overclocking

[short blade]

just saying

[faint tangle]

I did that the other day and the closest bench had a 1080 to

Ti*

[short blade]

if your 3060ti is anywhere near a 1080ti then something is seriously wrong with it

[faint tangle]

So either my 3600 is just that bad and unigen uses cpu too

[short blade]

unless it's like a 1080ti under LN2

[faint tangle]

Or my 3060 ti is awful yeah

[short blade]

no like

stock 3060ti should be beating OC 1080ti by a noticeable margin

[faint tangle]

I'll show you the results

Yeah I know

1080 ti is on par with 2070s

3060 ti = 2080s

[short blade]

my 3070 laptop beats my OCed 1080ti by like 15% lol

my 3070 laptop is almost exactly on par with a stock 3060ti

[faint tangle]

Yeah I know which is why I'm really pissed at skytech

This is gonna be my pc for atleasy 3 years

[short blade]

well that's not really skytech's fault

[faint tangle]

So if it's really that bad I'm pissed

Idk

If its both my cpu and gpu

[short blade]

silicon lottery is nobody's fault lol

[faint tangle]

Theyre buying parts thay are performing less than normal

[short blade]

when you buy a piece of hardware you are only guaranteed to receive stock performance

[faint tangle]

After everything else they cheaped out on?

[short blade]

you are receiving at least stock performance

[faint tangle]

Idk I don't trust skytech whatsoeber

Not to mention their customer support is awful

[robust aurora]

Which model is it?

[faint tangle]

2/10

[short blade]

do you really think AMD and NVIDIA take the time to bin each individual 3600 and 3060ti and sell bad ones for cheaper

[faint tangle]

Gigabyte gaming oc pro

[robust aurora]

Shouldnt be too bad then

[short blade]

3060ti is literally just bad bin of 3070 already

[robust aurora]

Unless its like an hp or something it should perform good on stock

[short blade]

dell cooler

[faint tangle]

Tho for whatever reason it's really good at mining

[robust aurora]

[faint tangle]

62mh/s with 65% power -350clock +1300mem

[robust aurora]

Thats great

[faint tangle]

Yeah

[short blade]

what is that in watts

[robust aurora]

More then my 3070 used to do lol

[faint tangle]

143w

[short blade]

oh that's really high

[faint tangle]

Wattage or hashrate

[robust aurora]

Oh mine did 59 on 120

[short blade]

should be doing 60-62 mh/s at 120W max

both of my 3070s do 62 mh/s at 115-120W

[faint tangle]

I'll prolly play with power limit then

[short blade]

my 3070 laptop does 67 mh/s at 115W

[faint tangle]

67?

[short blade]

i don't use power limit at all

[faint tangle]

Jeez

[short blade]

i use flat v/f curve and let it choose its own voltage

[faint tangle]

Wait how tf

[robust aurora]

I didnt push the memory too high tho cause i didnt know the temps i had

[faint tangle]

That's insane

[robust aurora]

Now I have a 3070ti so I can atleast know the memory temps

[faint tangle]

I was thinking about finding a miner and trading to a lhr 3070 straight trade

Had one set up but he ghosted me

Bc then I could just mine raven on a different algorithm

[robust aurora]

Hmm i would try to find some eth miners that will give you a 3070ti

[faint tangle]

It only targets btc eth and daggerhasimoto

[robust aurora]

So you can get same money with raven mining

[faint tangle]

Idk idt I'd find a 3070 ti trade

Maybe a 6800

[robust aurora]

3070lhr raven mining will give you like 20% less then 3060ti or 3070 non lhr eth mining

[faint tangle]

But even with 3070ti/6800 with my cpu it'd be such a bad pairing

[robust aurora]

Your cpu is fine tbh

[faint tangle]

It's 7700k level

[robust aurora]

For streaming and playing can use something better i guess

[faint tangle]

At 1080p 144hz not really enough

[short blade]

that definitely depends on the game

my 3600x was good enough for 1440p165 in most AAA games

[faint tangle]

Yeah 1440p

[short blade]

...it's higher than 144hz

[faint tangle]

And 3600x is a decent bit better than 7700k

[short blade]

got 180 fps in cold war

3600x and 3600 are basically the exact same thing

[robust aurora]

On 1440p tho its less cpu..

[faint tangle]

Yes

[short blade]

and idk where you're getting that the 3600 = the 7700k

[faint tangle]

Cinebench benchmarks

[short blade]

this is not true

[faint tangle]

Normally equals 8700k

[short blade]

144 fps is 144 fps to the cpu regardless of the resolution.

[robust aurora]

Well gaming and cinebench is different usage

[short blade]

cinebench is not at all representative of gaming performance

[robust aurora]

10700k gets like 3k less points then 5800x but almost same on gaming

[short blade]

i7-4790K would get smashed by threadripper 1950x in cinebench while beating it in practically every game that isn't star citizen or possibly HZD

[south sky]

5800x is for sure faster

a stock 10700K is like 5600x level in games

[robust aurora]

2-3%

not the 20% in cinebench

[south sky]

more like 30% in cinebench

[robust aurora]

Hm its 20

[faint tangle]

I'm about to run unigen superposition benches, 1080p medium, high, or extreme?

[robust aurora]

13k for 10700k and 15.5k for 5800x

[south sky]

10700k doesn't get 13k

[south sky]

stock it's more like 12.2

[robust aurora]

Mine stock gets 12500 but i have seen people get more idk

5ghz all cores I got 13300 i think

[south sky]

stock it runs into power limits

[robust aurora]

Well 20 or 30 doesnt matter, its just proving my point even more if its higher...

