#overclocking
1 messages · Page 67 of 1
@proven canopy do you know why my cpu is only using 1.28v when I have it set to 1.33 in bios
I don’t know if there’s a setting somewhere messing me up
You want some amount of voltage droop under load
Read up on load line calibration
Should be in the wiki, check the pins
I have it set to 4, the one asus recommends for oc but I’ve tried 2 as well
You want the setting that let's you run the lowest stable load voltage
Have to play with it and test to see
With 4 I can get almost all the way through a pass of cinebench r23, and voltage fluctuates between 1.279 and 1.284 but has blue screened right before the end
Which mobo?
If you have 7 or 8 levels, 5 is generally best, if you have only 5, then 3 is best
Strix/tuf have 7, maximus/crosshair have 8 and some primes have 5
@mint mountain listen to Fitz XD
I think it’s 7
So I’ll try 5
And it’s z590m plus
Prime
Maybe
Idk
The number is mostly just based on what you have
Run a higher llc if you expect more demanding loads
Well even with higher llc still won’t go above 1.284v in cinebench and blue screens after 15 seconds of running
Sorry I was sleeping. Got all those pings, lol.
It was still only 18-22-22-42. The guide I'm following basically says to tighten everything except the primaries first, then move onto the primaries.
I was testing 17-22-22-42 on 3733 last night. Seemed okay. 3733 with 16-22-22-42 crashed after about an hour I think.
but you arent able to get cl16 3600mhz?
I haven't gotten there yet. I figured I'd see what I could get on 3733, then try it all again with 3600 Mhz and then benchmark between the two.
try 3600cl16
3600cl16 is better than 3800cl18
I literally spent, like, 11 hours yesterday overclocking. I hate how long memory testing can take.
I've been doing 5-10 minutes between changes, while writing them down. If I hit errors in the 5-10 minutes, I'll roll back a change and then test longer.
yee its not the best
I'll keep playing with it. I eventually couldn't even get 3800 Mhz to work, so I'm gonna see what 3733 likes today.
And I feel like 3800 only didn't work because of the fabric on my chip. It previously (while decoupled a long time ago) hated any fabric above 1866
see if 3733 is crashing at cl 16 and cl 17 there is 0 point in bothering with 3733mhz
try checking 3600cl16
It definitely crashed about an hour into CL16, but CL17 was fine when I tested (only for 10 minutes though).
ur ram should be capaable of that
for 3733*
I see now. I just saw the Wikipedia chart for it.
I'll just go to 3600 then. I won't start at CL16 though. The guide I have recommends quite a few secondary timings first, then primaries
did 16 -20 - 20 - 20 - 40 work?
primaries are called primary for a reason.
you ALWAYS do primaries first
then secondaries
and then tertiary
what guide is this?
I linked it a while ago. One sec
I liked this guide because it offered a lot of granular details (like how to calculate tRFC and stuff). I found it as part of a collection of other guides on the overclocking subreddit IIRC.
strange that guide is very good and trusted
primaries still come first tho
also set the procodt to 36.9 or so
I'll try to find the original list of guides and cross reference them. I don't not trust you guys, but I figured I'd try to learn more about how DRAM works and common methodologies than just try a bunch of numbers and hope it works.
i see what you are gonna do for the next few weeks in that case.
lol I don't mind if it takes a while to learn this stuff. I originally just wanted to boost my CPU a bit, but after seeing the 2% gain from just a light 3333 Mhz overclock, I'm tempted to finally do a full dive into learning how memory works
as with every part in yur pc, the more you OC the more diminishing the returns get.
I would personally stick with a 3600cl16 profile as the basis to start
[Slightly Revised 1/20/2019] Set VDIMM and VSOC to baseline values. Set frequency to a baseline value. Primary timings: initial tightening and gauging baseline timings tCL, tRCD, tRP Set tCL, tRCD, tRP, tRAS somewhat loose. Unstable: Repeat until stable: tCL + 1, tRCD + 1, tRP + 1, tRAS + 2 Rep...
thats a decent guide to streamline your process.
I'll keep that guide in mind, but it doesn't seem to do a very good job at explaining how many of the values should be changed.
always by 1
there isnt another way
unless you have an electron microscope to observe the tiny imperfections in silicon
Then give it more voltage
But 1 what though? For example, it mentions VDIMM and VSOC. But I know for sure that 2.0V for SOC will kill the chip. 1.1V? Okay, that makes sense, but then what about tRFC that can change across 100+ (for example, from 666 to 485 in one of my tests).
It also says to establish a VDIMM and VSOC baseline. What is that baseline? Based on the example values given in other guide I linked, that could be many different values depending on the ram and chip.
that's terrible at "guiding" you what to do, if someone's asking for a guide
It’s set to 1.35
I'll keep that one in mind, but again, I'll probably cross reference multiple guides, as well as try to learn how memory actually physically works in the mean time.
But not using anymore voltage
then go 1.4 🤷♂️
Ok
Wait, it's set to 1.35V but droops to 1.28?
integral's is the best for new overclockers
I can get 1.45 to drop to high 1.2, with the equivalent of level 5, just has to do with current
Oh I see now. They said raising their LLC didn't do much for raising the voltage under load.
I mean, the highest level LLC should be keeping the voltage pretty steady at the set voltage.
Though I haven't done Intel overclocking in a long time, so I have no idea if it's wildly different than AMD when it comes to voltage and LLC.
LLC 2 for me on X570 droops about 20-30 mV, LLC 1 overvolts by about 5-10 mV and stays there. Not sure if Z590 or the specific MB brand does the same.
who wrote that google doc lol
higher llc will keep voltage monitoring software showing steady voltage, but transients will be just as bad, if not worse than a lower llc
here's 1.45v set in bios on the equivalent of level 5 llc (gigabyte high)
that's just r23
Like guy was asking how to do it
Thats just steps
Set VDIMM and VSOC to baseline values.
Set frequency to a baseline value.
