#overclocking
1 messages · Page 26 of 1
Installed the file, now what?
Reboot and try to run thaiphoon again
Which file
Didn't you say you installed it already
the file yes, did not execute it though
Then you didn't install it, you extracted it?
No, no. I downloaded it from the website. I have yet to touch it.
Then you definitely didn't install it
I was confused when you said this
Ahh my bad.
Now it's installed. gonna restart my pc.
actually, gonna update my gpu drivers while I am at it.
Haven't updated them in nearly a year.
back
How long does it usually take Thaiphoon Burner to read ram?
Less than a minute
Unless something is interfering by reading the same sensors, maybe hwinfo or something
Force close it and retry?
Sure
this is weird, I re-opened it but nothing is visible
Restart maybe
brb then
Any chance that with continued booting, the ram training time will decrease?....
Everytime I boot it's training ram for about 20s.
You could enable fast boot and it'll just use the last good trained values
Great, it's frozen again....
"DRAM Manufacturer:" is what you're looking for
at least it made it to 52% this time.
Make sure there's no other background programs running, any little thing can interfere
Discord is fine
I only have discord and file explorer open.
Even in the hidden tray
background programs would be nvidia geforce, razer synapse, steam
Gotcha. So remove razer before openoing thaiphoon on my next try?
I wouldn't mess with thaiphoon, much quicker to see what you have by testing what timings are stable.
Yeah it's nice when it works the first (or second) try but past that... might be better to explore other options
It just guesses based on spd
What do you think forks, Samsung?
idk, never used samsung
I thought I remembered something about tRCD not wanting to go below 39 on Hynix
I'd guess it's m-die, maybe a
I had clicked the first one right as you said that....
That would be slot 1 which has nothing in it
For some weird reason, Thaiphoon everytime I repeat this the % goes up. 35, 56%, 63%
Gotcha.
Maybe that's what my issue has been lmao.
restarting so I can try again cause it frozze
Back again, water delivery had arrived.
..... this is getting really annoying. Thaiphoon Burner keeps reading up to whatever % then freezing.
Probably better to find out the die another way then
When it works Thaiphoon is halfway ok at guessing but if it doesn't want to work then it won't
Yeah I think I am permanently done with Thaiphoon Burner, not once in a year of trying to get it to work has it worked once without issues.
Either way I am safe up to 1.5v on air so I am just gonna set that as my oc limit and depending on how the timings tune it will give good insight as to what ram it is yes?
Now question is, how does this cpu do in R24.
1.35v - 1.4v should be plenty tbh
But 1.5v would get you slightly tighter primary timings, which doesn't have much effect on DDR5
3 rounds of multicore in R24 good idea for how well the cpu is doing under load temp wise?
Or is 1 round enough?
Fair, though I am the type that likes to push it as far as I can for the fun of it until I hit dangerous boundaries when it comes to tech.
CCD2 is the x3D cores I am assuming?
You can tell by the clock speeds, the lower peak one is the 3D. You can always check in ryzen master.
Is the difference between the x3D and normal cores really only about 150mhz?
Cores 8-15 which I amm assuming is CCD2 runs about 150mhz faster on average peak
Try rule out a-die first. Set 7200c36 1.45v or something
On AMD that'll take setting uclk div 2 mode
I am on a 7950x3d...... I feel I may run into a different problem before that/
The board will default to it over a certain speed, probably.
Is 73.6c peak temperature on the hottest CCD good? CCD2 is 10c lower.
That's pretty good yeah
7600c36 or something xmp works plug and play on my 7950x3d and hero
.....
Just keeping it under 95C is a feat tbh
What kind of a golden sample do you have lmao?
LT720 push/pull P12s with TF7 paste really did it I guess.
I don't daily that system, just needed something to bench a 2080 super yesterday
Oh NVM that had the 7600x
But it worked on the x3d as well
I mean this is probably the simplest answer. I can just unscrew the heatspreader on my ram lmao.
The new versions of bios and such let you run much faster memory, 8000+ stable sometimes, but it has to desync uclk from mclk so it's not good for running daily
That said, I am eager to try out my system so I will do that later. Can the thermal paste on the heatspreader be reused? I assume it's uysing thermal pads.
Probably pads yeah so just don't tear them
It could be good for daily, depends what you do, but I wouldn't recommend it
Mostly adobe and I am getting into Blender.
6400+6800 is about the limit of ambient 1:1 depending on imc bin
Gotcha.
Hmmm, I feel tempted to bump up my ram speed real quick.
6400mhz a good start?
That wouldn't rule out samsung if that's your goal
Goal is to just bump up the clocks a lil before I relax and play some games.
If the goal is to find uclk limit on the imc then yeah
I will check my dimms directly later.
this is the main goal
gonna chuck 1.4v at it to start, currently at 1.35v
brb
Most ddr5 related voltages will stop scaling far before you hurt something
Just to make sure I understand, so DDR5 dimms will usually stop scaling long before the voltage gets high enough to hurt it?
where do I enable this?
6400 is pretty stable so far
Right click the program and run as administrator, it sets it automatically
Weird, I did do that.
Eh, doesn't matter much anyway. It just lets it use more memory per thread.
Also man, this 7950x3D IS AWESOME!
My frames have nearly doubled in most games and more in others.
and that's on higher settings too!
Considering the 5700g is basically a 3700X + iGPU, that sounds about right skipping 2 generations + 3D cache
xD, fair.
