCL stable at 26

I'm 8 minutes into VT3/FFTv4 so there's that at least

I'm gonna turn off the A/C in this room & let it run overnight. If it gets thru that, I'm calling it stable lol

but then again, 6400c28 made it thru 203 iterations of VT3 then starting acting crazy like 4 months later so who knows

[steady lance]

Back

[steady lance]

1.3 is safe though?

Also fCLK is unsynced from MCLK or UCLK. It is completely pointless to try and match it to the false idea of 3:1.

[steady lance]

.
@modern walrus what do you think my odds are of getting -35 at 6ghz to work on core 8?

Stable for daily to be specific

[steady lance]

Hitting 5.95 at -25 currently. -35 froze.

Welp bsod at -25

[wintry sleet]

But this doesnt do anything for longevity?

[red terrace]

he's reducing voltage. not increasing

[wintry sleet]

Yeag

[modern walrus]

ehh I dunno PBO is weird cuz some super light load can cause restart it seems

[modern walrus]

1.3v is the max the board allows since AM5 CPUs were blowing up like a year ago or whatever lol

the thing is, FCLK might be stress test stable but PCIe & even USB act up sometimes when you're really pushing it

or at least that's how it seems for me

I would get random USB disconnects sometimes & goofy driver errors with wi-fi card for example

then recently I started getting random restarts even on loads as light as stupid crappy Roblox lol

(my son makes me play)

messing with LCLK somehow mitigated the PCIe errors but exacerbated the random restart issue so I ended up updating BIOS & now I'm still here trying to get stable daily mem OC

& will likely have to reinstall Windows

but I usually reinstall like twice a year anyway

[steady lance]

Yes/No. It uses less voltage but at the same time tries to boost higher which causes other factors like power and such to come into play.

This is about performance not longevity. With and without the PBO the CPU will still last 5+ years if taken good care of

[steady lance]

Weird, SOC shouldn't affect that at all.

[wintry sleet]

Oh alright, I thought i heard some issues with x3d and voltages making it last not very long

[modern walrus]

oh I thought high FCLK led to issues with sound crackling, USB dropout, & other crap on 5xxx CPUs. I don't have wi-fi or GPU issues at stock settings

is OCCT VRAM + Prime95 Large FFT still best FCLK test? I may still be on AM4 mentally here lol

[steady lance]

Any test with a performance measurement that stresses the RAM and Cache is good.

I personally recommend either Karhu (Cache+FPU) or Y-Cruncher VT3 (sometimes called VST3).

[steady lance]

Your thinking about back in early 7000 release, the stock allotted voltages were too high which resulted in exploding and degraded CPUs (north of 1.4V SOC and 1.5 or higher core). Nowadays without disabling limits the highest you can go is 1.3v for SOC and 1.4v for core.

[wintry sleet]

Ah ok

So just turn pbo on and drop core voltage 10mv

[steady lance]

Not directly no. What happens is the data is being corrected when errors occur which in your case it seems it's correcting the incoming data from the USB devices is my guess.

[steady lance]

Start with per CCD and start with something unrealistic like -30 then keep reducing the CO till you find the stable CO that all goes on that CCD can do.

Once you have figured that out for both CCDs then start doing per core CO values. Drop the CO in increments of 10mv so -25 to -35 if erroring but no freezes then do +3 if it freezes do +5 and keep repeating this per core till you find the best stable Undervolt for each.

[wintry sleet]

Alr

Ty

[steady lance]

@frigid snow @modern walrus progress

[steady lance]

.

Are there tools that consumers can purchase to build their own RAM stick? I remember there being a tool for DDR4 that allows the user to remove ram modules from one stick and attach it to another one to create a RAM stick with only the best RAM modules.

[modern walrus]

You could swap out ICs onto an existing PCB you desoldered the existing ICs to but I'm not really sure how you'd bin the individual ICs

[static oasis]

I bought a 775 modded xeon x5460 for my optiplex 360, thinking it was an easy mod... but many links to modded bios files and sites discussing this one mod are broken.
can someone help me find a proper working guide?

currently on latest A7 bios, unsupported cpu post message

[proven canopy]

3dmark 11 physics works well.

[proven canopy]

You can do that with a hot air station. Get an 862D

[steady lance]

I feel pretty confident with my baseline CO now.

[steady lance]

Is that essentially a heat gun and soldering tool combo?

https://www.amazon.com/St-862D-Soldering-Station-Automatic-Desoldering/dp/B0B5XCTBMS/ref=asc_df_B0B5XCTBMS? found this from a quick search for the 862D.

[steady lance]

I am not exactly certain regarding the binning, if I had to guess it'd be through selecting the top 2 ram kits then mix and matching the ics till you yield the best results.

[proven canopy]

The atten 862d is much cheaper on AliExpress. I think I paid 165

[modern walrus]

local repair shop offered to put a new cap on 13600K for $100. I didn't even reply lol

[steady lance]

new cap?

[proven canopy]

Like a hat

[steady lance]

Is he talking about an IHS?

[modern walrus]

capacitor that I broke off at some point

should look like this

I can order the capacitor but my terrible soldering skills, need to go see the optometrist, & mediocre soldering tools might lead to bad outcome

Señor Forks could fix that in his sleep lol

[steady lance]

Oh.

[modern walrus]

ya pretty crappy. oh well

[modern walrus]

Damn... Just crashed after playing a game for about 10 minutes at stock settings & POST code is stuck at 00.

Thinking it's gotta be CPU or motherboard issue hmm

[red terrace]

it's a really small pcb you should be able to do it pretty easily

[modern walrus]

Tbh the part that worries me is how close it is to the SMD next to it

[red terrace]

as long as you go sort of quickly you should be able to just tack some solder onto each pad, then tweezers+iron the cap onto your tinned pads

the smd is small enough that you can probs heat it from the far end (away from the other caps), so tinning the pads is the only slightly concerning part

and atp you just use a fine tip iron and angle your heating element so that you just move with ur wrist more than your arm if that makes sense? gives you more control and stability

[modern walrus]

hmm okay. I've been meaning to stop being trash at soldering so mayb eI should just go for it lol

[short blade]

@modern walrus the ram fan bracket came in

it's perfect

exactly what I was hoping for

super excited to see results

[short blade]

preliminary results: 5C drop with side panel on, 14C drop with side panel off

not bad at all

well worth the $9 lol

[alpine mantle]

you got the thermals right

[steady lance]

I gotta get something like this.

[steady lance]

@short blade what fans are those?

They look nice.

[short blade]

which ones

have 2 different types of case fans in there

[steady lance]

Both look nice.

[short blade]

neat, 6400c26 should be doable

ram temp peaked 35c

[short blade]

https://www.amazon.com/dp/B0CP7LBJP2

https://www.amazon.com/dp/B0D7N5CK8V

two very different price classes lol

[steady lance]

needs more vdd.

Very rarely does Hynix DDR5 SR sticks do 6400cl26 with less than 1.65v VDD from what I have noticed.

Progress.

Just slowly testing one-two cores per night.