Primary timings: initial tightening and gauging baseline timing
first 3 steps give you literally no help
True
in step 5, why would you even touch vcore?
the rest of the "guide" is lower X timing by 1 and test
"tWR
If lower than recommended value, set tWR = tRAS - tRCD
Repeat until unstable or tWR = 8:
tWR - 1"
is just flat out wrong
Fr?
Its what the other guide links for amd platform
Like that guide links to a reddit post explaining stuff about amd ram oc
Which points to the google doc
I'm not sure why that Google doc was linked in that reddit post. It's pretty awful.
Maybe it's good from the perspective of seasoned overclockers that have more of a feel for these things.
I decided to kinda just throw together some values and see if it works. Mainly going a little bit above the recommended values (so a little loose on everything). I figure that if it was running at 3733 and 3800, it should be able to do generally loose values on 3600.
I also decided to do 16-20-20-40 at the recommendation of Craipop
Since I figure CL16 should be achievable if CL17 was fine on 3733, and CL16 lasted over an hour on 3733
So, my windows is corrupted 😂
16-18-18-38 tRC 40 wasn't very happy
I should practice running sfc /scannow more often
yay it recovered
OC @4.4GHz for 1.2V, successful OCCT 1h Linpack and 30m CPU Large-Extreme tests \o/
On air cooling solution.
It crashed in like 5 minutes XD
But OH, I think I messed up pretty bad. Apparently tRC is supposed to be tRP + tRAS... which in this case is 60... but I have it set at 44 😂
Jk, I do not either.
Many of the RAM values are based off each other, so you can do math based on some values to determine what other values are supposed to be
Neither did I understand this
I basically set a value WAY, WAY lower than it should be, and is probably why my system corrupted a little bit ago
Are you doing smaller FFTs? I think that's the most intensive one
Before it crashes, do any of the threads fail to calculate some values?
A thread will stop if the value is too far outside a range
I see
Hmm, you might not be giving it enough voltage, but I know you're on air cooling
I haven't overclocked Intel in a while, especially the older chips. What's the max voltage for that chip?
Yeah, I know it is about Voltage.
Most people seem not to exceed 1.3/1.35V
But I do not plan on stroking the limits.
It seems that 1.3V is the accepted safe for long term use
Neither in Voltage or Temp.
I reached 83-84°C in OCCT.
With only 1.2
Which is already kind of warm.
I'd say to lower clocks, raise voltage in 10mV steps (0.01V), until it gets too hot for your tastes (probably 85C in my opinion), back it off 10 mV then pump the clock. Then pump clock until it crashes, back off 100 or 200 Mhz
Though it seems like you're already there I guess, lol
Well, I spent my day doing that.
If you're at 4.4 Ghz, then 4.3 Ghz would probably be good. an extra 400 Mhz (if the max turbo is 3.9 Ghz) is pretty great
Well, not exactly that.
But first I tried achieving stable 4.3 in OCCT.
So I could see the temp it would reach then.
what air cooler are you using?
Hmm, yeah. That's a pretty decent budget cooler, but it can't do much.
A lot of people with 4690K I have seen on google seemed to have the same.
I have one on my old 1700X, max I could get was 3.8 Ghz, above the 3.5 Ghz base.
I would say to get a beefier cooler, but you could probably take that money and get a newer CPU and motherboard.
Oooooh no I cannot XD
lol
CPU + MoBo would cost me 10 times the price of a new cooler.
And I would have to change my RAM.
I thought 4th gen was DDR4?
What...
I could've sworn the only reason 4th gen wasn't compatible with 3rd or 2nd was because of DDR4
I chose the wrong generation to build my first PC with XD

I had a 3570k and I remember that I wouldn't have been able to upgrade to 4th gen
Wow, I feel extra jipped
Yeah, but 3xxx is third gen, not fourth.
This whole time I remember it was it being DDR4, but no, it was just Intel being dumb
Right. My goal was to have a 3570k, but then upgrade to a 4770k or whatever a year or two later
This was 8 years ago
I have tried putting my hands down on a 4770K.
I have been waiting like 10 minutes before the end of an auction on eBay.
It was 80€.
I looked for a 4790k and it's insane that they're all $100+ in the US
The 3770k was about $50-60 not too long ago
I got a 2600k for about $30
But 4 more people walked in the auction room and we started pressing the bid button like autistic monkeys.
Okay
Since I could not win the auction for 4770K, I backed off to the 4690K which was 100€ but on direct buy at a German seller (so very close to my country).
Was this a while ago, or was this a recent purchase?
Oh, you said you had a pentium, so it must've been kinda recent
How do you feel about the upgrade? Did it improve things?
January 2020.
Well, it was necessary as I wanted to change my GPU.
I was planning on buying both CPU and GPU on eBay.
I had a GTX 760.
I was looking for 1070-1080Ti, but those were too expensive for second-hand hardware. And I lost all the auctions for 1070Ti I found. Not to mention the new monitor I had to buy since the previous one was dead.
So I went for the second-hand i5 and a GTX 1660S (there was a discount on it).
I see
My Pentium could not have handled the GTX 1660S.
BTW, besides with OC Genie feature from the MoBo, I never tried OCing my Pentium.
How are your timings now ?
16-18-18-38 does not boot properly, but 16-19-19-38 did and is fine so far
I still have 1333 XDDDDDD
I'm considering just letting it relax back to 16-20-20-40 and leaving it there. Then moving on to PBO overclocking
So I guess timing are not a top priority for me.
Well, frequency is your 1333, timings would be the four numbers with dashes
1333 is usually CL9 or CL10
You could probably try
though I don't know how good a memory overclock would be
at least on that processor
Well, I do not think how I could get any benefit from this.
also, for reference, I have spent 3 days overclocking this ram now
Like, a solid 25 hours so far
Moreover, my sticks do not have thermal dissipation modules or whatever they call it.
So much patience.
A couple more hundreds of hours and you will reach the amount of time I spent modding my Skyrim again and again, hoping for a stable game.
In both case, you have to deal with hazardous instability issues.