For gaming the diffrenece is way smaller

[south sky]

anyways, the 7700K is still a very powerful chip for games

[faint tangle]

this was my closest similar bench

1080ti 8700k

only thing on this entire page is 1080tis

well one 2080

but you get the point

but then on the next page theres a mix of 2070 supers, 2080 supers and 2080 tis

this makes no sense

hell theres even a 3090 here

21302 for my score, if any of yall know how that compares

[south sky]

[lavish tundra]

on what benchmark

[faint tangle]

unigen superposition

1080p medium

i wonder if the 1080 tis i see high up are cli 1080tis

[ruby summit]

Can any overclocking nerd tell me if this kid is cap? He says he's doing this on a R5 3600 non x that he didn't specifically choose for good silicon, just random one off the shelf.

[zenith palm]

1.65V LMAOOOOOOOOOOOOOOOOOO THIS KID DEGRADED THEIR CPU IN MICRO SECONDS

1.35v is the max safe voltage for ryzen

[ruby summit]

Yeah that's what I was thinking

It's gotta be near dead

[lavish tundra]

1.65 would kill the cpu in a day

max

1.3 can degrade zen 2 chips prolonged

[ruby summit]

Lol ye

[dull ginkgo]

Sounds correct if they're actually pulling 1.65v

But that chip will degrade and it won't be stable in like a day xD

[ruby summit]

How convenient lol

[dull ginkgo]

lol

[ruby summit]

As soon as I question him about it he drops his voltage by .3

[dull ginkgo]

1.35v isn't even guaranteed safe

[ruby summit]

Yeah

Whats a 3600 run stock?

[dull ginkgo]

wonder if they ran their chip at 1.65v once and degraded afterwards and just thought something else was happening

[ruby summit]

Maybe

[zenith palm]

lmao for about 7 seconds kek

[sterile flame]

1.325 for zen1+ from my testing

2yrs

[dull ginkgo]

meh

depends on generation

zen 2 works a bit differently afaik

people had degradation reports at like 1.3v

[tall pelican]

just look at the nodes their on, zen was on 14nm (glofo?), zen+ on 12nm (glofo?) and zen2/3 on 7nm (tsmc)

[modern walrus]

5000D for $99 if anyone cares

https://computers.woot.com/offers/corsair-5000d-tempered-glass-mid-tower-atx-pc-case-1?ref=w_cnt_lnd_cat_pc_7_4

[south sky]

No way

try that on zen 3

you'll have a fun time

[short blade]

on zen 3 you'd probably see degradation at 1.35v within an hour

[south sky]

nah not that quickly

[lavish tundra]

1.3

for zen 3

1.25 for zen 2

[short blade]

i got degradation at 1.4v in 15 minutes

superpi, wprime, and cinebench r20

3 cores degraded

[south sky]

zen 3 literally has higher current density

power delivery is the same

more current per voltage

[lavish tundra]

So 1.25 is max safe for zen 3 as well?

I didnt understand

[south sky]

I don't know

[zenith palm]

Oh woops couldn't remember exactly just knew it was miles over the safe voltage lol

[south sky]

oh did you mean max stock voltage?

[zenith palm]

Ehh tbh I can't remember lol i had just woke up who knows what i meant lol

[tall pelican]

https://www.3dmark.com/3dm/63605883 that's pretty cool

[zenith palm]

22389 gpu damm

[short blade]

okay so I found the problem with my 3080

been sitting here for 3 hours trying to get it to work to see if I can avoid actually shipping it out

the problems occur if I attempt to turn on resizeable bar

when it boots to windows it says resizable bar disabled

bios freezes

any ideas?

[dark ferry]

I'd re-flash your system and gpu bios.

[rough cairn]

Most likely the problem is in the bios. Update to the latest. But then may still be a problem depending on the board producer. I still have the same issue on asus tux x570 3202 bios.

[short blade]

latest bios on both b550 tuf-plus and b550 steel legend

[left blade]

AC_PKKA#1131 has been warned

Reason: Bad word usage

[short blade]

please do not avoid filter in the future

one of the motherboards is not even asus

[muted night]

Anyone here familiar with memtest86?

[zenith palm]

Slightly

[muted night]

While running the test it displays my rams xmp profile, not my OC settings is that just a visual thing or is it not testing the oc?

[zenith palm]

Ehhh mb you hit exit without saving the bios

[muted night]

Nah it’s saved. Just curious if it’s identified as that only or if it changes the bios settings

[zenith palm]

Ehh idk I never even realised there was a thing to show timings in memtest86

[muted night]

I think it’s just identified as the xmp cause when I go back to bios after it still shows 3800 CL 14-14-14-28

[modern walrus]

What is the "IGP Ratio?" I cannot figure out what that acronym stands for...

unless it's Integrated Graphics Processor?

[muted night]

Yes

[modern walrus]

well no point in wasting time on that. Thank yo

you

Without getting into timings & all the other crazy RAM OC stuff, with XMP enabled, can you set the speed higher than the rated speed? i.e. if my RAM is meant for 3200, will setting it to a higher value increase performance or is that a bad idea?

[proven canopy]

bruh what do you think overclocking is

[modern walrus]

haha well I didn't know if I could increase that value without messing with timings

[harsh mist]

why does my 5600x lock at 4.2ghz

[proven canopy]

[modern walrus]

so if you increase RAM frequency, CAS increases?

[dull ginkgo]

depends

there is no simple rule

[modern walrus]

hmm okah

I set my 3200 RAM to 3400 & CAS is supposed to be 16 but HWiNFO shows Tcas as 19 now

I didnt' check it before tbh

I guess I should to get a baseline

What's a good testing thing for RAM? I was using OCCT Memory

Stock voltage is 1.35. When I increase frequency should I also increase voltage or leave it be?