[steady lance]

Yikes, indeed.

[short blade]

if it got this far with 1.61 then 1.62 should almost certainly work, no?

[steady lance]

If I had to guess you will probably need 1.63v.

Working on this CO for so long I am starting to wish I started with RAM tuning lmao

I am eager to get my hands on 10950x3D when that releases.

That and either this or the royal variant of this RAM kit.

https://www.amazon.com/G-SKILL-Trident-CL26-36-36-96-Desktop-Computer/dp/B0F1XN7DWN/ref=mp_s_a_1_1?

[steady lance]

I hit 19Ghz xD.

[short blade]

1.63v it is

[short blade]

getting game freezes with no driver timeout but no performance regression

can tell the game is still running in the background cause of third party app that tracks party dps still updating

but it just stops drawing new frames

have to kill and restart game

and yes i narrowed it down to vram

[red terrace]

fascinating

i've never seen this lol

[short blade]

same before this gpu

[steady lance]

Nice. I am surprised, your fCLK is working at 1.295v SOC?

[modern walrus]

oh wow that came out really good. Your system looks clean af

[modern walrus]

mine does when the rest of it does lol

btw I put 12900K/3080ti system back together cuz I'd previously taken the pump out. I had to fix a few pipes that looked kinda bad. Anyway, did all that, then for the heck of it, reseated the two 12vhpwr on the GPU on AM5 system & it stopped crashing

of course, this was AFTER I put that heavy heckin thing down on the floor to put 12900K on my desk

cuz it just be like that ig

so I'm gonna test the heck out of it & probably switch back

apparently I hadn't turned on this 12900K PC since April lol

I blame, uh, the heck is that thing called... contact fretting? or aliens or something I dunno

[modern walrus]

enable End task in Developer settings. Fallout 4 used to freeze so hard I'd have to restart cuz Close window wouldn't close it but End task is all like pew pew pew

[steady lance]

hm? I am confused.

Also some progress. -15 on Core 8 and -28 on core 9.

[short blade]

@red terrace it happened, finally got a proper crash from rdna4 vram

IRQL bsod which is either ram or vram

but my ram is stable

[red terrace]

https://tenor.com/view/regretting-thinking-nervous-macaco-monkey-gif-13105982953111325972

[short blade]

occurred way below the performance regression point as well

[red terrace]

that's wild

[short blade]

performance regression point on my card is 2810 MHz

[red terrace]

are you on samsung memory perchance?

[short blade]

bsod occurred at 2780 MHz

hynix

[red terrace]

interesting

i'm on hynix too

[short blade]

don't think samsung clocks this high

[red terrace]

2780 is my safe-stable setting

[red terrace]

really? i thought it clocked higher lol. i forgor

[short blade]

I heard samsung runs much lower temp but won't oc past 2600s

[red terrace]

interesting

[steady lance]

Does the vram maker on the gpus matter a lot for overclocking like it does with DDR5 oc?

@short blade

I assume that's what you guys were talking about about

[short blade]

yes

you want hynix vram on 9070xt

[steady lance]

So basically Hynix every thing.

Do Nvidia and Nvidia AIB cards all use the same vram or?

[red terrace]

depends on the SKU. currently you can get all three manufacturers in GDDR6 with Nvidia, just hynix and samsung with RDNA 4 afaik, and i beleieve it's just samsung GDDR7 atm on nvidia

[steady lance]

Thank you for the info. It was unclear from my Google research if the 5090 currently uses just Samsung or Hynix as well.

[red terrace]

there might be hynix and micron, i've only seen samsung though

[steady lance]

@frigid snow, @modern walrus, @short blade big news!

ASRock is releasing the Taichi OCF a 2 slot X870E board for OC!

https://videocardz.com/newz/asrock-unveils-x870e-taichi-ocf-its-first-amd-oc-motherboard-alongside-new-creator-nova-and-livemixer-models

6 M.2 slots (2x 5.0x4).

https://www.asrock.com/mb/AMD/X870E Taichi OCF/index.asp

[short blade]

one day i'll get a fancy board lol

im chilling on budget matx

4 slot, no clear cmos button

[short blade]

hey I finally figured out how to match my tPHYRDLs

now... is it stable?

tPHYRDL was 35 on A2 and 37 on B2 with ARdPtrInitVal on auto

setting it to manual 0 has both channels synced at 35 now

desynced tPHYRDLs were in the back of my mind bothering me a bit so hopefully this is stable

[steady lance]

Nice.

I am jealous you can reach 20,000% covering in 6 hours lmao.

What would you guys recommend using to test synced 1:1 vs desynced 2:1 memory performance?

[short blade]

finally broke 243 test speed

i think this is the new limit of my ram with the fan

not sure why, but tRDWR now works at 14 (used to not post at 14), dunno if it has relation to tCL, VDD, tPHYRDL, or whatever

tRCDRD definitely showed some scaling with VDD, tRCDRD 36 got to 500% instead of 7% before erroring, but def still not stable

also for some reason i was able to drop vsoc a bit so that's nice

i don't really understand tPHYRDL beyond "matching them is good", dunno if that's related as well

[short blade]

probably aida64 and a game benchmark

still perfecting the CO, currently at 24/32/30/30/30/30/36/36

[steady lance]

Nice.

[steady lance]

Thank you.

[steady lance]

Same, I am almost done testing CCD1's CO without a voltage bump.

[steady lance]

@short blade @proven canopy give this article a read.

https://videocardz.com/newz/colorful-launches-igame-shadow-iii-ddr5-memory-for-ryzen-9000-with-cas-latency-down-to-26

• 6400 cl28 1.4v
• 6000 cl26 1.45v

[short blade]

6400c28 at 1.4v sounds like a crazy bin

if true I imagine that bin is going to be rare

wonder what kind of voltage that bin would need for 6400c24

[steady lance]

I am keeping my eyes open for that cause at 6400cl28 that edges out 6000cl26 by 0.16ms

All while needing less voltage compared to G.Skill's current 6000 cl26 1.45 kits.

[short blade]

think i need around 1.5v for 6400c28

[steady lance]

It makes me wonder if 6400cl22 is possible at 1.6v.

[short blade]

probably not

[steady lance]

Do you think it would cost more or less than G.Skill's closest equivalent?

[short blade]

no clue man im on a $50 ram kit

[short blade]

interesting, 1.285 vsoc works now

I remember that errored very quickly before

not sure which setting has influenced vsoc

[steady lance]

What voltages have you changed recently?

[short blade]

dropped vddio from 1.4 to 1.1

dropped vddp from 1.0 to 0.975

raised vdd from 1.4 to 1.63

[steady lance]

The RAM was probably just lacking VDD voltage then is my guess.

[steady lance]

@short blade thoughts on why Core 15 (#8 of 8 on CCD1) past -40 CO it won't boost higher? It is not hitting fMAX and it's negative CO now at -60 is allowing it to boost higher in heavier workloads but it still won't boost higher than it did at -40.

Do you think it's it's limited by silicon quality or voltage?