The difference is Skyrim instability does not crash your PC.
Or corrupt your windows install I'd bet
But it dusts your fans off.
lol
And reset cmos can't fix skyrim :(
Lol
I mean, if you copied the game to another drive as a backup everytime you made a change, then maybe
I did not see this one coming 😂
Well, I used to manipulate my files that way.
But with an hard drive ?
I think once I confirm if this 3600 16-19-19-38 is stable, I'll leave it there, test, and then move onto PBO
It was taking forever to copy the 7Gigs of base game files.
lol yeah
Just gotta get yourself a good NVMe for way more than your processor costs, lol
BTW, how can tweaking something in the BIOS ruin your Windows install ?
Whatever pleases you Mister Einstein, where should I plug the NVMe on a Z87 Mobo ? 😂
BTW, that is one of the reason why I bought new SSDs a month ago.
Overclocking memory can make it unstable. Instability could mean that 1s and 0s get messed up. As a result, say, Windows does something in the background and it needs memory to change a value in the operating system.
If the data gets corrupted in memory and then it over writes that data into the operating system files, that can cause corruption
Including a 1TB one.
You could get a PCIe to M.2 card
Btw, you using 2t with gdm off or...?
The latest NVMe drives are only x4
1T GDM off... lol
Okay, I thought there were mechanisms designed to prevent this kind of errors.
Uhhhhhh
I guess GDM on with 1T, or 2T with off would be more stable?
That might be a reason of why if it's not booting..?
Yeah, the guide I'm reading did mention that
Gdm helps you get 1t with higher frequencies, but in return, iirc tRCD has to be even
I just know someone said 2T can reduce performance pretty significantly
That is cheating. And I do not have a real use for something this fast. I have a pretty low end old MoBo, so I am just expecting terrible bottlenecks.
Yeah, I know GDM at least didn't allow CL to be even. But back on the stock XMP, it was 16-17-17-35 with GDM On
Eh, tbh, I don't think it's that great a performance drop, anyways
I guess I could save this 16-19-19-38 1T since it seems stable
then try GDM and compare the performance
or 2T or whatever
BTW, why not boot on a Linux system instead of Windows ? If it can help you avoid Windows reinstalls every time you mess things up with your RAM ?
If it works, it works.
Todd Howards seconds that.
Gdm helps you when 1t doesn't work
Todd howards says if it ain't broke, don't fix it, if it is broke, fix it tomorrow
this testing program is on windows. And ideally, I'd want to make sure it is stable in windows
I keep my important files off my windows boot drive, so I can nuke windows at any time
He also said "It just works"
Yes
Actually, maybe I could partition it again and give, like 64GB to windows, then install the other programs to the remaining 275ish GB
I have a 40-60 GB partition for Ubuntu currently
Yeah, but having to go through the whole installation process, just to say "no, I do not want to share my location with you", "no, I do not want you to manage my bank account", "no, you will not have access to my pr0n browsing history", again and again... 
I don't mind it. Only takes a few minutes. I can go from nothing to installed in a few minutes
then just drivers
I used to always reinstall windows every 6 months, but I stopped doing that kind od
Yeah, but you need portable installations, or Windows is going to mess things up with filepaths.
maybe. I usually just remap the user files and programs
Remapping is a pain in the ass.
Maybe I could look into setting up an active directory or something, or using SCCM
With active directory, I should be able to map my user files and programs more easily
Like my fresh W10 install remembered the PATH to Winrar after I switched from W7 because Winrar was still installed on the D: Drive.
Hmm. Did you upgrade or format and install?
Format and install.
that's interesting. I've never seen Windows do that
Then I backed up my files from the D: to the E: (the 1TB SSD), and formatted D:.
Alright. I wonder if I should do 2T with GDM off, or 1T with GDM On. This is to try 16-18-18-38
2t more stable, 1t gdm more performance
Blend bad
Ah !
Smallfft is best for heat if you want heat
If you want to check errors just use OCCT :p
OCCT ran fine on 4.4GHz/1.2V
I thought smallerfft was best for heat, not smallfft
I changed to 1.22 because P95 Bluescreened my PC.
Smallfft is best for heat
Well, I'll try both.
Smallest fft is for something else, don't remember
BTW, why Blend bad ?
It's just low stress
Like my CPU runs 50c under blend, 105c max under smallfft when it was running 1.35v
Can I have a real time display of temperatures within P95 ?
No
Ok
Use smth like hwinfo64
It's smallfft, sounds normal for temp
95 !
Yep
doesn't like 16-18-18-38, or 16-18-18-40 with 1T GDM. Will try 43 and 2T, GDM off
98 !
I mean, I think I usually get 95C+ on P95 with my 3950X.
100
Might want to try and lower voltage if it goes over

Lol
But in reality, it almost never goes above 80C ,even on a 100% usage video encode.
Imma switch that off
:ln2: when?
lol, idk, maybe if I find a good/memworthy pic to turn into an emoji
I do not want to kill my CPU, it is the only one I have for LGA1150.
You'd only be able to kill it if you ride on tjMAX all the time
Or slam it with 1.5V or something
I do not know what to think.
The thing is I do not know how long such temps are going to be sustained.
I thought it was supposed to throttle when >90.
hmm. 16-18-18-42, tRC 62, 2T GDM Off doesn't work.
I just don't think 18 is possible for tRCD
I could try 15-20-20-40
what ic
I don't know the specific details, but Fal mentioned Hynix CJR
on the sticker, it's Rev 5.32 for the Vengeance LPX
I'm not sure where he found those details. Last time I tried looking for the chip types I had to dig through a ton of websites.
and ultimately gave up
Thaiphoon burner any help?
I'd go with googling what rev 5.32 lpx is rather than even bothering with thaiphoon
First results for googling "rev 5.32 lpx"

I mean, sure, but there's already two people in there that conflicted on CFR and CJR. Where did they even abstract that information from?
This means that there's some official documentation somewhere or something. Unless someone spent time removing the covers and looking at the chips.