[dull ginkgo]

testmem5 is good

voltage depends on the die

different die can take different voltages

[modern walrus]

When I emailed g.skill they told me leave it at 1.35

but of course they're not going to make themselves liable if I screw it up going over

at 3200mHz Tcas is down to 16

so it increases by 3 when the frequency is bumped from 3200 to 3400

Trcd jumps from 18 to 19; Trp 18 to 19; Tras 38 to 43; Trc 56-62; Trfc stays constant at 560. I have no clue what I'm doing lol

All I know is more mHz = good but higher CAS = bad

& I'm not sure what the trade off is

Time for some misguidance on youtube I guess

[dull ginkgo]

Well, that's just auto settings doing auto things, you can manually tone timings

[modern walrus]

I see. Is step 1: raise mHz until crash; Step 2: reduce a little after crash; then step 3: tune each until crash?

something to that effect

actually I'm not even gonna ask you how to OC RAM. I'll go see what nonsense YouTube can tell me

but is the goal to get mhZ as high as possible while still hitting stock values or lower for the other figures?

Jeez 3300 gets the same values as 3400. strange

Do you consider any sources on YT dependable? is this JayzTwoCents any use?

Does tCWL = tCL?

[dull ginkgo]

No, they're different timings, and well, ram is complicated

Arshia or alatron know a lot more about ram oc

[modern walrus]

The video says reduce tCL, tRCD, tRP, & tRAS first but I only have tRCD, tRP, & tRAS

[dull ginkgo]

If you're just messing around with primaries (tCL tRCDRD/tRCDWR tRP tRAS), there might be good sources on youtube, but secondaries and tert timings are weird

Oh yea I forgot about this, take a look at this
https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md

[modern walrus]

okay I'll stay out of advanced timing control & stick to standard timing control.

Thanks I'll look through this

"an example being tCWL = tCL - 1"

[modern walrus]

how high of an increase in frequency is typical?

[proven canopy]

yes

[modern walrus]

sounds like fun

it says start by setting (tCL-tRCD-tRP-tRAS. I have 0 of those on my asus b550-f

[tall pelican]

because asus is dumb and spells them out

[south sky]

fun fact, tCWL has no relation to CL at all

Timing tCWL based off CL is stupid

[modern walrus]

Apparently these are the corresponding names

you guys are missing it. I'm over here having a WHEA party by myself

What's the likelihood of actually reducing CAS latency?

How about setting stock timings then increasing frequency? I can't drop stock timings without WHEA errors for days

I think I had 11 before X1 minimized itself on startup

holy crap I ran sfc & got 102 of them. This isn't worth it lol

[dull ginkgo]

Ram oc is hell

[lavish tundra]

Yes

Also install windows on a spare ssd or hdd for ram oc

boot off that

[modern walrus]

Ya I'm gonna try to increase frequency above XMP, match FCLK accordingly, & keep stock timings & call it a day with the 5900x

[lavish tundra]

ram oc can completely corrupt windows

[modern walrus]

so far Intel (as usual) seems way easier

[lavish tundra]

to the point where you can wake up to have no windows installed anymore

[modern walrus]

I have everything on OneDrive & an external SSD with an image saved on it

[lavish tundra]

still irritating to restore the image when it can easily be avoided

[modern walrus]

true... ya I guess I'll hit up partitions soon & add one just for this

[lavish tundra]

not partitions

a different physical drive

It can corrup partitions as well

[modern walrus]

oh. Can I use an external SSD for that or no? probably not huh?

[lavish tundra]

Has to be internal

[proven canopy]

https://actually-hardcore-overclocking.creator-spring.com/listing/RAMTrainingPoster?product=623

[modern walrus]

then I need to clear my other NVMe

[dull ginkgo]

Yup

I assume those are post codes I'm too poor to see

[modern walrus]

wow. 33mHz increase with stock timings is still throwing WHEA errors like crazy

I'm going back to XMP on the 5900x.

fack this

[dull ginkgo]

I saw the shirt, I mean I don't have a post code display on my board lol

[lavish tundra]

Ahhhh

I dont either

or diagnostic leds

[modern walrus]

how to increase your RAM using overclock: buy better RAM

[dull ginkgo]

Yes

Be like me, try to tighten primaries, give up for half a year, try again, fail, and now giving up on primaries and just tuning secondaries and tert once in a while

[modern walrus]

ya I can see this taking 3 weeks

[dull ginkgo]

Bleach took about an entire week iirc for his oc

But bleach still isn't done afaik

[modern walrus]

I'm doing some DISM now

[lavish tundra]

run that as admin

[modern walrus]

somehow the 10900K RAM (3200) is booting fine at 3600

[lavish tundra]

and thank sippy

[modern walrus]

I'm doing both rn

well sfc finished so ya dism

sfc; dism; sfc

[lavish tundra]

just rightclick -> run as admin on the bat file

its way easier

[modern walrus]

oh shoot I've been wanting that

It's easy on Linux to create a command that runs on startup & stuff like that. I didn't really try to figure out how to do it on Windows yet

[lavish tundra]

Sippy made one

[lavish tundra]

there it is

[modern walrus]

I'm saving that rn

Windows Defender was not happy about that one lol

[lavish tundra]

never is the 1st time

[modern walrus]

I wonder why the 10900K hasn't crashed at 3800 yet

let's try higher

jk 3600

looks like 3700 is our winner

[lavish tundra]

ram timing?

no crash =/= stable

[modern walrus]

16-20-20-40

tCWL 16

[lavish tundra]

1.5hr long test needed to ensure stability

every time u change a timing

[modern walrus]

Now I probably have to clear heckin CMOS

oh wow BIOS came up

The thing is the thing I'm reading said start w/ 1.4v & 16-20-20-40. I think stock the timings are 16-18-18-[I forget]

38

Soo... if my timings go up 2 on 3/4 numbers but my frequency increases to 200mHz

Did I improve or get worse? lol

increases 200mHz rather

I guess I'll try to lower those 2 numbers back down

trying to lower CAS latency seems like a fool's errand

[lavish tundra]

Dont change stuff and say this runs

test it all

[modern walrus]

Well by "runs" I mean POSTs & doesn't have immediate WHEA errors

[lavish tundra]

If you change too many things and it errors you have no way of figuring out which timing caused it

[modern walrus]

I see

[lavish tundra]

Thats beyond stupid

[modern walrus]

I don't actually think that means I'm done & things went well don't worry

[lavish tundra]

I have had errors show up 1 hour into a very intensive stress test

[modern walrus]

but serious question tho, is a +200mhz increase at the cost of a +2 on 3/4 figures an accomplishment or a failure?