[short blade]

it was 1.4v for cl30

same as expo

[short blade]

no idea

dont have a fancy 9950x3d to play with

[steady lance]

Working tuning CCD0 now before I refine CCD1.

As it stands, CCD1 clocks between 5.8-5.89Ghz

[modern walrus]

I dunno if I like Asrock after my X670E Taichi experience tbh lol

grass is always greener on the other side tho ig

strange this is like the first 1dpc AM5 board

[tall pelican]

https://www.newegg.com/asus-rog-crosshair-x870e-apex-atx-motherboards-amd-amd-x870e-am5/p/N82E16813119734

[steady lance]

Hopefully it will be better than the Apex, I like the style more and on paper it looks more robust feature and spec wise.

[modern walrus]

oh damn I thought Apex was for Intel & Crosshair was usually the best Asus AMD board.

how the hell did Intel fall off so hard?

[modern walrus]

it's possible. OC Formula was a really good Asrock board & (iirc) the Riptide is really good for OC. I've just been frustrated with annoying little things with my Taichi at various times. It's a little unfair for me to knock Asrock tho ig since I'm using a delidded CPU with a direct die frame I took a Dremel to with a waterblock that's not for direct die that I basically had to rig up my own mounting hardware for lol

so not exactly a stock experience

[steady lance]

Maximus = Intel
Crosshair = Asus

[steady lance]

Yeah.... this seems like something else might be the cause lmao

[modern walrus]

I never really paid attention to the maximus part ig. There's no X570 Apex tho no?

[steady lance]

Not sure why but part of Asus's description is that the Apex is for overclocking enthusiasts. I don't think OC enthusiasts were buying AM4.

[modern walrus]

lol maybe for those crazy people who needed 2000mhz FCLK

I could never get it to run without 50 WHEAs/second

[steady lance]

xD

[modern walrus]

AM4 would do like 4800mt/s+

the problem is it was useless except for like ycruncher

[modern walrus]

8000mt/s DR? heck https://www.newegg.com/p/0RN-00MB-000S7

[steady lance]

?

8000 has been around for a while now.

[modern walrus]

2x32gb?

[steady lance]

Ahh I missed that. That's pretty good and fairly recent then.

Though 2x32 7600 has been around for a minute or two.

[modern walrus]

I couldn't get <7600mt/s to POST on 2x32gb lol

but that's using 5600c46 fancy memory

[steady lance]

die, heat, voltage?

[steady lance]

@proven canopy what are your thoughts on this ram kit that @modern walrus shared with us?

https://www.newegg.com/p/0RN-00MB-000S7

[proven canopy]

Seems fine

[steady lance]

Better or worse bin you think compared to the 6000cl26@1.45v kits G.Skill sells?

[modern walrus]

A-die, watercooled, & like anywhere from 1.3-1.65 iirc

[steady lance]

I think 8000 needs like 1.8v especially on low bins.

For 2x32 assuming both the board and RAM can take it.

This is based on the posts I see putting in the OC group.

[steady lance]

I finally almost have my Undervolt done per core, @modern walrus

4 cores left to go on CCD0 then I am gonna start seeing how far I can push the frequency with voltage offsets.

[modern walrus]

for VDD? heck I'll try it lol

[steady lance]

Currently averaging 5.75Ghz on CCD0 and around 5.85Ghz on CCD1.

[modern walrus]

does this board do 1.8? I'm not even sure haha

I mean DDR4 boards usually did 2.2 for memory voltage so probably

[steady lance]

Once my situation stabilizes and I have some pocket change, I am gonna get the stuff I need to mount my AIO as a direct die cooling solution.

[steady lance]

Mine let's me give the ram up to 2.4v sooo

Try at your own risk lmao

[modern walrus]

doesn't noctua make one?

[steady lance]

Yeah but it's a giant air cooler.

With how often I move my PC that'd be a very stupid idea for me.

[modern walrus]

ya & I doubt they'll sell the bracket alone or if they do, it probably doesn't connect right

[steady lance]

Bracket is sold separate for that cooler and is only compatible with that specific one.

[modern walrus]

ya figures lol

I understand why people use air coolers but I don't care for them

[steady lance]

btw as it stands, I am looking into getting a job as a full time ranch hand here in alaska soo if that happens I am gonna be enjoying winter for oc time.

[modern walrus]

whaaaa

radiator out the window benchmarking

[steady lance]

1-3c nights with drops into the sub-zero zone.

[modern walrus]

use car antifreeze cuz you'll actually need it haha

it's almost 10pm here in Miami. It's still 85F outside lol

[steady lance]

Yikes.

It's 64f here.

Nights drop to 12c during the summer.

[modern walrus]

damn even that's good benching weather

[calm grail]

I gotta wait till winter

[steady lance]

I don't think I have had a dual CCD in the past where the CO values on both CCDs varies so much.... -15 all the way to -45.

[modern walrus]

my memory OC fails after like 26 hours of VT3/FFTv4

I'm tempted to just not care & leave it

seems like that could be some power saving related crap I dunno but I'm sick of fighting with it

maybe my win11 install is just derp

could also be tFAW 16 ig

[steady lance]

Do you have PDM (Power Down Mode) on or off?

Or are you talking about Windows Power Settings?

[modern walrus]

I figured windows, chipset, or some other nonsense. Issues are with Power Down Enabled or Disabled & MCR same

starting to think SOC is degraded or something

what's strange is memory OC would survive FFTv4/VT3 ycroncher for like 20+ hours

but Expedition 33 crashes quick

but Expedition 33 runs fine at stonk settings so that at least eliminates GPU or PSU being the issue

[steady lance]

@modern walrus my CPU definitely can do 19.8Ghz, definitely...!

[calm grail]

The heck

[steady lance]

It's a misread lmao.

CCD0 does 5.8Ghz though.

[calm grail]

Nice

[abstract merlin]

How do misreads happen

What makes it report that

Or 100,000 C

Oh its the software misinterpreting the data from the cpu

Well thats one of the reasons at least

[modern walrus]

dayum & I can't even get my borked FCLK to hold 6000 1:1 rn lol

might be stretching clocks a bit

[steady lance]

FCLK at 6000? Did it actually post lmao?

[modern walrus]

lol I mean to hold 6000 1:1 with FCLK at 2000

everything runs fine at stock settings tho

not sure if it's CPU, board, or memory that I borked

Can't even keep this running

[steady lance]

Ohh okay.

[steady lance]

Borked? How could you've done that?

[modern walrus]

tbh not sure but there are two possible explanations that likely go hand-in-hand

(1) running 1.285 SOC at LLC 2 for almost 2 years

(2) Asrock motherboard

supposedly they were frying 9800X3Ds somewhat frequently so degrading my CPU over a couple years doesn't seem too far-fetched

[short blade]

found out something interesting

SCLs 5/5 are unstable when gpu is loaded

temps are fine, tRFC/tREFI are auto, etc

narrowed it down to SCLs only

SCLs 5/5 errors in aida full system test within 30 minutes

SCLs auto passes overnight

[modern walrus]

found out something interesting myself... at 6000c26 1:1, VSOC 1.255 fails OCCT VRAM/P95 Large FFTs in two cycles. In my continued throw spaghetti effort, I got annoyed & dropped to 1.1 VSOC thinking I'd increase by .01 everytime it borked out. To my surprise, 1.1 VSOC not only POSTed, but is running thru tests. There's been an error on Worked No. 1 so far but that's much better than my usual mess of the whole system restarting

[short blade]

what fclk?