I have seen some reach 4.5GHz with less than 1.15V, while I am here...struggling with my 4.3GHz hoping to stay below the 1.2V.
Did not win the manufacturing lottery.
Now I'm reading that it could be CJR or AFR for 5.32
I'm just gonna try CL15 and see what happens. If it doesn't work, I'll do 16-19-19-38, 1T GDM off and call it a day, since I know that's at least 25 minutes stable.
Did you not just say you were calling it a day should the 16-19-19-19-38 work ?
kinda yeah
Like 30 minutes ago
but I realized I never tried CL15
A man, true to his words !
Overclocking isn't the worst addiction to have
@orchid flame What's lpx rev 5.32
It is, BTW.
Otherwise it is called masochism.
You cannot do that just for fun
a worse addiction would be spending money on top end parts, regardless of value, and not overclocking a K chip on a $300 board
You could have put it in a simpler way.
"A worse addiction would be being drop dead rich."
lol
My brother has a 10900K on a $250 Z570 board, 64 GB of RAM, 1000w platinum evga, and a 3090
no overclocking whatsoever, in a Lian Li O11 Dynamic XL
NZXT Z73
makes me kinda sad, especially that his PC without overclocking is faster than my 3950X overclocked
Though, I guess it's kinda cheating, because they boost to 5.1-5.0 Ghz no problem out of the box
Well, it is kind of a factory OC.
Ver 5.32 is supposed to be CJR
yeah
You were right Fal, I did get 3600 CL16
just with 16-19-19-38
I did also manage it on 1T with GDM off
Oh nice! I figured you could get it there. I'm guessing it needed a bit more voltage?
Huh. Weird. Well if it works, it works.
3733 and 3800 were kind of working at CL18, but was not stable enough
I think my 3950X doesn't like FLCK above 1866
Pretty common for the 3000 series to not like over 1800. Past that it's up to binning and luck.
yeah
hello all, i am new to playing games on a Pc i am currently running a gigabyte 2070 with i7 9700k on a aorus z390 pro wifi board. would it make a substantial difference in FPS if i were to OC and would it be worth the extra amount of stress i would be putting on machine?
I also kept the tRFC (apparently) a little conservative at 280ns, which is the top end of of the 260-280 range specified for CJR on the guide I was using
Overclock your memory and the GPU and you could get 10-15% more fps @radiant turret
I am already sustaining 92°C with 4.3GHz, 1.185V, no wonder it would go 98°C on 4.4GHz with 1.22 and still keep crashing.
98 at 1.22?
the hell
what cooler lol
i ran the ram up to 3200 with the xmp in bios already is that what you would consider an Overclock or are you talking manually go up
212 Evo
1.35 On cpu? Can you cool it and how long do you need it to live?
Not a single one XD
yes i can cool it, uh
Ewww 212
this battery replacement needs to get here faster. My Pixel 3 went from 100% to 14% in 4 hours and I think the back is bulging a little more than yesterday
normal lifespan
It is a decent cooler considering the price.
Just not meant for OCing.
gammax 400 is better
No it's not considering coolers half the price cool better
212 is a prism but bigger and looks worse
Dude, I am running a Haswell processor.
I built my during Haswell era.
There was not many coolers this efficient in this price range.
Yes, back in the day it was good but these days...
I can attest to that. 7-8 years ago, it was Evo 212 or $100+ AIO
These days do not matter since I built it 6 years ago. And prices are not the same in my country as in the USAs.
so should I call you 4690K or Echec?
Account name
Yeah, yeah.
that's what it says your username is, lol
Did not think he would check it out.
well, it's weird to call someone 4690K
Just need to click a name to see it
I was going to say "Echec built it a while ago with a pentium", but you said it first, lol
Well, "EchecCritique.exe" is because I always prove unlucky when it comes to random.
I can generate statistic anomalies.
BTW, someone mentioned OC-ing GPU. Can it be done efficiently on a Zotac card with original air cooling system ?
noo! 16-19-19-38 threw an error!
"Ha ha haaaa ! You didn't say the magic wooord !"
I was able to overclock my 2070 Zotac, but it was the non-A versoion or whatever, so it was limited to 200W

For GPUs, you usually run out of stability before you run out of cooling.
There's two 2070s, a normal chip, and a chip limited to 200W
Did not get much of a choice as to my 1660S.
You underestimate the power of the meme that is Zotac
I got a Zotac Mini, which turned out to be the 200W limited one.
Unless it is implied.
I also killed my 2070 by touching two wires on purpose-accident.

yep
I would throw myself out of the windows for ruining my 200€ card.
I would not stand it.
So, stock values, right ?
stock values for this ram is 3000 Mhz 15-17-17-35
Ah !
I will have it at 3600 Mhz 16-20-20-38 for this next test
How much of an impact is increasing CAS latency while boosting clockspeed to an extra 600MHz ?
higher frequency usually means lower access time at the same CAS
So it is correlated ?
see "effect on memory speed access"
Column Address Strobe (CAS) latency, or CL, is the delay in clock cycles between the READ command and the moment data is available. In asynchronous DRAM, the interval is specified in nanoseconds (absolute time). In synchronous DRAM, the interval is specified in clock cycles. Because the latency is dependent upon a number of clock ticks instead ...
yes
Okay
So you can do some math beforehand to get an expectation of how efficient your RAM will be according to the latency/frequency tradeoff ?
So for 3000 Mhz CL15, it's:
10.00 ns 11.00 ns 12.33 ns
That's first word, fourth word, eigth word. Basically just means parts of data
yeah, basically kinda
Okay
But there's more than just CAS latency. CAS Latency is just the first number of that 15-17-17-35 sequence
You can calculate the cas latency in ns by using this equation: "CAS"/(Speed/2000)
Oh, you can use the freq/2000 calculation for any of the numbers? At least for from/to nanoseconds?
Someone post the excel sheet
Close enough #cpus-mobo-and-memory message
Is the CL determined from the four numbers sequence ?