Like I don't know exactly what the end game is

I assume it's get lowest figures with highest frequency

but I dunno which is more important

[lavish tundra]

depends on how much worse the other values got

and if they are even stable there

might have to bump em up or down to get stability

[modern walrus]

Like is 3200 16-18-18-38 better or worse than 3600 16-20-20-40?

[lavish tundra]

no

3600 18-20-20-40 still beats 3200 16-18-18-38

btw what die do you have?

if it was a stick of 3200 16-18-18-38 its probably c die

which doesnt do well above 1.35V

[modern walrus]

I gotta check if it's samsung or what on G.Skill TridentZ 3200C16D-32GTRZ

[lavish tundra]

Do that first

[modern walrus]

probably given the name

[lavish tundra]

take a pic of the label and send it here

like the label on the stick

[modern walrus]

okay hold on

I thought I had a picture. I'll shut down the computer & take it out

[dull ginkgo]

When did you buy that kit?

Oh wait

Hmmm

Thought it might've been too old, but looks like it might be too new for data from what I'm looking at

[modern walrus]

5/31/21

so ya pretty recently lol

[dull ginkgo]

"and some sticks start with L42, A42 or 043 for an unknown reason."

Damn it

[lavish tundra]

yep 043 got me too

[dull ginkgo]

But uh, assuming the rest of the naming is correct

That might be hynix afr

Which would be oof

[modern walrus]

ya it's hynix

[lavish tundra]

very oof

[modern walrus]

lol don't like ooof

Is XMP 2 not available for 10900K?

[lavish tundra]

There is no xmp 2 profile on the ram

just 1 xmp profile

[modern walrus]

but it says on the RAM

XMP 2 ready

[lavish tundra]

XMP 2.0

not 2 xmp profiles

[modern walrus]

oh

well that's no good

so explain the oof?

hynix sucks?

what's max voltage you should use with RAM?

[dull ginkgo]

Not hynix sucks, but afr sucks afaik

[modern walrus]

um... I have no idea what that 2d part means

[lavish tundra]

afr is the name of the memory module die

like the chips

on the pcb

that do the storing

[modern walrus]

oh okay

so it's only partially trash

[dull ginkgo]

Well wait

Looking at it again, looks to possibly be apparently 16 bit ajr?

No overclocking reports at this time but its presence in more potent G.skill Ripjaws and TridentZ bins like 3600 18-22-22 and 3600 19-20-20 (codes S820A, S821A) seems quite telling.

[lavish tundra]

still not much better tbh

or is it?

[modern walrus]

I have no idea

[lavish tundra]

asking mr1111

[dull ginkgo]

It says no overclocking reports at this time sooooo

It's very new it seems

[modern walrus]

is 1.5 safe voltage?

Intel says don't go over 1.65

like my other RAM is 3800 & G.Skill said 1.4 was highest...

I hate being so ignorant all the time lol

but increase voltage doesn't necessarily increase stability right?

[dull ginkgo]

Voltage depends on the die, and well, we don't have too much data on the specific die if it is ajr

@tall pelican you think it's ajr or afr? And do you know anything about ajr?

[tall pelican]

ajr is pretty much just 16gbit djr

[dull ginkgo]

Sooo 1.65v gonna be fine with it if it is ajr?

[tall pelican]

I personally would stop at like 1.45 since it just clocks high and does similar timings, since I dont know what lithography they're on

[modern walrus]

What about this one?

[tall pelican]

and afr depends on 4gbit vs 8gbit

that's mjr

[modern walrus]

Wow you really know everything

[tall pelican]

not really, just know how to read

[dull ginkgo]

Now that I look at the first one again, it has to be ajr since it's 16gbit

[modern walrus]

I think I know how to read but I definitely dont know what to read. There's tons of useless nonsense out there from what I've experienced...

Meaning?

Where do you get your info btw

[dull ginkgo]

Should try 1.45v like fitz says, there's still hope, djr isn't that bad

[modern walrus]

What about the 2d one?

[dull ginkgo]

Currently just cross referencing the overclocking wiki for the data on the dies

[modern walrus]

I guess messing with both of these probably isn't a good idea anyway

[tall pelican]

the 042 code tell you

[tall pelican]

this is 8gbit afr

[dull ginkgo]

Is it?

Wouldn't the S mean 16gbit?

[tall pelican]

afaik no

but let me double check

[dull ginkgo]

Hmmm

X standing for the die density: 4 for 4Gbit, 8 for 8Gbit, S for 16Gbit;

[tall pelican]

I might be getting corsair and gskill mixed up

yeah, the S is 16gbit, so it is ajr

[south sky]

AJR is apparently 18nm alius (same as CJR and JJR)

according to thaiphoon

[tall pelican]

and ajr is on?

[south sky]

AJR*

Auto corrected it haha

[tall pelican]

so most likely stick with 1.45 on ajr?

[south sky]

yeah, unless if there are xmp kits with higher

[modern walrus]

So 1.45 for the 3200

This 3800 doesn't appear to have any room for oc...

if the WHEA errors are cpu bus/interconnect errors, does that indicate anything specific?