I'm holding out for the next gen to see how the performance bump is. My processor still works great for me, the things that hit it the most that I do don't need the 3D cache and prefer raw clocks.
8000 is supposedly 20-25% better than 7000, which would mean a standard 8000 chip will be almost as good as the 7000X3D ones. 8000X3D would go crazy.
It got pushed back a little, April-May
2024 or?
If 8000x3D releases you bet I will probably upgrade, xD.
gonna see if my cpu can handle 6600mhz brb.
Besides DRAM voltage is there anything else I can do to encourage handling of higher memory clocks?
SoC voltage will help with uclk and fclk
There tends to be a sweet spot for that somewhere around 1.25v, it's worth playing around with if you feel like it's on the edge of stability
Oh and VDDG CCD 1.05v
Jumpers and buttons are usually fastest, 5-10 seconds
VDDG and VDDP are good up to about 1.25v but you'll start reverse scaling before that
SoC is touchy, too low or too high and it's simply not stable when you're pushing it
Gotcha.
Is there a clear CMOS button on the X670-P that I missed by any chance?
I can't seem to find the clear CMOS pins on this mobo?
The jumper pins are between the RGB and the serial connection on the bottom left
Labeled CLRTC
My layout is different.
It's in the same spot for both the wifi and non wifi models of Prime X670-P in the manual
Thank you very much.
Touch the pin with a metal screwdriver while the PC is powered off right?
Press the power button a few times after touching pins?
Not necessary
Ahh okay.
It's not like removing the battery, the jumper directly drains the caps to ground
What do you mean? It turning red doesn't mean you can't, it's just warning you it might be higher than you mean to do, potentially dangerous on some memory kits.
Does MSI automatically set the uclk for me?
Yes
Oh neat.
No, no I mean if I put in 1.5v it reduces it to 1.43
Oh, I had to enable high voltage mode.
Just to make sure, for FCLK I should really just keep upping it? No ratios or anything?
Yep
It's permanently desynchronized from the memory clock so no latency penalty for going out of any ratio
Gotcha.
Watch for WHEA errors with higher fclk, it's the last thing on the list in hwinfo
Is being able to run 6400mhz on 7000 series good?
Gotcha.
It's good but not anything amazing
I think I may need to mess with VDDP and SOC for 6400+.
That or I try higher dram voltage first?
Is 1.35v to high for soc voltage?
Seems 1.3v is the max soc I can do, and it let's me get into windows with 6600, still crashes when load.
testing out different vddp voltages
Man, I am all the way up to 1.2v VDDP I really hope this is enough.
@sudden torrent any idea what else I could try to get 6600mhz to work without destncing?
1.2vddp has caused it to not wanna boot.
Don't go higher than 1.3, stay under 1.25 at most
Higher than 1.3 was known to literally explode am5 chips at the i/o die
Oh
6400 is good as it is
Will bring it back down to 1.25v so then
My mobo let's me set it up to 1.3
Felix you know anything else I could try?
Increasing dram voltage did not help as far stabilizing goes.
AM5 is very on or off for stability
i.e the settings will definitely work without a doubt, 99% guaranteed
Or you hit that slight 1% where it's a touch unstable
Or it's horribly unstable, bad settings, just won't work at all
This is my experience anyway
6600 is unlikely to work unless you have a lot of luck and great settings on your hands
For me 6400 worked without issue. 6600 is being a pain.
I can boot and do most things but when load hits then poof at 6600.
This is absolutely unstable
Just settle for 6400
Test it properly, and settle
Yeah fair. Also found out my PC boots with VDDP up to 1.15v
Anything past that it's hit or miss.
Should I do FCLK or cas next?
You can do it while on, but yeah
Did you figure out what IC you have?
Not yet, wasn't in the mood to take out my ram and unscrew the heatsink lmao. Will do that tomorrow.
I wouldn't do that, unsure how the heatsink is attached to your sticks, but yeah
Kingston Fury Renegade, it has 2 screws on each dimm.
So I am assuming I can just unscrew it.
Yeah, but whatever thermal tape/stuff between the heat spreaders and the ICs might not reassemble so well. I think you could figure out if it's a/m/b within 30 mins of testing
I am done with frequency, should I do cas or FCLK now?
Post a zentimings
Okay.
Loading up the profile for what got 6400mhz stable, I started with 1.4v.
I wonder if the default 1.35v would be enough for 6400mhz.
Booting into Windows ATM.
Sticking with 1:1
Unsure if the voltages there are reporting correctly, but you can't go wrong with 1.4/1.4 vdd/vddq for daily
What's the xmp of the fury renegade
7800?
Oh missed that lmao.
2x32 6000cl32
Odds are A, especially as to my knowledge they reserve A for renegade
Ahh okay.
You could test it by trying to boot like 7600 in complete desync
On a new enough agesa it might do it
Almost every chip can do 7600 boot on a-die, most can do 7800, some can do 8000, few can do 8000+
Sure, guess there's no harm in trying.
Yeah I've never tested desync so I wouldn't know
My x3d rig is currently apart, about to stick rom's 1600W in it
Just gotta boot my PC up first lmao.
The ram training time is horrific sometimes to where reboot is faster then waiting.
Guess that's the price I gotta pay for 2x32.
speedge https://i.imgur.com/DgGeAeG.png
Pic of the build btw.