CL is the first number, aka CAS
So in "15-17-17-35" it is 15, right ?
yes
Okay
Does that even exist ?
I don't think so
yeah if you wanna run 7-48-48-89
No, not possible
Benching on 4800c14 is meta
or 4000 12 11 11 28
Rocket lake can do 5100+ with straight 14's
I will never have the money for that s*** anyway...
DDR5 is a different story, theoretically we can get reduced latencies with the new bus design
I don't think it actually has any significant performance anyways. Only really used by overclockers for peak numbers
Okay, so less than half as mine XD
Hey @proven canopy do you think it would be worth pushing my b-die kit to 4800, or should I keep it 1:1?
Ouch, my P95 got the long face. 90°C in average for 1.185V, 4.3GHz, I guess I will call it the right spot to stay in.
I know der8auer did a video on desyncing the FLCK and memory
I forget what platform you're on, but 4800 cl(anything) is not going to be daily stable
Zen 3 / X470
IIRC the result was that desyncing FLCK was not worth it
lol no
he even tried 5100 Mhz ram
5100 ram is ass most of the time
19-26-26-48 at 1.5v
yeah
that's actually pretty good
I'm sad 16-19-19-38 threw an error but I guess it's whatever. I just hope 16-20-20-38 is stable
Nobody who buys 5100 ram runs it xmp
Shouldn't really be doing that with anything rated over 4000
Especially on an asus board, 
So, what are the three numbers following CAS ?
system agent gonna assassinate the cpu
This content hopes to define memory timings and demystify the primary timings, including CAS (CL), tRAS, tRP, tRAS, and tRCD. As we define primary memory timings, we’ll also demonstrate how some memory ratios work (and how they sometimes can operate out of ratio), and how much tertiary and secondary timings (like tRFC) can impact performance. -
of course i don't get dynamic OC switching on b550e
Oooh !
I bet I can get a 10900K to run at 6.5 Ghz on air, but probably only for about 2 seconds
Okay, those are those things that wear weird names.
Which comes near to 0 interest XD
Except for flexing on Reddit.
You bet?

lol
He dares you.
The naming makes sense on latencies once you know what the acronyms mean and how they work together
Kill it, buy him a 5900X, rebuild the system all without him knowing 😂
I think he'd notice the drastic increase in thread count
I know, I was just referring to a lot of comments I have seen on this channel, where people would say "Okay tKchépaci is at 16, tQchépaça is at 16" and all those things that made me think those people were diagnosing someone and checking his vitals numbers and all.
LOL
"Look, he lacks magnesium and he got severe hypokaliemia."
Someone talk me out of doing 1.55v daily on my memory, I'm considering pushing for 3800C12
I mean, if you have a spare kit, go ahead 
1.55 most likely wont get you 3800c12
I tried up to 1.65 on my good kit, and was barely boot stable
When you fall short on ideas of what to do with your money XD
if you have just 2x8, go for 3800c13 gdm off
I'm already doing 14-13-13-28, I might just do that
if I had that kind of money, I'd already have purchased 450 TB worth of hard drives
And yes I do have a spare 3200 C-Die kit
Of magnetic tapes.
no, 7200 RPM HDDs
No, mag master race
I could get a 45 drives chassis and 45 10 TB drives
worst writing speed evah
probably about 642 TB usable
Get 10 and make it a 1PB raid array
Yeah XD
Reason: Bad word usage
oh woops
GG
can't say the f word I guess
I was gonna say, if I had f you money then sure, 100 TB ssds
Half of my vocab consists of bad language, but I manage to stay unnoticed here.
They're 40k US each
Holy crap
yeah no way, that's just wasteful
At least they're SAS drives for servers lol
12 Gb speed
Does not matter if you have to lose 100TB of storage for a single drive failure.
I was joking about the raid array, I know it's best to do it differently
And all this for hosting a Webhost-grade quality hosting service XD
I wouldn't mind getting PCIe 4.0 NVMe drives (TLC preferably) and doing, like, 8 of them in RAID 0 😜
Raid 0, aka YOLO raid
SLC master race
I was actually planning out a 8x 1 TB SSD server that would've been RAID 0 for a place I used to work at. Was going to be a lancache
20 Gbps SFP+ teamed or 40 Gbps QSFP
Make sure to have some sort of fault tolerance or error checking in place with all those disks
Yeah. I know that many enterprise drives come with automatic error checking, and it would've had ecc memory
oh gosh I really hope 16-20-20-38 is stable. I'm kinda tired of Ocing the ram now, lol.
That's... not sufficient. At least have a couple parity drives.
20m in, no errors
🤞
I guess so. Could make sense in a lancache setup where it barely writes, but mostly reads.
RAID 5 I guess it was? Not sure what the equivalent in ZFS is
Yeah raid 5 is raid 0 + parity
Raid 10 would be even more ideal, with the speed and size of 0 with the redundancy of 1
Ah, I remember now. RAID 5 has basically 1 drive of redundancy. Write speed of one drive, but read of all drives.
Would've been great for a lancache actually. Max speed on a single provider was 1 Gbps anyways
after this memory test, I'll do an encoding benchmark and then start working on PBO for my 3950X
Question, my Small FFTs has been running for almost an hour now, and it is 2 am. Any idea how long it is going to last ?
P95 runs infinitely IIRC
What ?
Yep, it runs as long as you allow it
I think 1 hour is fine. Though, you could run it over night
I ran a Blended test yesterday, and it stopped eventually.
I wouldn't do that one overnight tbh, too much risk
If you run it overnight and it is still running in the morning, it's a very stable OC
really?
PC should just reboot
Oh, not small ffts overnight
yeah
you'd probably want smaller fft whatever is a little less than that
though I could've sworn that smaller fft was actually more intensive than small fft
If it overheats and everything is working right, yes ideally it just shuts off. There's always the chance it doesn't work right and gets stuck or melts something
ah I see
And say, I was happy with 4.4GHz, but I had to go down to 4.3 😭
I'd take stable and smooth over 1-4% faster
It is "smallest"
And yeah, it is more intensive than "small".