[south sky]

the interconnect on ryzen is the infinity fabric

[modern walrus]

hmm okay cuz I set it to half of the RAM frequency

[modern walrus]

50 minutes 16-20-20-40 at 3500mHz & no errors on HWiNFO doing OCCT Memory at 85% usage

[lavish tundra]

OCCT isnt great

TM5 extreme anta better

[zenith palm]

Is it overheating, the board good enough ?and turn on pbo too

[modern walrus]

testmem5? Mr1111 said to use that one

Is that the same as MemTestHelper?

How's MemTest64?

[zenith palm]

Testmem5 and you want to run the extreme anta77 config

[modern walrus]

Any chance you have a link? I kept going in circles

[zenith palm]

@modern walrus https://www.overclock.net/threads/memory-testing-with-testmem5-tm5-with-custom-configs.1751608/

[modern walrus]

Thank you very much

[zenith palm]

Np

[modern walrus]

Load config & exit>Anta777 Extreme>Run as Admin>Watch the fun?

(from notepad file)

My fans are very confused. They know things are going crazy but the CPU isn't that hot

keeps saying, "to enable AWE, you must run with administrator" but I did... is that just a PSA or is it not recognizing it's being run as admin?

[zenith palm]

You don't actually need to run as admin, but yes, the next time you just need to open normally, the config will load the same

Wdym AWE?

[lavish tundra]

Its fine

run it as not admin

[modern walrus]

Oh crap well it's been running as admin for like 30 minutes now

[lavish tundra]

thats fine too

if running it as admin errors it out run it as not admin

[modern walrus]

okay will do

[lavish tundra]

errors it out as in at launvch

if you see an error after it launches thats mem oc

like by launch error I mean it literally doesnt allow you to start the app

popups and then quits the app

[modern walrus]

Ah okay

So far so good I think

How long? Like 1.5 hours minimum?

[lavish tundra]

itll tell you when its completed

it pulls up a popup saying testing complete no errors found in extremely bad english

if it finds no errors

[modern walrus]

okay awesome

Thanks

[modern walrus]

jeez 2 hours 20 minutes still going

[orchid flame]

That doesn’t make a lot of sense…huh

I woulda thought it’d be on DJR node

Given how it behaves, it’s tolerances and lack of wall

[modern walrus]

XMP on my 5900x puts command rate at 1T. Is that likely why I can't get past the rated 3800mHz? I've read 1T is better than 2T as far as command rate goes tho so probably better off leaving it?

[zenith palm]

Getting over 3800mhz depends on both your ram bin /die and your cpu bin as well, you could be stuck at 1900mhz fclk

[modern walrus]

oh okay. The RAM is 3800 out of the box. FCLK can go up to 2k I think. I have it at 1900 to match the 3800

but even 3866 with FCLK at 1933 throws WHEAs

It's like there's no OC headroom at all

even at 1.45v

Which out of the box it's rated at 1.4

RAM is freaking complicated. Almost as annoying as Curve Enhancer for 5900x

[south sky]

considering S-Die vs B-Die it makes sense

They can be extremely different

I thought it would be too but I haven't seen anything pointing to it being 17nm

[steel zephyr]

I'm back to 11th place on the Port Royal hall of fame. I updated to a BIOS on the graphics cards that supports resizable BAR and got a small gain. https://www.3dmark.com/pr/1119823

https://www.3dmark.com/hall-of-fame-2/port+royal+3dmark+score+performance+preset/version+1.0

Its difficult because I have one 3090 that doesn't overclock well with the GPU, but the memory overclocks well. The other 3090 has really awful memory that starts crashing and artifacting with a +900 MHz memory overlock with ambient cooling, but the GPU overclocks much higher than the card with good memory overclocks. I'm envious of all these people who I see getting +1400 MHz to +1600 MHz memory overclocks on Kingpin 3090s with ambient cooling

GPU cooling is a 280mm radiator submerged in a cooler with water and ice.

[zenith palm]

Ngl, noone uses that, use timespy to get a more indicative result

[sterile flame]

ur telling a high who regular gets high scores on almost everything?

[steel zephyr]

I'm also 1th on the 3DMark Hall of Fame for TImeSpy Extreme graphics score. https://www.3dmark.com/hall-of-fame-2/timespy+graphics+score+extreme+preset/version+1.0

[sterile flame]

hhahahahhahahha

[dull ginkgo]

Port royal is fine, but that's more about RT cores

[zenith palm]

Just saying lol, the others aren't used a huge amount

[zenith palm]

Ohh nice

[steel zephyr]

24th overall on Time Spy Extreme Hall of Fame. I don't have a W3175X or 3990X so I can't get close to top 10. https://www.3dmark.com/hall-of-fame-2/timespy+3dmark+score+extreme+preset/version+1.0

[dull ginkgo]

Couldn't find you in 3dmark hof, but 4th on hwbot ain't bad at all

Considering top two were ln2

I like the setup tho, big ol bucket of ice

[steel zephyr]

its cheap and easy for cooling. However it has its limitations.

[dull ginkgo]

Yea, can't really do sub ambient

[steel zephyr]

It looks like I'm still 3rd on HWBOT. I'm thinking about trying a glycol chiller or dry ice next time.

[dull ginkgo]

You can probably use a similar setup for dry ice

[steel zephyr]

I was thinking about something similar to the Jayz2cents method with dry ice. However using different fluids so they don't freeze so easily.

[tight plinth]

OCCT isnt great on Intel, on AMD it usually works nicely

[lavish tundra]

Not for ram errors

OCCT passed OC error in TM5

[tight plinth]

There are always some exceptions. That's why you use multiple tests, not just one

[zenith palm]

putting salt in the water, might help as it lowers the freezing point of water btw

[tight plinth]

be careful with saltwater tho, metals usually dont like saltwater

[steel zephyr]

the water won't get any colder than the ice.