[modern walrus]

2000

[short blade]

fclk has negative scaling with vsoc

[modern walrus]

nothing fancy at all

I ran 2133 with 6400 1:1 for months

[short blade]

wouldn't expect 2000 to be a problem but that could explain why dropping vsoc could improve stability

[modern walrus]

at 1.285 VSOC LLC 2 lol

I figured I degraded it

[short blade]

hmm interesting

[modern walrus]

BZ claims FCLK is most stable between 1.1-1.2

heck I keep getting worked stopped on P95 rn

but I'm wondering if I should focus on VDD/Q & VDDIO now

since this is significantly different behavior than the usual reason tests fail

which btw, running ycruncher FFTv4/VT3 doesn't seem to show FCLK instability well

I could run thru hours & hours of ycruncher only to open Expedition 33 & get immediate restart

[short blade]

personally I have 2200 fclk rock solid stable at 1.295 vsoc so 💀 dunno

[modern walrus]

I knew VSOC was high & thought I should try lower but I thought like maybe 1.15 was the lowest I'd start messing with it

[modern walrus]

at 6400 or 6600 1:1?

[short blade]

6400

can't run 6600

[modern walrus]

1:1 seems to scale with VSOC

[short blade]

yes

[modern walrus]

I haven't really tried to POST 6600

[short blade]

UCLK scales positively with VSOC
FCLK scales negatively with VSOC

[modern walrus]

I couldn't get 7600 1:2 (not that this is in any way relevant lol)

[short blade]

I'm running 6400 1:1 with nitro 1/2/1

[modern walrus]

lol gotta balance a seesaw ig

[short blade]

nitro 1/2/0 doesn't post at 1.3 VSOC

[modern walrus]

I've always left nitro on auto which iirc is worst

[short blade]

auto = 2/3/1

[modern walrus]

tbh I don't really know what Nitro does exactly ngl

[short blade]

massive massive performance difference

like 6000 1/2/0 probably beats 6400 2/3/1

[modern walrus]

I'm thinking dropping VDDIO should be my next try. It's at 1.23 rn

[short blade]

afaik it only scales with VSOC as well

[modern walrus]

does nitro tune timings that aren't exposed in BIOS or something?

[short blade]

I need 1.295 VSOC to run nitro 1/2/1 at 6400 but nitro 2/3/1 can run at 1.27

[modern walrus]

this is the first shimmer of hope I've seen in days of doing this off & on now

[short blade]

yes, exactly

[short blade]

I have mine at 1.1 lol

VDDIO >= VDDP + 0.1V

I have VDDP at 1.0 and VDDIO at 1.1

stable

[modern walrus]

oh heck

it just did it

total system restart

the hardest part of DDR5 OC is trying to forget everything that worked for DDR4 OC I swear lol

[short blade]

you might have to do some digging for this but try setting ARdPtrInitVal to manual 0

[modern walrus]

ngl, I don't think I've ever even seen that setting before

[short blade]

with ARdPtrInitVal on auto I had tPHYRDL= 35 on A2 and tPHYRDL = 37 on B2

[modern walrus]

is it with Nitro settings?

[short blade]

setting it to manual 0 synced tPHYRDL to 35 on both channels

reminds me of RTL/IOL on intel

[modern walrus]

lol I miss figuring those out

[short blade]

no, mine is somewhere in amd overclocking menu

[modern walrus]

the latency benefit was insane

[short blade]

might be able to just search for it

yeah, setting ARdPtrInitVal finally let me break 243 MB/s test speed in karhu

[modern walrus]

hmm didn't find it

you're on Asrock board too right?

[short blade]

yeah

[modern walrus]

do you set Memory OC Mode to Enable & then Memory OC Profile pops up? I wonder if this would help

just stumbled on it now lol

I have Read & Write Drift Adjustment but I don't see anything about initial values anywhere so far

[short blade]

don't think so

[modern walrus]

do you mess with LCLK Frequency?

I'm gonna see if 6600 1:1 even POSTs lol

[steady lance]

What is this setting and does it matter if both DIMMs show the same value?

[short blade]

it's a lever for controlling training of some timings that you don't have direct control over

tPHYRDL being the one that shows up in zentimings

if both dimms show the same value it might mean your board is autoing to 0 or 2 or whatever

if it's not 0, then 0 should have a performance increase (if stable)

I noticed that my tPHYRDLs match at 6000 with auto and unmatch with 0, so I don't think it necessarily means 0 = match

so like at 6000 maybe
ARdPtrInitVal 0 = tPHYRDL 33/35
1 = 35/35
2 = 35/37... etc

and at 6400 maybe
0 = 35/35
1 = 35/37
2 = 37/37... etc

but yeah reminds me a lot of RTLs/IOLs

[short blade]

no, honestly have no idea what it is

google suggests that it's the speed of the link between the IF and the PCIe bus, and is generally pointless to overclock as you're not bandwidth constrained there

I generally find that if you have to get too deep into the esoteric stuff to buy a semblance of stability, that semblance tends to be very fragile

maybe downclocking LCLK could improve bclk oc potential on boards without eclk, but otherwise doesn't seem relevant

[calm grail]

What in the acronym vomit

[short blade]

from an outside perspective there is little difference between overclocking and the ramblings of the utterly insane

perhaps even from an inside perspective

I am not well

[calm grail]

U alpha and ggood are legit the only ones who talk here

[short blade]

that's not true

we're only 95% of the channel not 100%

[modern walrus]

no waaay

crashed lol

[short blade]

have you tried increasing vddp? SCLs?

[modern walrus]

I tried VDDP at like 1 before but haven't really messed with it otherwise

[short blade]

try up to 1.1-1.15

increase VDDIO to 1.25 if trying 1.15 VDDP

[modern walrus]

I'll try it

oh those should be linked like that? are other voltages linked like that?

do you set VDDGs?

[steady lance]

lmao.

[steady lance]

agreed.

[steady lance]

I just noticed for some reason, my tPHYRDL is 36/38. Both were equal prior to me using bCLK.

[red terrace]

i will try to rectify this. i think i wasn't a member when i was messing with my 9070, and my a770 is effectively unmoddable. b580 soon tho

[steady lance]

Looking forward to it.

[red terrace]

it's here!

[calm grail]

Absolute w

Titan might be able to help get u to like 3.5ghz

[red terrace]

i alr know how to mod it, but it's a new card with an actual warranty so i'm going to daily drive it for a day or two to make sure it actually works before i go soldering

[calm grail]

Alr

[red terrace]

who is titan btw?