The newer agesa is faster to train imo
I am on newest bios.
Need to clear my CMOS, 1.2v VDDP seemed to off messed up my bios or smth.
Nyoom
Booting into Windows and testing if 6400mhz is stable at the xmp voltage of 1.35 instead of 1.4
Xmp is 6000cl32 @1.35v
ycruncher vst, or just running 2.5b would be the first thing I try
Then memtest/tm5/etc overnight
Yeah I did TM5, VST, cycle 18 OCCT and one other y cruncher test
I went for a wide range rather than thorough coverage
Figure, might as well let it test while you 
Current settings. Only thing different from XMP is 2200 FCLK and 6400mhz instead of 6000
I would use y-cruncher 1B to gauge your performance gains. Fun to see the time go down as you tune.
Starting at trrds, I'd try:
4
8
16
8
16
auto
Will look into that later.
Easiest way to run it - https://benchmate.org/
7zip is also pretty good
@steady lance TBH, I'd start trying these timings a few at a time, test a bit, save bios profiles as checkpoints, then let it run tm5 overnight
You can use mlc gui or something like 3dmark to gauge your fclk performance btw
Hey! I finally got cl28 to boot.
Ahh ty.
Fclk seems even less black and white on AM5 than it did on AM4, it's hard to tell if you're using the best setting without testing
For mine it's 2133, even though 2200 is perfectly stable
Not really much point to pushing cas latency down just fyi
True but it doesn't hurt.
It benefits a couple of resulting timings, but cas latency itself does basically nothing
3dm11 physics is one of the best ways to see if you're getting fclk regression
Noted
3D Mark 11?
Someone put me onto mlc gui, so I ran 10 tests of each setting for an average per
Make sure you're testing to make sure you're not losing performance by dropping cas that low
I can lose performancd by dropping cash to much?
Yes, it sounds weird right?
Yeah. Also meant cas not cash lmao.
Just run ycruncher 1b through benchmate after each adjustment as a quick check
I'd have much bigger issues if I kept dropping my cash lmao.
Oh wow my ram sticks are really cold despite being 1.5v dram
Will up it to back to c28 then run tests tomorrow.
Aight.
It doesn’t benefit that much tho
Almost no performance improvement
One could say, a non-zero % improvement
If you count 0.2% as non zero
got these timings stable so far.
You're supposed to start with the tertiary timings, that's where your biggest gains are
Also tfaw is a good one to start with too, speeds up ram testing
SCL and tFAW yeah
I usually just go 4/8/16 to start, test with ycruncher 1b as a quick performance test
I see. Going to finish up primaries later then do that next.
For reference
Don’t bother with primaries, do those last
They barely improve performance at all
Start with secondaries, then tertiaries then finally primaries
Easy timings for AM5 https://m.youtube.com/watch?v=dlYxmRcdLVw
Text version: https://www.patreon.com/posts/low-effort-rank-77403831
My Patreon: https://www.patreon.com/buildzoid
Teespring: https://teespring.com/stores/actually-hardcore-overclocking
Bandcamp: https://machineforscreams.bandcamp.com/music
Twitch: https://www.twitch.tv/buildzoid
Twitter: https://twitter.com/Buildzoid1
Facebook: https://www.fa...
@proven canopy any idea if C5 post code is typically motherboard or cpu?
re-insert the dimms
Yeah no tried this a good dozen times now, still C5
I've managed to get it to train with only a dimm in slot 2 (2 slot board) but it would freeze or load super slow into the bios
I think the slot or cpu is buggered
My memory kits heating up to 56C at OCCT memory test
is this safe for ddr5 kit ?
What OCCT settings should I run for mem test?
Gonna do some ram oc so and I decided to start with tertiaries like you guys said. These are the tertiary timings right?
Yes, the first ones to do are the ones that end in "SCL" to 4 and tFAW to 16. That'll speed up your memory testing a LOT and net a decent overall performance gain.
Gotcha.
Doe tertiaries prefer any kind of ratio or?
Just set SCLs to 4. Time to see if it boots then if it's stable.
How do I restore a setting back to auto in MSI bios?
4 did not boot. Upped dram from 1.47 to 1.5 and changed SCLs to 6.
Just booted
SCL 6 is stable.
Welp SCL 4 did not boot but I at least got a dram light this timf.
Time.
Gonna try SCL 5 at 1.5v dram
SCL 5 does not boot at 1.5v
Moving onto Tfaw.
Hmm, low on time. Will tune tFAW tomorrow.
How much time should I wait on AM5 for boot from bios?
Loosen tCL to get SCLs to 4, it's worth it even on DDR4 and even more so on DDR5.
Primaries should be on auto, they're literally the last thing to touch on D5
Gotcha. Will try that later.
Right now I just want my PC to boot.
For some reason my SCL did not work on second tries and later.
Fal'Cie, what OCCT settings do you recommend I use for ram stabillity?
85-90% usage, the rest on auto/default, that's fine for a quick test
Gotcha.
What were the other programs you guys have been recommending to me for ram performance tests?
TM5 with the anta extreme profile loaded is good
I would go yc 1b/2.5/5b whatever the highest you can run, then tm5 overnight or memtestpro if you have it
Throw in linpack, yc vst/vt3 if you want
If you have a negative voltage CO set, test with corecycler
I believe that's what I have.
How do I confirm I have the profile loaded?