I was thinking about Propylene glycol and dry ice next

[zenith palm]

Ohh right forgot that

Or you could bite the bullet for an ln2 pot lol

[dull ginkgo]

Too much and annoying

[steel zephyr]

three LN2 pots. 💸

[dull ginkgo]

Also can't exactly daily on ln2 pots

Have to uninstall them for daily use

[tight plinth]

[insert LN2 loop here]

whatever that thing was called

[dull ginkgo]

[Insert KINPIN GIVE US ROBOCLOCKER NOW]

[steel zephyr]

With my current ice and water setup I can easily switch it back to a daily setup.

[dull ginkgo]

Ya

Take rad out, put it somewhere, done.

[steel zephyr]

Switching to Propylene glycol and dry ice just means draining and refilling the part of the loop the GPUs are in. Same with a glycol chiller.

[zenith palm]

Can't exactly daily an ice bath either lol

[tight plinth]

Get an ice machine

put it over the cooler, get some drain thing if water overflows

[zenith palm]

Big brain

[steel zephyr]

I know someone who dailys a glycol chiller. He has it hooked up to a larger Yeti cooler and has fittings for the GPUs and CPU lines. He sets the temp to the just above dew point when not overclocking.

[tight plinth]

What mem die do you have?

getting some good mem OC can push you a bit higher

[dull ginkgo]

You need to remove the pot and reinstall cooler with ln2, with ice, you can literally just daily it and not fill it up with ice lol

[zenith palm]

Yeye i saw that after i responded lol

[steel zephyr]

I'm using underclocked Samsung B die. Its CAS16 4000 MHz Trident Z neo at 3800 MHz. With the slight speed drop I can't improve the timings that much.

[dull ginkgo]

Alatron and arshia could probably change that xD

[tight plinth]

put a fan over it and OC it

bdie is sensitive to temp, scales with voltage, and is resistant to voltage.

[steel zephyr]

My 5950X isn't completely stable above 3800 MHz memory speed. I have a 10900K CPU and motherboard on the way with 5033 MHz RAM.

[tight plinth]

Ye, but you can tighten timings

[zenith palm]

What timings did you try?

[dull ginkgo]

You can do 1.5v for b die

And just tighten the timings

[tight plinth]

1.55v can be daily-ed with good cooling

[steel zephyr]

I think I tried 15-15-15-34

[dull ginkgo]

What bios version you on?

[tight plinth]

gdm off or on?

[zenith palm]

^

[dull ginkgo]

Agesa 1.2.0 helps a bit with getting fclock to 2000mhz

[steel zephyr]

Latest BIOS from MSI for the MEG Ace x570.

[dull ginkgo]

Alright

[zenith palm]

Not beta bios tho right?

[steel zephyr]

I get bluescreens at 2000 fclk

[dull ginkgo]

Mhm, still no guarantees, but bit more people can do it

[steel zephyr]

no beta BIOS

[tight plinth]

Also, do note some boards have weird fclk holes

[zenith palm]

You should be able to get 1900fclk with nice timimgs tho

[tight plinth]

I think it might be Asus only, but idk

[dull ginkgo]

B die can be worth actively cooling for daily

But well, you should be able to do a lot more

[steel zephyr]

I figured out I can probably mount a 120mm fan blowing directly on the RAM with a little bracket.

I'd need to mount two pieces of 90 degree angle metal to my top radiator.

[tight plinth]

If you're benching, make sure it's one that can move a lot of air. Might as well not care about noise and do max RPM

[dull ginkgo]

Is the xmp 4000 16-16-16 at 1.4v or 16-18-18 at 1.45v? Could check some predictions based off xmp

[steel zephyr]

its the XMP 4000 16-16-16-36 1.40v version.

[dull ginkgo]

Theoretically, at 1.5v, it should be able to do 3800c14, although the rest of the primaries and secondaries and such will be a bit wacky

[steel zephyr]

I don't have a lot of experience with RAM overclocking. Its been very time consuming so far to make sure its stable.

[dull ginkgo]

It's time consuming and weird and hard from what I've seen...

[zenith palm]

Yeah but it can see a big boost especially on ryzen

[dull ginkgo]

Oh right, @steel zephyr what's your soc voltage at?

[tight plinth]

Before venturing into RAM OC, def make sure everything is 100% stable. It isnt fun when you failed a stability test with checking a bunch of RAM related settings trying to understand what went wrong when it wasnt even a RAM related thing

[dull ginkgo]

Mhm

[steel zephyr]

I tried SOC up to 1.2 but didn't see an improvement in stability over the "auto" setting.

[dull ginkgo]

Yea there isn't much gain beyond 1.15v

[tight plinth]

dont set to auto, ideally keep near 1.15v or lower

[dull ginkgo]

But it's good to set it to 1.15v

[tight plinth]

Dont want more heat if it wont do anything

[dull ginkgo]

Soc voltage can help you get fclk up, could test 2000mhz for that again

(if you haven't tested already with manually set soc voltage)

[short blade]

I run rev. e at 3800c14 with 1.105 vsoc

[tight plinth]

do vddp/vddg help with fclk? Cant remember now

[short blade]

vddp can help up to 1.1

lowering vddg iod can help with stability

vddg ccd I'm not sure

[tight plinth]

I know those ones are whack cuz sometimes lowering can help with stability

must be very temp sensitive or smth

or sometimes can really dislike voltage

[short blade]

but yeah up to 1.1 vddp can help stabilize higher fclks

[steel zephyr]

I tried 1.1 vddp and 1.15 SOC with the RAM at 4000 MHz XMP. So far its been stable for 15 minutes in AIDA 64 stress test and Intel burn test maximum setting. I'm going to try prime95 next.

[tight plinth]

generally would do TestMem5 with extreme Anta config and OCCT

or medium if you're just benching

longer runs are more for stability at max temps

[steel zephyr]

Prime95 crashed after a little while.