[calm grail]

This dude

[red terrace]

ah

[calm grail]

He technically has the overclock record for a photon b580

[red terrace]

i have the record for fastest 7800x3D in my town while raining on 8/11/2025

i also have wr for 9070 in a bunch of benches and used to have a 6800 wr. driver updates nuked my score tho, everybody got like a 3% perf bump and i'd already sold the 6800 by then

[inland swift]

YO

GOOGLE AI IS SAYING MY POST

[calm grail]

W

What's the query I wanna try it

[inland swift]

Chat im so sigma

That even ai has recognized my genius

[calm grail]

Ac w

[red terrace]

first b580 mods today when i'm off work

[red terrace]

forgor to upload this, but 10th place graphics score, third place by user. i'm being beat out by a bunch of reruns from the same two fellas lol
https://www.3dmark.com/spy/58108203

unfortunately or fortunately my LE a770 arrived and i'm more interested in benching that atm, so more b580 scores will need to wait

[abstract merlin]

How should I go about rdna 4 undervolting

Should I increase negative offset as much as possible and then increase frequency

I can do -70 +300 mhz but at lighter loads I was crashing

or rather the whole computer freezes because linux

I was streaming and it crashed with the additional load from encoding

I increased to -65 and thought I was good until I had a host migration and went to really low loads

Then froze again so I moved that to -60 and was stable

So I started to think maybe +300 is too high

I turn that down to 250 and now I do -70 no problem

ngl

think im just splitting hairs here

[red terrace]

are you looking for power savings or better performance?

[abstract merlin]

Perf

[red terrace]

i do negative voltage offset and then core freq offset after that, but when done properly there should be very little frequency offset headroom left, since ur basically just dragging each frequency point leftwards and running at moar mhz at a given voltage

[abstract merlin]

Thats what I started to think last night

[red terrace]

nagtive V/O is just a more consistent F/O basically

[abstract merlin]

Yeah when I got the card I dialed in both very quickly

I should have done offset and take time with that

Then move on to freq

[red terrace]

mhm yah

before i modded everything i think i was like -125 v/o and like +75 core

not much headroom since ur just maxing out frequency with the V/O slider anyway

[abstract merlin]

Lol this is basically like curve optimizer pbo

Wow how did I not make that connection earlier

[red terrace]

yeah exactly lol, just balancing PL instead of balancing thermals int he case of Ryzens

[abstract merlin]

So I think I know what was going on earlier

The frequency offset was high so it took every chance it could to hit the target

But it goes too high and freezes the system

[red terrace]

ah yah

you have to get an EVC next mate its so much fun

[abstract merlin]

Edible voltage controller

[red terrace]

nummy

[red terrace]

https://www.elmorlabs.com/product/elmorlabs-evc2se/

[abstract merlin]

So because your offset is so low you basically get a big clock increase right

[red terrace]

which message is this supposed to respond to

[abstract merlin]

here

[red terrace]

ohhh

yah exactlt

pull the v/f frequency a bunch so it clocks super high via negative offset, then get whatever remaining headroom is there with core offset

[abstract merlin]

And because the freq offset is not a lot, you wont have the issue I was having with lower loads

The majority of frequency increase comes from lowering the vf curve

[abstract merlin]

right

[red terrace]

it still happens just less often. it'll still be running more frequency at same voltage, but the boost algo is a bit happier this way usually

[abstract merlin]

Damn literally pbo but larger numbers

Yeah rdna 2 was a lot different

With 4 you just slide a slider

[abstract merlin]

But like what I meant was because the frequency target is just +75, it wont be shooting too high in lower loads where the clock goes up

I lowkey think it was trying to hit 3.6 because I saw some clocks to go 3.55

[red terrace]

:aware:

[abstract merlin]

It was trying to go super high with the framerate cap on

But having that lower rids that issue right

[red terrace]

but yah idle clocks/voltage should be normal-ish since it's mooostly just messing with 3d clock curve

idle power states shouldn't be radically altered by negative CO

they will a little but not a lot, vs like idle voltage +300mhz with core offset

[abstract merlin]

That I know, but I will see how messing with just the offset goes

I wanna see if that frequency target I set was causing this

Lol 3.3 ghz at 145w

Nutty

[red terrace]

crazy, lmk what you hit under load once it's dialed in

[abstract merlin]

I do remember seeing -110 not crash but then it did

But at that time I was also touching frequency target too

So Im just gonna send it to -110 rn

[abstract merlin]

Fr

[red terrace]

i got voltage control working on my a770 but it's so jank
just saw a random spike to 1.4v at idle which is uh- not great. and locked to supposedly 1.15v spits out like 1.3v under load which obviously i did not run for more than a moment. absolutely cursed

[modern walrus]

rn I'm at 1:1 6200c34 1/2/0 8x/8x. I'm gonna see about dropping tCL. I've been really lazy about getting this OC working lol. I tried disabling GDM but it wouldn't train. Do you know if it's as darn near impossible to get GDM off on DR on AM5 as it was on Zen 3? I'm only at 1.23 VSOC so maybe if I blast it it'll GDM off.

[modern walrus]

does it give you LLC control?

[red terrace]

i don't think so. i'll have to check when i get home

[short blade]

I dunno, for me GDM off was as simple as flipping the switch but it was one of the first things I did

it's probably your SCLs

GDM off stable with SCLs 5/5 but no post at 4/4 at 6400

GDM on allows SCLs 4/4 stable at 6400

[steady lance]

What's Poppin?

[modern walrus]

with dual-rank (2x32gb) a-die?

current situation btw

I'll deal with tRFC later meh

I think that's the timing I pay least attention to ngl

AVX512 is silly on AM5. Why does it run at like 53C? why not bump frequency a little?

[red terrace]

no LLC on the i2c, i think the mp2979 datasheet is public so i can figure out if it's possible to adjust through manipulating resistances or smth

i kinda get the feeling it's not super worth tho

[short blade]

I have 2x16 a-die

probably same thing

just try raising SCLs to 5/5

[modern walrus]

5.45 hours & it errored out lol

[steady lance]

Your vSOC is fairly low. Have you tried 6400 at 1.22-1.3v?

[modern walrus]

I'll try loosening SCLs. I got this far at least

[modern walrus]

I ran 6400c28 for months then everything went bananas out of nowhere somehow lol

[steady lance]

what vSOC I am curious?

[modern walrus]

1.24v; LLC level 2; MCR disabled; PowerDown disabled; & dGPU only mode enabled

I just set 6400c30-39-39 & FCLK 2000 for testing's sake with same VSOC

will see if it POSTs

[steady lance]

Hmm, most 7000 series can't run 6400 truly stable at 1.24, might be worth trying 1.26-1.28. I find it usually best to find the what appears stable then add 0.1-0.2v as a buffer

[modern walrus]

1.275 no POST either

I'm kinda over 6400 lol

I got 6600 to POST before but not stable lol

heck

[modern walrus]

something seems off lately with FCLK I dunno. it doesn't seem to like >1.255

1/0/0 Nitro doesn't wanna POST

[steady lance]

Do you mean 1/2/0?