It says in the test window when you start
yc stands for Y-Cruncher right?
Yes
Down where it says "customize:" it says the loaded profile. You want the extreme anta one.
Run it as admin if you're not. If you are, then don't. That usually clears up the error.
Gonna reinstall TM5 first.
TM5 refuses to load the config?
It's not giving any errors now but when I load it, the config is not ABSOLUTE
When you hit "load config" and it closes, when you open it the config isn't the one you selected? Is that what you're saying?
Yeah.
tFAW 16 is stable.
You can pretty much follow these timings. You have dual rank memory so the things he says "only relevant for dual rank" you'll want to tune too. The old fashioned "trial and error" way.
https://www.patreon.com/posts/low-effort-rank-77403831
Oh and if you're running at 6400 you'll want tRFC to be a bit higher than 500 probably, tRFC 2 and 4 don't matter since AMD doesn't use them.
Gotcha.
Yeah I am running 6400.
Is turning fast boot on a bad idea if I am overclocking ram?
Cause the boot times are horrendous.
Some say memory_context_restore off as well but I haven't had the longevity with am5 to know if that tracks myself
Again, fact of life for AM5
Well in any case, I am trying to tune TRDRDSC rn.
Call this bad advice if you will but for AM5 I really just shotgun almost all the numbers in one sitting
It either works well, works terribly, or fails when the dimms are too hot afaik
It's pretty black or white in my experience
(ddr3 corsair dominator ram fan ftw)
I nearly bought one of those
Years ago
I have an extra if you want it
Also idk why the guide says to set TRDRDSC/TWRWRSC to 1 when it's 1 by default lmao.
nvm you live in AU
I'm good lol
Some boards will train it to 1, some won't
Oh really?
Depends on dimms, IC's, bios etc
Mine has it listed as 1 by default.
Motherboard training algorithms are created by people
So, ymmv
It's better to list the change in the guide no?
Otherwise you'd have no idea
do you know what DIMMs you have? in terms of IC anyway @steady lance
Does upping CPU SOC voltage help with tuning tRDRDSD?
Likely Hynix. 6000c32/38/38/80
Higher cpu pops
Yeah not DRAM lmao.
Does upping VDDP voltage help with tertiary timings by any chance?
Cause I can't get tRDRDSD lower for the life of me so far.
Not really
You can really push dram vdd/vddq together to see what results you might get
But pushing beyond like, 1.4v? You'll get someone low returns imo
I am running 1.5v dram ATM.
That's pretty high for 2x32
Be a warm dimm
If it was M that would throttle in stress
Warm but not hot after multiple TM5 runs.
Gotcha.
You'd know if it was M by now
Haha well that answers things.
Daily up to like, 1.6v is fine
Before you want a fan
But you should already see marginal scaling by now for little changes
Good to know. I am gonna try 1.6v and if tRDRDSD still doesn't go lower then I give the f up.
Cause this is like my 7th try now without boot.
If it's not working for terts I doubt upping vdd/q would do much if anything
Ddr5 is like
You set terts first
And never go back
I love ATX boards now lmao. The CMOS pin is so easy to reach compared to my old board.
Yeah if it goes from perfectly fine to not booting at all
I'd say that number just won't work
I wonder if I can get SCL 3 to boot with 1.6v dram.
I'd maybe even step it back an extra magnitude to make sure it's stable
I really doubt lol
Worth a try?
Timings I'd worry about would be primaries and trfc, maybe trc, tcwl
The rest should be set and forget
Gotcha.
If it won't boot with a secondary it probably never will (under reasonable circumstances)
Same for terts
Intel it's sort of the same but it takes a lot more patience with voltages for D5
I am a try SCL 3 at 1.6v dram for the fun of it. I give up on tRDRDSD/WR
Seriously....?
Cause I can't tell you how many times I had to clear CMOS by this point...
Seeing the CPU and RAM LEDs turn on Everytime I boot my PC scares the life out of me haha.
Yeah imo just avoid pushing the envelope
AM5 is very black and white
Unlike intel which is a voltage and stability puzzle with D5
Ohh this is interesting not in windows yet but I got to ram training.
Btw my dimms seem to hover around 50c when under load is that good or bad at 1.5v?
I am more worried about during After Effects lmao.
My D4 kit went through hell whenever I did Adobe AE.
How long should I wait before I consider the ram training a failure?
This bend in the bottom fans bracket is going to drive me insane.
It just occured to me I have been trying ram settings meant for a 2x16gb kit at 6000 on a 2x32gb kit running at 6400... No wonder they aren't working the same.
got tFaw to boot at 14
Decided to tune tFaw some more. Upped DRAM to 1.6 till I can't reduce timings anymore then I will reduce dram till unstable.
Does it hurt to have it lower though?
Really? That seems fishy to me xD.
I didn't make that rule ask anyone else lol
Well tFaw 14 is stable so I guess I will just leave it at that if it doesn't hurt me. My dimms are averaging around 58c during load in TM5 at 1.6v.
Up to around 62c peak so far
What about tCWL 🧐
Reliable enough
You can factor in a trend from a relative reading
Dimms peaked at about 65-66c.
Just finished the TM5 test.
Is it safe to say that this is Hynix A die then?
Heading to bed. Night guys.
FAW stands for Four Active Window. It's the minimum number of clock cycles you can put 4 "row active" commands in. That command happens to be controlled by the SCL timings, which are set at 4. 4x4=16, so that's the lowest tFAW can currently go. You can set it lower, but it will still take 16 cycles for 4 activate commands.