Yeah, that is what I think too XD
My CPU does 4.7 all core but uh, it gets HOT. I need a bigger AIO or something, but to get that I need a new case or a shorter GPU...
what AIO do you have now?
I guess I could try OCing my GPU a bit.
If I've learned anything, the only logical upgrade from an AIO is almost always a custom loop
I would not spit on a 10% perf increase.
ML240L, cheapo one I know but it worked for my old 65W CPU overclocked
Hmm
I could do a 360 AIO on the front but my GPU is too long
I guess you could try the Arctic 280 or 360, but otherwise I'd just look into custom watercooling
This is the logical next step yes, to finally make the move to custom cooling. My current case will not support that though, I'd get a 5000d or something but then I need a new desk too (or mod the one I have).
someone I know has a 5950X in a custom loop with a GPU and it maxes out at 65C-70C stock. But my understanding is that Zen usually is hotter on stock than stable OCs usually are.
Cannot wait until I can build my next PC with a freezing color scheme.
Which case do you have?
Meshify C
Hmm. I feel like I remember some case that supported 360mm and 280mm with plenty of space, but was pretty small
It does support a 360 on the front, but my GPU is too long. It's only 8mm from my front fan right now.
Zalman Z11
oh wow
I think you can slide a 360mm in it.
Thinnest rad I found was 20mm
But it is ugly AF
This mini USB about 1.5mm from the fan blades. Lol
which is hilarious because I forgot how asetek mounts work
I forgot you can set them at about 30° increments
I hate being late to go sleep.
how long did it take to get to 95C?
And did it hit 100C then throttle?
Also, you do have a fan at the front of the case pushing air towards the heatsink, right (not including the fan on the heatsink)?
Who are you asking ?
You
I ask because I had forgotten to plug in the front fan on my nephew's meshify case. Sliding it up, adding a second fan, and plugging them in reduced temps by 10-15C
I have two pushing, one right on top of the radiator, and the other at the back.
They do not seem to change speed.
I see
My case is a Zalman R1 with stock fans.
you may want to look at the fan settings in your bios
I guess they are on max speed stock.
oh, really?
I am going to check this.
oh my
hmm that's pretty high already I guess
I mean, it may help, just depends on what you are okay with for sound
Well, I do not mind the noise.
I already have a "lame extractor hood", the compressors of my fridge and freezer whistling H24, so...
lol
I got to go now, thank you for your advices, the talk and all. Night :D
good night!
40m was good, so I'll call that stable
wrote all my settings down on a piece of paper just incase, lol
CJR
given the PCB
should wall near 4000
thanks
or do 4500+
sir is this CJR?
yep
5.32
jedec
Oh
Don’t look at JEDEC Refresh tbh
I do wonder how much higher tREFI will be on DDR5
Smaller nodes means smaller capacitance and less leakage, means higher tREFI
And lower tRFC as well
inb4 same because jedec isnt original
10ns tRCD is bound to fail
at least at/above 3600
also
why is GDM off?
I was advised that GDM off can allow for slightly faster speeds.
I guess it wouldn't hurt to turn it on
but I'm kinda tired for today, tested my 16-20-20-38 twice and it's basically stable, so I'm gonna move onto PBO.
GDM mostly affects CAS. If that's an odd number, GDM adds 1 to CAS. Without GDM on it'll default to 2T command rate, which tends to be slower than increasing the CAS.
Wait, I'm confused
should i overclock my gpu?
Everytime I used GDM, it would force odds to evens. But you're saying that without GDM, it can run odds, but sets 2T?
So GDM with odd numbers allows it to run at 1T, but not as fast as 1T without GDM?
But that doesn't make sense, because it forces it to evens
Idk, my OC is stable on 16 with GDM off and 1T. moving onto PBO after a benchmark
...no thats the opposite
anyways
if you want to run CJR at ideal
you'll mainly be running 3800 16-20-20-34 or smth smiliar
and since you're on AMD
3800 CL16 was unstable, even at 22-22-42, but I did run it on 1T with GDM off
Maybe I'll try it some other time
GDM is for stability mostly
you'll be aiming at about 280ns tRFC
it does help stability
especially at higher clocks
Yeah, I currently have 280ns, which works out to 504 for 3600
what vDIMM?
then the /1.346 and /1.625 rule for 2 and 4
Ryzen tends to be less stable with odd number CAS at CR 1T, so GDM is to help with that
1.45 is max I'd daily
lower tRCDWR and tRAS
change this
1.05
assuming this is 2 dimms, change your RTT
rest looks good
isn't tRAS usually tCL + tRCDRD +2?
tfw you wrote it in a notepad instead of using notepad
what are we, Jufes
Yeah, because 99% of my time was either in BIOS or with the test running, so my computer was basically frozen
which test? TM5?
Yeah, with anta777
extreme or whatever
I just run occt 
So what would happen if I accepted my current OC and waited to do those modifications you mentioned until later? Is it that much performance?
I've basically done this for 30 hours straight and I'd like to move onto PBO, since people said PBO would be insane compared to my current 4.15 Ghz OC.
On a 3950X
is your batch newer than Feb 2020?
3950x
I bought my 3950x in I think JAnuary?
meh
you'll have some issues with PBO
but thats fine
feb 2020 was a massive mask and litho level change for Matisse
of course it was -_-
So is that why the benchmarks I see online are, like, ridiculously higher than what my 3950X achieves at 4.15 Ghz?
dpeends
but to give you an idea
before Feb, 4.4g AVX2 stable was considered "insane"
afterwards, it was pretty average, with chips even hitting 4.6
because it is the same chip
we see this sorta stuff all the time
Reason: Bad word usage
lol
Early B die sucks compared to new B die for example
8700ks were better before the 8086k came out
11900K Engineering samples DEMOLISH retail chips
so on so forth
So I should wait to buy a new chip until at least 6 months after release?
theres a version of the 3900x that does 4.9ghz
jesus christ
its super rare
I'm looking to buy one
So is that a more detailed look into "silicon lottery"?
meh
thats moreso variation
It's another layer to just manufacturing tolerance?