[steady lance]

Tried raising VDDG or VDDP?

[steady lance]

.

I am almost done with my CO values.

Then I am gonna start futsing with voltage offsets.

[steady lance]

Probably low VDDP/VDDG would be my guess.

I found that raising those voltages helped stabilize fCLK for me.

[modern walrus]

VDDP I tired up to 1.15 iirc. The VDDGs don't do well >1.050 it seems

[modern walrus]

How high? I think auto is .840 on my board

For VDDGs I mean

[short otter]

The voltage vs achievable clock speed curve for my 5090 is driving me insane

.91 MV I pull about 420 watts under max stress load (about 380 while gaming) and has room to go to about 2800mhz before it starts having issues.

In order to hit the 2900 it needs 1.05V and pulls 560 watts

That's actually insane

140 watts to achieve 100mhz higher stable is like actually crazy I don't get why it's such a harsh drop off. My 4090 was a lot more gentle with the curve

What's funny is at full 600 watts I can only get 3050ish

Which means that to get that extra 250mhz I need almost 200 watts of power. Absolutely insane. I think they really could've made this a 450 watt card

[red terrace]

this is how most generations are these days. my 3090s could run at like 250w and perform better than stock. 4070 ti super at ~190w

[steady lance]

VDDG should be safe up to 1-1.1v.

[steady lance]

Tuning for CO on Zen 5 is a bloody nightmare.

[hallow ermine]

@everyone

[calm grail]

No

[hallow ermine]

No

[pure widget]

I’m on a 9800x3d getting 20727 multi core score on cb23 is this good?(expo 6000mhz cl30 pbo+200)anything else I need to do?

[steady lance]

Does anyone here know how VDDP MISC Voltage affects DDR5 on Ryzen 9000s?

[steady lance]

Well I am back to tuning my ram. Unfortunately, 6528mhz at cl26 just wasn't happening without really high voltages to I opted for 6400cl26 with Nitro 1/2/0 and FCLK 2200.

Also, since I am now in Mexico I had to redo a lot of my other stuff since I am now living in a ambient around 24c vs 8-9c.

[steady lance]

Trying to see if I can get FCLK 2233 or 2267 to post and maybe stable before futsing with the RAM any further.

[red terrace]

4th place 7900 xt score, 2nd by user. graphics only bc 7800x3D moment

[steady lance]

@short blade any ideas on how I can stabilize MCLK beyond loosing nitro, raising VDD, and raising SOC?

[red terrace]

mclk or uclk?

[modern walrus]

heck I've been trying to get 6200c32 1/2/0 with GDM off on DR & it takes a month to train & is one heck of a hecker

so far my only useful observation has been that GDM off 1/2/0 needs significantly more VDDIO than GDM on 2/3/1

I can't POST <1.4 VDDIO

whereas GDM I'd run VDDIO at only 1.24 before

I'm thinking I might just do 6200c30 GDM on 2/3/1 cuz frankly, I'm sick of doing this lol

[steady lance]

you switched to DR?

[steady lance]

I can get 6600 to post and run fine in Karhu but for some reason VT3 keeps crashing my system even at 6400.

Also 6800 1:1 posts for me now that I have updated bio, @modern walrus.

[steady lance]

My issue with GDM has been that the I can get into windows but it then immediately crashes.

[modern walrus]

I've always been DR for AM5

two 32gb sticks are dirt cheap

btw I can get 1/2/1 to run fine but 1/2/0 will fail stress tests & 1/1/1 just doesn't POST lol

[steady lance]

I can get 6600 1:1 at 1/2/1 stable

[modern walrus]

ya I'm at 6200. 6400 might be in the cards but meh

6600 will POST & get into windows but that's about all it will do lol

I think 7xxx v. 9xxx might be involved there

I miss Intel lol

[steady lance]

this new bios has been nice in this regard.

[modern walrus]

which board are you on again?

I'm on latest for X670E Taichi

[steady lance]

Refurbished Excellent (Amazon) MSI X670E Gaming WiFi

[modern walrus]

ah jeez I just saw there's another BIOS update lmao

2 days ago

WHY DON'T THESE BOARDS HAVE DUAL BIOS

like seriously wth

my Z690 Dark K|ngP|n has FOUR BIOS options lol

[steady lance]

I miss Kingpin boards, never used one but they looked awesome and I never once heard anything bad about them

[modern walrus]

both my Z790 Apex & Z60 Unify-X have dual bios

since my AM5 system has been a work in progress, I've been on the Z690 KP with 12900K & 3080ti mostly for the last 6 months or so

I'm using some ancient M-die on here but this is the only board I know of that can run 1T

I've been running 7000c32 1T since like 2022

for daily

[steady lance]

back

[steady lance]

1T means GDM disabled?

[modern walrus]

I think it means 1 command every clock cycle (this is Intel so I'm not entirely sure if there's an equivalency lol)

[steady lance]

gotcha.

[modern walrus]

I was gonna update BIOS but tired of fighting with this. I'm not updating BIOS. I might be able to tighten timings, especially tCL, make other changes, whatever, no, I'm over it. I'll get a new CPU in the next year or so & care again then lol

[steady lance]

the tuning never stops!

[steady lance]

Trying to stabilize 6600 1:1 since it's less stable after the reseating.

I am gonna need a better cooling solution in the future.

Playing with 1.85v VDD

[steady lance]

I am frustrated, since reseating 6600 1:1 is not playing nice.

[steady lance]

@modern walrus @short blade thoughts on the paste application? Took my TF7 paste and put about two very small drops then spreaded with a spatula.

[steady lance]

Figured I'd try a flat spread vs x or dots this time around.

[modern walrus]

seems good for not liquid metal lol

[modern walrus]

for what? tCL 20?? lol

get 5-6 Pokemon cards & a fresh razor blade, it's time to take those spreaders off & watercool haha

[steady lance]

6600cl26 1:1, posts with 1.6-1.7v but not stable.

[calm grail]

lol i thought you scratched the top off or sm

[steady lance]

Thankfully not lmao.

[steady lance]

I tried 1.9v VDD but no luck, it generated to much heat to control.

I was trying to get 6600cl24 to post.

.
Also frustrating is that I am like 99% certain during travel whoever dropped my PC case damaged my PSU. It's reacting very poorly to power spikes.

So I am limited by what apps and such I can run until I replace the PSU.

[steady lance]

@modern walrus could you remind me how low I need to bring my tRFC latency to check if I have Hynix A vs M?

[modern walrus]

I think it's 120ns for A-die

You can check CPU-Z SPD tab iirc & it'll tell you if it's single or dual rank. Afik 32gb sticks are DR A-die. I assume 48gb sticks are DR M-die

[steady lance]

Mine are DR Hynix.

Haven't tested on this new bios but on previous bios versions I was stable around 130-140ns.

[steady lance]

Decided to focus on 6400 1:1. A 1.7% boost to speed is now worth nearly 15% increase in voltages that are very close to the danger zone.

[steady lance]

@modern walrus thoughts on 1.19-1.196v SOC for 6400 1:1 with nitros 1/2/0?