One thing I noticed is Intel Burn Test seems to crash or report errors faster with unstable memory overclocks than TestMem5.

[proven canopy]

Don't use PG for anything XOC. If you're looking for a heat transfer fluid between dry ice and a rad, use 100% iso, acetone or kerosene work - but can dissolve stuff you don't want dissolved.

If you want a sub-0 coolant, use a 60/40 water/ethylene glycol mix. Green prestone works well.

If the jug doesn't say what's in it, check the msds.

Also I'd probably still keep vsoc set manually to 1.15 or so in case the motherboard has it's own idea of what's safe under auto.

[steel zephyr]

The problem is I'd need something that won't freeze when there is dry ice in it and also won't damage water blocks. Automotive antifreeze doesn't do great below -30. I thought acetone wasn't safe for water blocks.

[proven canopy]

Acetone will destroy almost any rubber/plastic, yeah.

Automotive antifreeze - the eth/diethylene glycol stuff can be good lower than -30, depends on the ratio with water.

https://cdn.discordapp.com/attachments/812875858544820248/862398099419168788/20210707_192152.jpg

Not my pic, someone in another OC server built a phase change chiller, using glycol/water mix

[steel zephyr]

What's wrong with food grade Propylene glycol? I know a few people using it in their glycol chiller setups for overclocking.

[proven canopy]

It will turn to slush when it gets cold

Marine/RV antifreeze is often PG based, it's use case is just to prevent pipes from bursting. Terrible if you want any kind of fluid flow sub 0.

[steel zephyr]

Propylene glycol has a freezing point of -60C according to the manufacturer.

[proven canopy]

Yes, and I'm saying a PG/water mix will turn prohibitively viscous and slushy to be pumped around at any useful flow rate well before that point.

[vague gull]

cherry slurpee juice

[proven canopy]

It's josh's diy phase chiller from borg if you remember him

[nocturne pecan]

I just up graded my computer with an Aorus Z590 Elite Ax motherboard with a Intel I7-10700k processor. Gigabyte claims I can overclock this board up to 5.1 ghz with little effort. Can anyone send me a step by step link on how to do this successfully??

[tall pelican]

medium or high llc, voltage to what you can cool, and see how high it'll clock

the motherboard wont have any issues, its all up to the cpu and cooling

[sterile flame]

is overclocking a 2060 fe worth it or better stock?

[tall pelican]

do you need 5-10% more performance?

[nocturne pecan]

Ok thanks

[modern walrus]

Christmas in July miracle guys

or is +300mHz nothing to write home about? I still know nothing about htis

[zenith palm]

+300mhz ram oc?

[modern walrus]

on a more negative not tho, I got my custom loop liquid & I have no idea how I'm gonna drain what's there

yes

3200 to 3500

survived anta777

[zenith palm]

Without touching timings?

I would try an hour of occt as well btw

[modern walrus]

I had to go from 16-18-18-38 to 16-20-20-40

[zenith palm]

Nice

What voltage?/die

[modern walrus]

That's good? lol I have no clue bro I swear

[zenith palm]

Depends on the die tbh

But ye

[modern walrus]

1.4 & um I dunno it was kinda up for debate this morning

[zenith palm]

And the voltage

[modern walrus]

1.4 is the voltage

[modern walrus]

this

[zenith palm]

I see, that's not very safe for some dies btw

Ohh if it's g skill we can check sticker num

[modern walrus]

stock is 1.35 from G.Skill

ya but it was unclear apparently

cuz it's really new. I got it 05/31 & it was produced 4/21

[zenith palm]

@modern walrus it's hynix a die I'm pretty sure

I have no idea about it tho

[modern walrus]

I know that much from BIOS

but as far as what a hynix die is I have no idea

I'm the blind squirrel who found a nut rn I guess

meanwhile my other G.Skill memory has zero headroom

[zenith palm]

Hynix is the company instead of samsung, crucial etc

[modern walrus]

I couldn't even get it +33mHz (with FCLK adjusted accordingly)

I figured Samsung would be the best

[dull ginkgo]

16 gbit ajr from what we looked at last night iirc

[modern walrus]

ah okay. I got the sense it was inconclusive

[zenith palm]

Ohh, ajr is hynix a die or?

[dull ginkgo]

I was stuck between it being either 8gbit afr and 16gbit ajr but forgot the S meant 16gbit

It's hynix

[modern walrus]

Does that mean I'm overvolting it & it will melt or something?

[zenith palm]

Iirc ajr is pretty decent so you are overvolting but should be fine at 1.4v, 1.5v if it has a fan

But don't quote me on that

[modern walrus]

Is it true more voltage doesn't necessarily mean more stability?

I'm tempted to just leave it as it & not even mess with the timings

I guess I'll mess with them. I only benchmarked at 89%

wait no, if 1,991 users benched at 87%, that's overall. I'm in the 60th percentile. That's trash

[zenith palm]

This is very much die dependant

@modern walrus i would at least tighten tfaw even if you don't touch any other secondaries

[modern walrus]

okay & primaries?

[zenith palm]

If you could lower to 16 18 18 38 or 16 16 16 36/8 if possible

[modern walrus]

I'll try it out rn

same voltage?

(as a suggestion not you're liable if something happens)

Let me rephrase: would you add more voltage?

[zenith palm]

I mean, try one timing at a time, but see if it's stable at same voltage, no need adding unnecessary heat

[modern walrus]

okay will do

Do I really have to run 1.5+ hours of stability test every time I change a number?

[zenith palm]

Unfortunately

[modern walrus]

lol this is brutal

[zenith palm]

I mean you can try them all at once, but if fails you won't know which one you gotta tune back

[modern walrus]

Ya I see what you mean

I guess in the meantime I can try to figure out how the heck I'm getting all the water out of this loop

Might just try to connect a shop vac & run it for like one second at a time lol

[zenith palm]

Yes lol

[zenith palm]

No idea what the vac is for tho?