I do this on principle, for everything. Let them work out the bugs first. I don't need to be an early adopter.
these are changes to stepping, node improvements and mask improvements over time
3950X was one of the first few things I bought brand new. Only other time was the GTX 970, which turned out to have the 3.5 GB VRAM issue.
Alright, so. I remember the last time I was really trying to overclock my 3950X, I played with per-CCX in Ryzen Master (since my ASRock X570M Pro4 doesn't support that at a BIOS level).
I think the spread was 4.35 Ghz, 4.40 Ghz, 4.5 Ghz, and 4.575 Ghz. Though, nothing above 4.35 Ghz was really that much better for R20 scores, and ultimately pulling the clock down to 4.25 Ghz was faster than anything I had ever done with 4.3 Ghz or higher.
So in that case, would PBO even be worth it? Or should I just set it back to 4.25 Ghz and call it a day?
I think the spread was 4.35 Ghz, 4.40 Ghz, 4.5 Ghz, and 4.575 Ghz.
thats good
what does it drop to in Prime?
were you on small FFTs?
you hit 4.6ghz Prime95 stable?
I do know that 4.35 Ghz ended up having rounding errors in P95, so that's what prompted me to eventually drop to 4.15 Ghz.
I was on small FFTs, because it was my understanding that it was more intensive than smallest/smaller, or whatever the top option was
Oh nonono
Well
I can't remember
I don't remember if I even did P95 with the per-CCX in ryzen master
my 240mm rad was letting it creep towards 100C so I wasn't playing with it
before I forget, I'm going to benchmark this RAM OC
so yeah, the most stable I got from what I remember was 4.35, 4.4, 4.45, and 4.5. But it barely did anything for R20 scores so I gave up
4.575 was the highest I got on a single CCX without the system crashing, but it wasn't necessarily not throwing errors
Alright, the results are in.
Original memory speed of 3000 Mhz 15-17-17-35 ran the video encode at 641.78 seconds.
3333 Mhz 16-17-17-35 (which wasn't really refined further), encoded the video at 629.25 seconds.
3600 Mhz 16-20-20-38 encoded the video in 620.35 seconds.
Which basically means that a 25 hour video encode will now only take 24 hours and 10 minutes. Pretty cool.
It doesn't seem like a lot, but I guess because it's something I'll be doing a lot, it feels more meaningful, I guess 10,000 vs 9,760 in R20
Knocked almost an hour off that 25 hour video, not bad for only changing memory.
yeah
I know I could pretty easily just push it back up to 4.25 or 4.30 Ghz on the CPU, but I figured I'd give PBO a try, since people have claimed it can help push individual cores or CCXs up during a heavy load.
It does usually jump between 70-100% utilization a lot, so maybe it can
3.4% improvement on your encode times, that's about right I think
I just round up lol
or actually 3.276%, but close enough
Could say 3 1/3 % or so
yeah
I still can't believe that I've effectively had 3600 Mhz CL16 RAM just sitting in my PC this whole time
That's the power of overclocking, unleash that potential!
btw, I'm guessing leaving it at 1.4V is fine? No need to try to bump it down near 1.35V again?
You can try it lower but 1.4v is perfectly safe
If you do try to bring it down, go in 0.01 volt increments and test after every change.
1.45v max safe daily imo
That's what I heard too, 1.45 is safe for most dies and 1.5v is safe for b-die on the daily
According to this guide, in order to do PBO properly, I need to record stock behavior for p95, cinebench, and gaming.
Which is annoying, because I did this whole RAM OC on the 4.15 Ghz OC 😂
I wonder if I could just set the OC back to auto mode
but I guess a factory reset would be better
then apply the RAM OC to the stock
Save your overclock profile in BIOS before you change it, then if you want to go back it's easier. No need to clear CMOS.
I can still do a bios reset without clearing CMOS, unless that's the same thing
It sets all the BIOS settings back to 0, but keeps everything else, like my OC presets
Your BIOS must have storage for the profiles then.
RAM overclock won't affect your CPU behavior. It's mostly watching what cores boost to what frequency, memory won't affect boosting.
wow, this guide basically ended up being "do what you're already doing"
So no PBO, probably, kinda
it's a pretty awful guide to be honest
No TL/DR, read or don't read, w/e not bothered either way.
Overclocking Zen 2 is not as straight-forward as some would like to believe, it is not an Intel, so stop treating it like one.
I am not a fanboi of either camp, so please don't think I'm bashing Intel methodology, it's simply...
It's a header, then just gigantic paragraphs of instructions mixed with reasoning.
That is definitely not a friendly guide. I think I know a better one.
Skimming some other links, it almost feels like I'd get a better boost by just going back to 4.25 Ghz all core and just limiting my games to the two middle CCXs where I can
I already have an inherent issue with the CCX design, getting 1-10% more performance depending on the game by just using ProcessLasso
It just sucks, because some games I play don't allow you to change the affinity
If only you could set it so PBO works on half and you could do manual OC on the other half.
I mean, I'm sure if I actually followed the recommendation in that guide, I could probably find a PBO setting that allows basically a 4.25 Ghz all-core, while bosting to 4.4-4.5 where it can on some cores to help with game performance
But that would probably take another 3-5 days, and I've spent 3 days already on this RAM OC
So it sounds like I could probably get away with 4.25 or 4.30 Ghz, and boost the voltage up a bit (since it currently droops to 1.26V on the second highest LLC setting).
I do know that Zen doesn't like when you cut it short on voltage. It's kind of like the ECC memory on the 3090, where if you go too far on speed or low voltage, it will start to take a performance hit.
I wish it worked like Intel where it could still automatically boost, but you can specify a minimum all-core under load
But then I guess Ryzen wouldn't be all that fun, would it?
Is there any chance that a higher synced FLCK can allow for slightly higher core overclocks?