Gotcha. Thank you for the explanation.
Seems my ram at 1.5v 6400mhz, tFAW 16 and SCLs set to 4 is semi stable in my most ram demanding game ark.
Definitely not a heat issue cause my ram was not hot at all, just slightly warm.
You can check dimm temps in hwinfo.
I think most measure from the pmic, unsure
High temps just mean lower stable refi mostly
My 4x 32 DR a-die is unstable at 60c 65k refi, but stable at 90c+ 32k with a hair dryer on them
Lmao, an actual hair dryer or a active ram fan.
The build was going into a server for a customer, needed to test max temp stability.
Did you actually use a hair dryer?
Seems tFAW 16 is stable.
haven't had any crashes in Ark:SE.
Then again the crash could've just been cause of the game lmao.
Yes, it worked well for testing.
It's a good idea to run furmark while you do the ram test to check things like trefi
GPU ramped up under the fans isn't as dramatic as a hair dryer but does dump heat over the dimms all the same
I turned SCL up to 5. Now my ram has stopped causing sudden crashes.
This is highly likely in my experience lol
AM5 is black and white I keep saying this lol
Lmao especially considering the game I was playing is Ark.
Yeah so far it seems SCL 5 is perfectly stable.
What does upping VDDP voltage affect for the ram?
VDDP is one of the fastest ways to make your memory unstable
Gotcha, so don't mess with that.
What timings would upping SOC voltage affect?
None in particular. That's for frequency mostly.
Tuning tWTRL.
32 is XMP, 28 booted. gonna keep tuning in groupings of 4 till unstable then increase by 2.
My current timings.
Lmao.
tWTRL 28 is stable
Can upping Dram voltage to 1.5v cause certain timings to be unstable?
I see you got some liquid nitrogen cooling there
Running a round of TM5 to see if tWTRL 26 is stable.
tWTRL 26 is stable, 24 unfortunately gave BSOD.
I finally figured out what was causing the occasional ccrash lmao.
At least I think so.
Seems it's SCL.
Fal'cie do you by chance know which timings?
I always tune my tertiarys and secondary so this question is kind of irrelevant but I think this servers better then the internet in terms of actual answers
I've gotten 6200 working but what's the best xmp you could probably get on a 12th gen ddr5 controller varying between 6400-6800 at like 1.4 volts?
Depends more on the motherboard than the controller
And yeah we've got several of the top hwbot overclockers advising us here
Still on a 4 slotter BUT Hi falcie! Good to see you're still moderating
I kinda disappeared for...a year? Or so but you've always given good advice and it's just generally good to see ya in chat
It feels like coming back to animal crossing and the villagers forgoring
Depends on IC and board, but in the right circumstances the 12th gen ddr5 controller is very quick
Well I have samsung B-die Rev 4.1 in a 4x16gb (I know i shouldnt be using 4x16) and a z690 aorus xtreme
5s16b?
Uh
You'd probably be lucky to get 6200 to even make windows
On that combo
Z690 gigabyte has arguably woeful ddr5 overclocking
The boards were lucky to get anything over 6000 in some cases
And 4x16gb ram can be more problematic for stability over 2x32gb
I have 6200mhz working so you'd say that's peak?
I would say that's really good yes
5S16B peaks around like, 6600, 6800 daily to my knowledge (it may be a little better nowadays) but you have 4 sticks, that's pretty impressive
While a little sad I guess that's a pretty good answer
I would also check you aren't slamming the hell out of your cpu based on your voltages
If you have crazy high imc volts like vdd2 well north of 1.35V, I'd consider toning it down
I believe cpu vdd2 is the one to worry about
It might be cpu vddq, been a minute, need a refresher
I'm a voltage coward I don't put anything above 1.4
People use 1.45 and I cannot bring myself to lmao
There have been observable cases for intel 10nm to have degradation over 1.35
That's good info there
As well as ICCmax should stay under 200A iirc typically
It's been a long journey and a lot of scamming off of buildzoids more extreme settings
Since I've been out of the space a while I hear things every now and then, I'd definitely double check these things
Honest truth:
There are observable cases with raptor lake and refresh that too much multithread torture can lead to degradation
I haven't seen alder lake, but wouldn't count it out
It's all 10nm
From what I hear 12th is a bit under so I like to keep my expectations low
Under?
I have undervolted P cores and overclocked e cores it's funny because it's 4.9ghz all p core 4.0ghz all e core and about 28256 cinebench r23
My config is goofy because I don't believe in over volting for .1ghz
We are at like llc low and .08 offset it's prolly like 1.28v and I just leave performance on the table
Ye
I've been running this since Ig all of 23?