[steady lance]

Some progress

[steady lance]

Got fCLK 2200 to work with tweaks to SOC, VDDG, and VDD-MISC.

[steady lance]

Managed to get my tRFC below 120ns.

Going to do a long test to see if 117ns stays stable.

[steady lance]

@modern walrus @short blade no errors and ram temps never exceeded 47c!

1:1 6400cl26, Nitros 1/2/0/8x/8x, 2200fCLK, tRFC 375 (117ns), max tREFI (65535k), GDM on.

[steady lance]

FINALLY! I have reached 300MB/s test speeds!

[steady lance]

I have set:

• Frequency
• Infinity Fabric
• Nitro
• Tertiaries
• Secondaries
• Primaries
• tRFC and tREFI are as low and high as they will get.

What else can I do to improve memory performance?

.

I will post a ZT screenshot with Karhu and HWinfo later after the test has had some time to run.

Here is what it currently looks like though.

[steady lance]

I am going to consider VPP 1.8v stable and VDDQ at 1.4v and VDDIO 1.45v stable, going to try lowering VDDQ and VDDIO some more later before cranking down on VDD.

[calm grail]

I overkloc rgb

[steady lance]

Got my voltages down to:
SOC 1.22v
VDDG 1.17v

VDDP 1.15v (might try lowering)
VDD_MISC 1.22v

VDD 1.75v
VDDQ 1.32v
VDDIO 1.45v
VPP 1.8v (might try lowering further)

[steady lance]

You should overclock EVERYTHING!

Even your brain, xD.

https://tenor.com/view/alita-battle-angel-alita-army-gunnm-gif-23816665

[calm grail]

[Overclocks water bottle]

[steady lance]

Extra hydrating water!

[calm grail]

Fr

[steady lance]

I am annoyed.

I have managed to lower everything but VDD.

Some timing I tuned when I was at 1.85v VDD refuses to be stable at even 1.8v.

[short blade]

I'm jealous

I think 2200 fclk is actually not completely stable for me

never errors in any test and I see performance gains compared to lower fclk

however, once in a while when gaming at 2200 fclk I get a massive stutter (framerate and audio) that doesn't occur at 2000 fclk

so I might pursue a 6200/2067 daily, not sure yet don't really have time to mess with the oc atm

[steady lance]

What is your SOC, VDDG, and VDD-MISC voltages at?

SOC and VDD-MISC should always be higher than VDDG by around ~50mv or more to be stable. (Difference is board dependent).

For me I am running VDDG at 1.17v so SOC and VDD-MISC are both at 1.22v.

@short blade

[short blade]

2200 fclk profile is
vsoc 1.295
vdd misc 1.1
vddp 0.975
vddg 0.96

tried raising them all individually, didn't change anything

[steady lance]

Hmm I'd try raising VDDG and keep VDD-MISC at least 50mv higher.

Did you try that?

I found that VDDG is a Low Dropout Voltage for VDDG and that if VDDG gets to close to VDD-MISC voltage it can cause instabilities I think the reason was due to to much noise or something like that.

@short blade @modern walrus @calm grail what do you guys think of the ram tune?

[calm grail]

That is a lot of acronyms

https://tenor.com/view/dumb-cat-kitty-stupid-funny-gif-23207376

[steady lance]

I tightened every timing.

[calm grail]

Which will make it faster, i am given to understand?

[steady lance]

Slightly high peak fps but in games like Halo and Rivals which love tight timings, it means very high 1% lows when people start bombarding the damn maps.

My worser tune in the past netted me about 10% higher fps in rivals and around 7% in Halo.

[steady lance]

Modded Minecraft LOVES fast ram.

Went from around 100fps average in my modpack to around ~160fps average.

[calm grail]

Oooh

[steady lance]

Well I am annoyed.

So since doing my timings I have been working on reducing voltages, well although the reduced VDDQ and VDDIO was stable in synthetic tests, it crashed in Rivals.

VDD 1.85v
VDDQ 1.45v
VDDIO 1.5v
Is both fully stable and frankly I am at my wits end with reducing voltages lmao.

[steady lance]

.

Decided I want to tune my CPU cores back to 5.8/6Ghz so I am gonna redo my RAM with bCLK at 6515 1:1.

[steady lance]

Decided to instead focus on 6460 1:1 since that gets me the additional 50mhz I need from bCLK.

[steady lance]

Got 6500cl26 1:1 at 2200.86 fCLK to work

[steady lance]

Trying to get GDM disabled.

[steady lance]

I am just gonna settle for 6500cl26 1:1 with fCLK at 2200.86.

GDM off refused to work and going any higher than 6500 had instabilities in one shape, form, or another.

[wooden vigil]

Who here

can like cook with laptop

OCing

[wooden vigil]

Over the weekends

Btw

[steady lance]

What laptop if it's Zen4/5 I might be able to help.

[wooden vigil]

Zen 4 pretty sure

Stuck at car rn

Check later

[wooden vigil]

Alr

I don’t have the specific model

But

Current performance is

Like

110 average fps

On cs2

1440p

Other games are unclear

[dull ginkgo]

Unfortunately 7535hs is Zen 3

It's a bit funky

[wooden vigil]

wait WTF

bro

well that explains

i didn't know stuff about PCs when i bought this laptop lmao

[dull ginkgo]

That's after the time when AMD started making their laptop chip names harder to understand, instead of x000 being the number you look at for the architecture, it's 00x0 that's the number that tells you the architecture

The 7 in front only tells you the year, not generation

[steady lance]

Hmm a bit outside of my knowledge.

Does your CPU support PBO and Curve Optimizer?

[wooden vigil]

curve yes PBO no

i think curve yes

wait

have to double check

[wooden vigil]

is it safe to overclock on asrock steel legend mbo ?

for 7700

[steady lance]

All ASRock boards are safe, just don't put in any risky voltages because the boards allow you to.

Also use HWinfo to monitor the voltages supplied and received.

[wooden vigil]

is

okay

i need a list of cards that peforme around a 4070 super

and i can OC it to a great amount

[proven canopy]

Country Cup 2025 has started. Do your part. https://cdn.discordapp.com/attachments/1428494010196033757/1433892134427295774/25001-CC25.jpg?ex=6909a35d&is=690851dd&hm=ec9cc5ed5422ffe027d7ba9f01705d7684e49954a35d893f513e203363c4f94e&

https://hwbot.org/competitions/Country Cup 2025
Let's get another USA win on the board

[radiant turret]

is anyone familiar with OC/ undervolt 7900XTX?

I was wanting to do this on adrenaline.

[steady lance]

Haven't tried it on Radeon cards personally but it should be like PBO.

Raise the clock, test in games and benchmarks like Port Royal I think it was called. Check if the score went up or down, visual artifacts, etc then rinse and repeat till you've maxed your clock. From there undervolt and use the same methodology.

[alpine mantle]

Looks like installing the custom contact frame back on correctly allowed the 13600k to push even lower voltages (used to run -0.095mv), will do some further testing later but so far its lookin good!