Because all my previous OCing was at 1500 Mhz, not 1800 Mhz
No, fclk won't have anything to do with the core frequency. It's more like how the cores communicate with the cache and each other.
fclk is connected to ram not cpu tbh
yeah but ram speeds affect it not cpu speeds
Kind of......
connected as in affected by
Well, you can desync them
FCLK should match ram speed
but it's not necessarily affected by it
You could even run fclk 1800 with RAM at jedec
true
Is 4x8 still better than 2x8 in AMD 5000 series/dual channel? Without manual ram tuning and only factory DOCP used?
Eh
Its better on all platforms
Does anyone know a good free alternative to 3DMark ? (I would like to bench my GPU for OC)
Just got my 3600X stable at 4375mhz all core at 2.6V
There's the time spy demo version on steam that's free. You could also do heaven benchmark.
Yeah, Heaven is the one I chose finally.
Thanks
Seems doing pretty good.
But I guess I should not expect huge improvements since my GPU is already reaching 75°C.
You could set a more aggressive fan curve, but that will only do so much. It shouldn't throttle too much until it's over 95 at least.
Does not seem bad (I do not know the average result on this kind of benchmarks).
For the fan curve, I do not have any BIOS options that would let me change them.
I guess I am supposed to do this through Windaube ?
The gpu fan curve you'd adjust in afterburner
The idea is to do a before and after benchmark to see what you gain
I'd be concerned about that huge drop to 8 fps, or maybe just a background task
Yeah, I noticed some slight stutters during the benchmarking process.
Besides Discord and the Windaube Antimalware service, I do not have major background tasks being executed.
Afterburner is MSI Software right ? Would it work on a Zotac GPU ?
yes
It's for all gpus
it works on everything
Okay, right. Thanks
Best all around gpu overclock software too
There's a monitoring software you can use your phone as a second screen for
Oh ?
Since my phone got a way better screen than my actual Desktop screen...
I would like to be able to display the image on it
No, it would just read the sensor data
Already tried an Android app for my tablet, but it sucked.
Imagine shilling 2R/channel
Cringe
Also nice Sion reference
Reason: Bad word usage
Out of curiosity, I launched a scan. I can see it moves the clock slider overtime (it is on +135 at the moment) : is it just testing how much I clan speed up my card ?
Yes, it's increasing and testing each step for stability. Once it finds a step that's unstable, it will drop that back and test the memory the same way.
Manual overclocking is preferred over the auto option, but it's a good place to start
The benches with global points are most relevant
Oh, thanks.
So I should try "Unigine Superposition - 1080P Xtreme", I guess ?
Sure
Heaven is fine btw
Honestly just pick up 3dmark for $5 lol
And use time spy gt2 for testing, that's what I do
Yeah, I guess I can just use the numbers auto mode is going to give me and tweak those.
Is it not 25€ ?
That is the price I have seen on Steam.
Very often goes on sale for 5
If you buy on steam, just extract the keys and run the standalone
Not very sure to understand what you mean by the "extracting the keys" thing.
But I take it is a collection of benchmarking programs I can run independantly, right ?
Yes, link is in the pins
It got me in for a few seconds, I am not going to lie :
I was like "OOOh, you are so much ahead of your time !"
Thank you for pointing the pinned messages BTW.
The scan finished with +118MHz core clock
and +200 MHz memory clock.
Oh wait, "Results are considered unstable" 😢
So right now I have a r9 3900x, asus tuf 3080, 32gb corsair vengeance rgb pro ram at 3200 mghz, and a crosshair 8 hero. I "overcloked" it to the dchop thing in the bios but I don't rly know if I should overclock anymore, what do u guys think?
💰
Question: How does it come to that conclusion "https://outervision.com/b/ipXXIN" ?
Do not tell me I will have to change my 500W PSU 😫
Never ran out of power, BTW.
PSU calculators are kinda bad. Your 500W will be pushed over that occasionally but can still handle things except in specific cases, like running furmark and prime95 at the same time.
Of course, which model of 500W power supply you have is more important, since if it's a bad one that doesn't have OCP or OTP then it may catch fire.
can i oc my rtx 2060 with 500 watt bronze psu?
It is a Corsair CX500 80+Bronze
CX500 is good, it'll shut off if it's pushed too far for too long
Which psu and what processor do you have?
i7 10700f and 500 watt bronze from coolermaster
That is reassuring, thanks 😃
i7 10700f and 500 watt bronze from coolermaster
Mr1111 walks into da plaz
Btw, that sounds like old cx, like green label, since I believe modern cx are -50w
@open surge what color label is on it?
Yeah, it is green
@sterile flame Is the middle finger icon allowed ? XD
It is 6 years old BTW X'D
I'd ask mods before I asked admin...
It was just a joke, there is no one tagged with "@sterile flame" anyways XD
It is like if I were asking God.
Welp, says it's group reg and DF, dunno how good the actual protections are
(role is not all uppercase)
Wait a min...why did it not suggest me anything while writing, then ?
It's some poor guy being pinged
Case matters? I'm not fully fluent in my Discord
Oh, okay, did not know that.
Maybe it is a toggle option in the settings.
Yeah, you have to work harder on learning declinations.
Roles correctly being pinged would show the correct capitalization
And I'm not on desktop so I can't check, but on desktop, you can right click the @ and if it shows you options for profile and such, it's a person
Yeah, and because it did not show me anything, I asserted no one had such a tag. I would not have taken the risk to stroke the Ire of some Admin for the sake of a silly joke 😂
That description could be any one of a half dozen power supplies, but they all range from bad to explosive. You can try overclocking but have a fire extinguisher handy if you push the system for extended periods. I know the 10700 is a very thirsty CPU with power limits disabled.
this psu
Uhhhhh
DANGER
Tier E
admin isn't a pingable role
Yeah, but it is not "10700" but _"10700F" _with an "F", like in "Fails to preserve itself from Power Supply burnout."
🔥
(not here at least)
Ya