So you're probably fine
When I say over 1.35 I say in the sense of load voltage at full
I am just an absolute nut and can't leave settings alone
I guess optimally if I could afford to shill out swapping from xtreme to tachyon and 4x16 to 2x32 would be best but I just don't want to
If it ain't broke don't fix it
I wouldn't really bother
Samsung really good from testing I should say as these are my 5600mhz 1.25 volt set from ddr5 release
If I didn't mention
Honestly unless you went full ham, made a reasonable exchange to something like a z790 white apex or a unify or edge itx
Or maybe gigabyte z790x
2x32 or 2x16 Hynix A
You wouldn't see meaningful improvement
And I say that after sunk cost already implying it wouldn't be lol
Cos even tachyon won't do that well
I've heard that apex and the evga two slotter do best I guess I said tachyon more as a 1:1
They do ye
If you want best eatx, probs z790 kp
Best overall, white apex
Best gigabyte? Z790X
Z790X they finally kinda figured it out
Best bang for the buck would be the itx msi z boards
Grab one of them used you're definitely underpaying for close to peak
As I said Definitely glad I came here memory reviews online are just sales pitches and compliments
A lot of oc places will just tell you to get the best
Or recc some stupidly expensive kit if it's not t-create 5600 CL46's
There are a lot of ways to skin a cat, but I would also objectively weigh up the benefits
As I said its kinda lazy for me but kinda not
You would see a significant double digit improvement with these changes (and a newer cpu) but it's still widely exaggerated benefits, the value really doesn't exist
I'm no frame chasers for sure
Like I'd still say a 12900K is perfectly viable
It may not run every gpu at 1440p, but it's more than ample for everything else otherwise
There's one avenue left untouched I'll ask you about
Sure
Alder/raptor have a pretty infamous flaw I'm sure you know and that would be the ring/cache clock with e cores enabled
Oh wow didnt know they fixed that for raptor anyway I can take it from 36 to 41-43 without too much testing at the upper limits
Should i?
Okay so I already leave it at stock 36 the 41-43 was purely just seeing but you're saying I could leave it at 41? Most likely but obviously stress it?
I would ye
It's not a huge gain but a gain nonetheless
I just would always check and double check
You seem to get how I do so ill trust that
Sure some people like to disables stuff for peak p core but I just want my whole chip overall better
I'm the kind of person that rather do a whole lot of new work to check my work rather than find out later and reinstall my windows

I hate redoing things
R24 was my bane
Ye
R24 forced me to re evaluate stable settings that work with 1000% coverage
Oh I thought you meant R23
But im also on windows 11 and I've been told it's scheduler goes schizo with stress tests
I wouldn't use cinebench for stability
Anyway I'll bbl
Good chat mate
Hit us up if you got more questions
I gotta run
Should i undervolt my 6950 using amd adrenalin? Or do you recommend other software?
If you recommend afterburner over that one i can do that, if it's additional software to one's i don't have its one of the twos
Personally I use MSI Afterburner. Pretty easy to use and customizable.
Use morepowertool
MPT stonks
RDNA 3 no mpt FAT L
I fixed the issue!
It was my llc
I was so confused why lowering my undervolt wasn't fixing my stability but it never was that it was my vdroop
Going from Normal to low brought me from 1.174 to 1.224
I can now run the stuff I thought was just bad scheduling crashes
Which I never actually set by the way
That's actually the gigabyte z690 auto
Just straight maximum vdroop?
But like easy win there I'll take the +50mv over setting an avx offset
fr? wth
Yeah it also just froze at one step above auto so I'm keeping it at low which is two steps above auto

What's fantastic is when a higher llc level makes the voltage RISE instead of droop
(Proceeds to set it to ultraextreme)
Here are my current ram timings, what should I tune next?
I've tuned, tFAW, SCLs, tWTR L/S, and Frequency to 6400.
rrds to 4/8/16
Are you referring to RDRDSD and RDRDSC?
I got tRRD S/L tuned to 6/8
What should I tune next? Also, should I be concerned if HWinfo says that my PMIC temperture exceeded threshold? My Ram hovers around 57-62c under load according to it.
tRDRD = 8
tWRWR = 16
is that good?
How do I figure out what to set tREFI to?
I've finished all secondaries and tertiaries.
you can max that out if thermals support it, i just set mine to like 50000 because the gains after that are pretty small with lots of heat gain
At what thermals should I stop?
i dont remember I havnt touched ram and while sorry
Ahh okay.
Currently my tREFI by default is around 12000 or smth last I checked, it always changes when I reboot since it's on auto.
just set it to 50k and forget it should be fine
thats what actually hardcore oc said to do and he seems smart
Worth a quick try then.
To my surprise, my mobo actually sets most timings near it's limit by default pretty accurately.
Been using it for references and usually it's spot on
Meanwhile me with my ram idling at 45
DDR5? that seems way to low.
Most people don't push their ddr5 past 1.35v so that's about right with a good heat spreader
"Ideally" which isn't always "realistically"
You must have good airflow
That's under heavy sustained load?
Mine hover around 55-62 under heavy sustained load at 1.5v.
Yes
Albeit my fans are pretty low RPM wise, could probably turn it up to help since I am using a Fractal Torrent.
What die are you using?
I likely have Hynix A die, wondering if it'd be fine if I pushed the voltage higher.
The problem is there's no fan near the memory to blow air between the sticks
A die
I cranked up my fans from 20% to 100%. According to HWinfo after a 8HR OCCT run 85% auto, my ram was around 45c average.
Any tips on how to install a 120/140mm fan for the ram in a fractal torrent without it looking horrible?
I hung an 80mm or 60mm from the grille above the memory
imma attempt to undervolt igpu voltage for no reason later
Is it fine if I pushed my ram voltage to like 1.6-1.65v range?
Actually, on second thought would the returns even be worth it in the slightest?
I dont think 1.65 is safe for daily, I don't know about temporary
I've also never tuned ddr5 so take what I say with a grain of salt
Gotcha.