[steady lance]

Great to hear it!

I've been doing my CO values with my CPU in V-Cache mode. Current values with +200mhz are:

Negatives in mv

• 90
• 25 (currently testing this core)
• 10
• 10
• 10
• 10
• 10
• 10

[steady lance]

Pain! @alpine mantle had to lower core 1 to -20 because with high temp flucations core 1 was being a pita.

Not thorough testing but cores 2 and 3 were doing -40 to -50ish.

[alpine mantle]

is this Ryzen undervolting?

[steady lance]

Yeah, undervolting and overclocking my 9950x3D.

[alpine mantle]

how long you've been at it lol

[steady lance]

I applied a +200mhz overclock which each core can do. And then undervolting to keep temps low because to maintain 5.75ghz flat (flat being +5 seconds at a time) I need to have tempsp within 47c in single threaded loads.

[alpine mantle]

thats nuts

[steady lance]

Issue with undervolts is that stress tests don't catch idle freezes from low load.

[alpine mantle]

Curious on the max power draw on all core here

[alpine mantle]

What software you used for stress testing? C23, OCCT, or something like P95

[steady lance]

Well with 5.7-5.8/5.9-6ghz on each ccd from my prior OC in alaska I was pulling 290-330w in Cinebench. That was with around -30mv +200 and bCLK tunings though I also used a positive voltage offset of like +15mv.

[alpine mantle]

because on my end using OCCT found errors in my UV fast during stress testing and after fixing those issues I see no errors on idle either

[steady lance]

Karhu, OCCT, Games like Ark and War Thunder, and the hardest test of all the low load CrunchyRoll streams.

[alpine mantle]

Dude hit stock 14900k level of wattage

so this OC is gonna be your daily or just to hit max scores

[steady lance]

OCCT is good for load tests. Issue with undervolting on AM5 is when the cpu decides "Hey let's try doing say 5.8ghz effective clock for as long as we can on something that doesn't need it at all because our temps are low and then the temp spikes drastically causing it to error and freeze windows".

[steady lance]

Daily.

[alpine mantle]

oof thats finicky

[steady lance]

Lmao yes. Note that was in Alaska during the night with -12-14c ambient too.

CPU maxed around 74c.

[alpine mantle]

wow, if I had a ryzen chip I'd probably have a hard time finding a daily profile

crazy

[steady lance]

Undervolting is a pita on AM5.

Setting the boost clock is easy though.

@alpine mantle once my finances improve my end goal is to delid this 9950x3D and eventually look into a daily sub-ambient or just above the dew point solution which I have a few ideas.

[alpine mantle]

This kinda reminds me of Undervolting on my RDNA 3 card, an absolute pain to deal with because of its wacky boosting algorithm

[alpine mantle]

Dang, its gonna get chilly

😳

[steady lance]

My most appealing idea currently is getting a chiller from Alphacool to get the water just above the dew point and using it in combination with a de-humidifier to lower the dew point of the room.

[steady lance]

I wanna get my load temps down to the 30c range.

That with a little extra voltage should enable 5.75/5.95ghz all core clocks.

[alpine mantle]

my aircooler on the UV'd 13600k hits up to 80c

nuts...

[steady lance]

not bad for an air cooler.

[alpine mantle]

Thermalright doesn't disappoint

[steady lance]

Currently my game temps are in the 55-60c range with clocks doing 5.6-5.7ghz in gaming. In lighter tasks I see 5.75 pretty easily.

I am heavily temp limited.

[alpine mantle]

tbh with these high clocks were there a noticeable increase in performance with some titles?

Because there has to be a diminishing return

[steady lance]

Since I am bottlenecked I can't speak for highs but my lows have gotten drastically better.

[alpine mantle]

thats already a plus

getting 1% lows up is great

[steady lance]

Went from my lows being nearly half of my high to almost matching it.

I usually play with my fps capped to 120 or 144 and my lows are like 100 or 120 depending on the game.

I play Rivals, Halo Infinite, Ark Ascended, War Thunder.

[alpine mantle]

some of those do have a hit on cpu

[steady lance]

In very CPU heavy titles like Ark I see my CPU around 40% usage despite having a 3070.

So it's definitely working overtime there.

[alpine mantle]

what a whiplash, i forgot you had a 3070 despite it being shown on your username

[steady lance]

In lighter titles it's solidly in the 30% range.

[steady lance]

Lmao yep. I am saving up for a RTX 5080.

[alpine mantle]

[steady lance]

Good artist.

[alpine mantle]

no supers?

[steady lance]

Currently the Supers are either delayed or cancelled.

[alpine mantle]

well its a rumor really

OOF

5080 it is then

[steady lance]

Apparently it's because of the high demand for VRAM.

Also, I can get RTX 5080 cards at MSRP pretty regularly.

[alpine mantle]

🔥

[steady lance]

My plan is save up enough by Hanukkah to afford:

• Corsair SF1000 PSU
• RTX 5080
• Either a new case or a cheap Direct Die loop for my cpu or potentially some good ram.
  @alpine mantle

[alpine mantle]

ram is so expensive i wonder if the prices will drop during the holidays

[steady lance]

I wanna buy a 2x24, 2x32, or 2x48gb kit of 6000cl26 ram at 1.45v.

Capacity depends on if I can find a use for more ram.

[steady lance]

I should be able to run these kits at 6400cl22 with like 1.7v

[steady lance]

@alpine mantle my current kit is running this (it's 1:1)

1.85v VDD, 1.5v VDDQ, 1.5v VDDIO

[alpine mantle]

Nice kit, voltages high imo

[steady lance]

true. I went the direction of, okay how much voltage can my ram take while still being able to cool it? Then cranked the timings down as tight as I could.

Gotta love Hynix A for being able to take voltage like a champ.

You'd need to use over 2v and have bad cooling to kill these things.

[alpine mantle]

dayum

ill stick with 1.4v lol

[steady lance]

Are you on Hynix A?

I wanna get a good 2 dimm board with the ram mentioned above to try and cut my timings down by half or maybe run +8400 2:1.

[alpine mantle]

Forgot what die my ram is but it's something from CMXT (changXin Technologies)

It isn't great tbh, but managed to tweak it to hit 3600mhz CL16 @ 1:1

[steady lance]

Sounds like a brand rather than a die.

[steady lance]

Above average, I assume you are on AM4?

[alpine mantle]

well it is a kit from silicon power

I'm running LGA 1700

[steady lance]

Ahh I don't know how LGA1700 scales for ram.

Surprised your using D4 on that platform though.

I thought only 12th gen had D4 and D5 support?

[alpine mantle]

nope up to 14th gen tho

[steady lance]

Considering 13th and 14th gen can do high 6000mhz on D5 I am assuming you are severely board and or ram limited rather than IMC.

[alpine mantle]

running DDR4

[steady lance]

I am aware.

Did you try +4000mhz ram?

[alpine mantle]

I tried overclocking it to 4000mhz+ before but it sure didn't like it

and tbh ram tweaking is too much of a pain to deal with

its on a whole new level of patience