I will keep it at 1.5v for the time being then.
Should I do FCLK or Primaries next?
I already did secondaries and tertiaries.
It's fine smol risk
Safe is up to around 1.55 ish
Have fun

I upped voltage to 1.55v DRAM. Tuning tREFI, is there any specific formula I should use or just keep upping it till instability/artifacting?
Decided to up my dram to 1.6v and see how much tighter I can push my timings since my thermals are presonable now with the fans set to a higher rpm
btw MSI sets my DRAM, VDDQ, and VDDIO voltages to the same, so is it a concern if the other voltages are up to 1.6?
scl 4 is now stable!
just installed a fk120 fan incase of high temps.
directly above my ram blowing tons of air down onto it.
temps are looking quite good. and so far, timings are dropping pretty nicely.
Dang this fan is powerful. Makes the 140mm fans I have from Fractal and my P14s/12s feel like a joke. If only it had rgb.
I am so far reaping some decent gains from going to 1.6v.
Dropped tRDRDSC timings by 3 so far, and SCL dropped by two and stable.
By installing the fan my ram temps despite being upped from 1.5v to 1.6v, has dropped about 20c on average during load to the low 40s range from mid-high 60s.
Out of curiosity, may I ask what setup you're running?
Thanks for the idea. Worked out nicely.
My current settings (passed TM5), gonna run it overnight on OCCT to finalize stabillity.
Part List - Intel Core i5-13600K, GeForce RTX 3060 Ti LHR, Phanteks Eclipse P360A ATX Mid Tower
The ram is at 3600c16, but i haven't actually tuned it. Just spent like 10 min getting that to work with some settings sips told me
I see.
Idt I will actually tune it unless I go to something similar in perf to an xtx
for some reason I am getting t he odd feeling that I actually have Hynix M die.
I do plan on doing another build soon, I've been thinking of taking the mobo/ram for that and going to a d5 mobo
timings seem to scale a little too well with voltage.
That would be ideal
true.
cause my ram for the most part is doing single rank Hynix A die timings pretty well at similar DRAM voltage.
Well I am a go to bed now, nice talking to you, Bacca.
Neat, D5 has definitely been more fun to tune then D4 in my experience.
Actually slight issue, my mobo is atx and I wanted to do at most an matx mid tower build
Was just mapping it out and realized
I could do a 7600 itx build
I'm either gonna do budget with like a 12100f or smthn along those lines, or 7600
you'd know if you did because 1.6V would DEFINITELY need active cooling
It's A die
1000%
My M die can overheat at like, 1.45V lmfao
well not overheat but spew a buttload of errors at like, 80C+
I think you need to loosen something, my tune is quite similar but I'm only on 1.4V
I've got a fk120 fan blowing directly down onto it.
I see. Yeah probably A die then.
This is with active cooling?
no
My ram was averaging 55-62c under load at 1.5v (prior to having active cooler) is that overheating?
.
Also, is there anything I can do besides opening up my ram to figure out if it's A or M die? Like apply certain settings or smth in bios.
@kind walrus, should I be worried about VDDQ and VDDIO voltage also being 1.6v?
Seems your VDDQ/IO is also synced?
After I tighten up my timings at 1.6v, how should I tune tRFC?
My personal opinion is that's a lot of voltage for 6400
this kit is m die
if you were running something really tight like cl28 at 6400 I'd understand 1.6v
not a lot of point to it tho, frequency is what counts
I haven't tuned primaries yet
I know my ram is capable of cl28 though.
Primary timings have next to no effect on D5 anyway
It's not worth increasing voltage for them
pretty much
if you're struggling to stabilise with 1.6v I would also say good luck with lower primaries
I only upped to 1.6v to see how far I can push my ram tbh. previous timings listed were stable at 1.5v.
Why is that?
That's a question for the engineers
I'm not one, so I only know what the buttons do and not why they do.
I assume that it's related to how each stick has two 32 bit channels instead of the single 64 bit on DDR4
Cas latency itself offers basically little to no improvement, it's only the resulting timings that are influenced by it
like TCWL and TWRpre
that have actual gains
and those aren't that big either
tWRPRE I believe isn't even on AMD anyway iirc
or not exposed
this is even more so on DDR5 btw
Gotcha.
Managed to drop tWRRD from 8 down to 3.
Lmao, HWinfo thinks my ram is in the negatives for a moment
You guys think my ram will survive at least two years with this voltage?
don't really know tbh
we've only had ddr5 out in the wild for like less than 2 to begin with
somewhere near 2 years
I see.
Btw I just got tRDWR 14 to boot, gonna try 12 now.
Is there any point for me to tune tRDRDDD/tWRWRDD?
The guide Fal sent says that to leave on auto if I am running multi-rank but did not explain why.
tRDWR 12 did not boot, trying 13.
Hey it booted!
Bloody hell?!
Each D5 dimm is drawing 30-35w watts of power.
Is that normal?
Thank gosh lmao.
That's better
Lmao, that scared the life out of me.
That's exactly how much I'd expect
I was like how the heck does ram draw 30+ watts
Btw tRDWR 13 seems to be stable.
With that, my tertiaries have been fine tuned.
Now back to re-tuning secondaries.
Btw my ram is averaging about 50c under sustained load.
If I turn my ram fan from 80% to 100% that drops to 45c.
At 1.6?
Yes.
I swear that's A die
Is M die really that much hotter?
Should I do tREFI or secondaries next?
Trefi do 64k or 32k or don't bother
Nothing in between?
Arguably one of the more important settings
Eh, 50K?
You want that ideally as high as possible
Is there a formula or is it like other timings where you just keep upping/lowering it till unstable.
If you need to halve it then you want other things loose imo
Not to my knowledge, and lowering one by one you'd be dead before you finish
💀
True. So 10k at a time then drop by 5k?
I tried 50k earlier back at 1.5v which caused some visual artifacting. Should I try it again now that I am at 1.6v?
I'd just try max trefi
64k
50k
32k
I wouldn't bother with anything in between
You'd have an impressive level of patience to truly root every number out
xD I have been at this for weeks lmao.
I spent just over a day on mine
I hear there might be a formula for calculating tRFC or smth like that, is it true or false?
Yes
There is
But on D5 not a lot of point to it
You get some gains the lower you go but you do add instability very quickly
Could you tell me the formula in case I wanna try it?
I like to run trfc quite loose, so my ram doesn't care if it's 80c
Can't remember
Ahh okay.
Should I just set it to auto for a baseline then keep turning it down till unstable?
64k tREFI booted.
So far no artifacts
My board has the tREFI up to like 65.6k if 64k is stable should I just max it?
Oh dang, with the tertiaries and tREFI tuned in, my DRAM read and write bandwidth both went up by 3-4Gbps.
You could max it
If you wanted to
If you auto it then boards will train like
1.6K...
3K...
Useless numbers
Yeah, my board trained it to 12.6k.
Wow, tREFI really made a difference in my average read bandwidth and a nice buff to my write bandwidth. Is that normal?
Write went up by about 3Gbps and Read went up by 4.5Gbps.
Yes that can happen because it doesn't spend nearly as many cycles refreshing data
More time for other commands
Gotcha.
What DDR5 heatsinks would you guys recommend? I plan to do a lot of D5 OC for the foreseeable future and I figured it woulf be nice to have a good set of heatsinks to use across the dimms I buy.
I got 65k tREFI to boot, for whatever reason 65,535 refused to boot.
I would just strap a good fan on top of them imo
I got a nice fan on them already, think it's a FK120 directly above zip-tied to the grill in my Torrent.
Was just thinking that it'd be nice to have a visually nice and very capable heatsink that I could transfer over as I get more kits.
Ah
the FK120 is somehow pushing more air and static pressure then my fractal torrent's 140mm both at 100% by a large margin
albeit is definitely louder though
The case fans have resistance from the filters
Comparing it to my exhaust filter when I have the side panel off.
resistance is definitely noticable on my rad.
Ek monarchs if you want something to assist but really the fan should be enough
loop the ram
How hard would that be?
I honestly considered the idea of a 140mm rad in the back.
Occured to me that my CPU has an iGPU. Will disabling it help me with my ram oc?
Generally no unless you're doing some crazy XOC stuff
Extreme Overclocking. LN2, LHe, even dry ice. Anything subzero cooling I'd say qualifies.
Ahh gotcha.
Glad you tol;d me.
How hard would doing a 140mm ram loop be?
And would there be any noticable benefit over just getting a good pair of passive copper heatsinks?
Real world at reasonable daily voltages, there won't be any difference between a fan over the ram and direct liquid cooling
If you were trying to push it as hard as you could at risk of breaking it and going for 1.9v 12000 c30 or something then I could see looping it
Gotcha so no point in doing a loop.
Any good passive copper heatsinks you guys would recommend?
Also, Fal do you know the formula for tuning tRFC?
What do you mean by "formula"? Like how many cycles it spends refreshing per second?
I read somewhere that tRFC is set based on like one timing * another timing or smth like that.
It did not specify which timings.
There's really no point
Aren't the EK Monarchs all water-loops?
No
I went to their site and I did not see any passive.
You can just buy the sinks
They have the option to be looped
But you can just get the sinks and they work
Oh okay. So they work as passive and have the option for looping?
Mhm
JESUS?! 140$?
...... the website had them listed at like $40.
that's why...
On Amazon, someone is reselling them for 3x the price.
DDR5 sticks that I got from Kingston.
Kingston fury? Or renegade?
The tRFC number is related to frequency vs nanoseconds.
As a common example, B-die DDR4 is known to be easily stable around 140ns. Lets say you're running the memory at 5200c22 and want 140ns. You'd set tRFC to 364. This may help https://gcdr.net/ramcalc.html
My plan is to get a nice pair of heatsinks I carry over to all my ram in the future.
Fury.
thanks, would this work the same for D5?
And just a warning, if you intend to delid ram every time you get it, more power to you, but it carries the risk all the same
Same formula at least since it's time based
Hair dryer
I got one that gets pretty hot.
Until warm enough that the heatsinks peel off gently with the glue soft
But not too hot
As you might cook the chips
So I put in my first cas latency and my frequency then put in whatever tRFC value it gives me?
Gotcha. So warm it up and slowly test if the heatsinks peel off as I go.
Frequency, CAS, then either change tRFC (ns) or tRFC1 value and you'll get the other value automatically
CAS shouldn't play into it iirc
It's like
Ns = expected freq/2000 or something
It's been a while but I definitely don't remember using cas
It's in the form that's the only reason for it
Use zenTimings for my tRFC input value?