[modern walrus]

to try to pull all the water out of the loop real quick lol

[proven canopy]

Blow it out with compressed air

[modern walrus]

unfortunately I don't have a compressor

[orchid flame]

Forks 🔫 https://docs.google.com/spreadsheets/d/1q2JgzpUt3qHzAnGXHeSKY0_Jlh4d-k5vRbplVPb-V1Q/edit?usp=sharing

gimme XMP bins to add

or manual OC
just gimme gimme

[tall pelican]

the only xmp worth adding, 3200 16-18-18-38

[proven canopy]

https://www.twitch.tv/antitux
It's Frosty Friday!! Let's get our #1 scores back! - LN2 Extreme Overclocking

[orchid flame]

I’ll add 3200 14-14-14, 15-15-15, 16-18-18
3600 14-14-14, 14-15-15, 15-15-15, 16-16–16, 16-18-18, 16-19-19 and 18-22-22

[modern walrus]

I changed the liquid but went too heavy on the red part of the mixture

[orchid flame]

Few others for higher clocks

[left blade]

Arshia#2457 has been warned

Reason: Bad word usage

[orchid flame]

Average stuff

And then I’ll add the DDR3 stuff

And the 8400 DDR5

After that, move into manual OCs instead of XMPs

[south sky]

yuck

those calculations don't make sense

doesn't make any sense

neither does first bit access

and some other things

[orchid flame]

4000/freq?

why?

[south sky]

that's the calculation for that other latency

[orchid flame]

based on? or is it just an estimate

[south sky]

I'm looking at yours

[orchid flame]

oh that number I set to different things for DDR3, 4 and 5

[south sky]

oh I see

[orchid flame]

just an estimate

cause theres no all encompassing

[south sky]

anyways that's what happens in a idle to idle single read burst

tCCD is between bursts not bits

[orchid flame]

tCCD is burst to burst delay?

[south sky]

essentially for the same type

[orchid flame]

so a single command does the whole burst?

[south sky]

cas to cas command delay

yes

[orchid flame]

so finding the first column happens during tCL/tAA?

[left blade]

alatron978#7416 has been warned

Reason: Bad word usage

[south sky]

as soon as the read command is addressed the data is moved out of the memory into the prefetch buffer over the next internal clock

[orchid flame]

right but it needs to select the column first.

that cant possibly be instantaneous

[south sky]

Hence why tRTP and tAA can't go below 4 on ddr4

It picks the column then it starts transferring the data

[orchid flame]

and then moves down columns in the burst

[south sky]

it has an entire physical clock cycle to do it

and 3200mt/s ddr4 only runs at 400mhz so it's a long time

[orchid flame]

ok so 8 bits of data

transferred over 4 clocks

wait no

[south sky]

1 clock

[orchid flame]

thats 7 clocks

it takes 7 whole ticks to do 16 bits of data out

[south sky]

Time breaks

[orchid flame]

so 8 clocks

so 2 bits/clk

1 in rising clock and 1 in falling clock

[south sky]

per IO bus clock yeah because it's DDR

But the internal bus transfers the data over 2 clocks in that case

[orchid flame]

do you mean 2 bits/ 2clk or 16 bits/2clk

it'd be 2/2 right?

[south sky]

16bits in 2 clocks

8 bits per clock

[orchid flame]

...how
how many transmission lines are there

how would 8 bits of data be transferred in the singular clock

it cant be over a single line
or 2

[south sky]

because the internal memory is 8 times wider then the external memory bus

I'm pretty sure I went into it in the guide

you can read it there if you want

[orchid flame]

no but they need some level of delay

no?

[south sky]

no because it happens in 1 clock

so the data has to already have been transferred

and ready

[orchid flame]

so literally just tRC?

[south sky]

depends

[orchid flame]

either than or tRAS.

[south sky]

those 3 things can limit it

[orchid flame]

tRC< tRAS and what?

[south sky]

well 4 things

tRC
tRAS + tRP
tRCD + tRTP + tRP
tAA + BL

[orchid flame]

read operation

for write its slightly diff

[south sky]

yes

[orchid flame]

ok, so I just run an if statement with all 4 of those things combined finding the largest?

[south sky]

but you are assuming read anyways

for read yeah

[orchid flame]

and that should be the limiting factor

[south sky]

however you can have them overlap if tAA+BL goes longer

[orchid flame]

this is odd, I've always unserstood it as find row, find column, where first bit will be located, as that data is being grabbed, next bit, next bit, so on

[south sky]

just puts it out of phase I guess in that case

when it finds the column all the data is there

then it gets moved into the prefetch buffers

[orchid flame]

right

ok interesting

@south sky wait

total latency wouldnt be just

tRC
tRAS + tRP
tRCD + tRTP + tRP
tAA + BL

whichever is higher

it'd be all of those +tCCD

no?

wait so why does tCCD have such a large impact on BW

and more importantly, no impact on latency

[south sky]

Just a sec have to take out the garbage

[south sky]

yes

[south sky]

no because it's a single burst

[south sky]

it's the gap between bursts, if larger then the burst length you get clocks that do nothing

[south sky]

it doesn't impact latency for a single burst, it does when you are measuring the latency between bursts or the latency for a large amount of data (such as 100MB)

So like tCCD 8 on DDR4 causes half bandwidth since 4 clocks are reads or writes, then 4 do nothing

[orchid flame]

btw what are jedec tRTPs again?

I look it up and all I see is the minimum at 4nCK

[south sky]

minimum for what

spec?

[orchid flame]

yes

[south sky]

yeah

[orchid flame]

right

thats the minimum but will it just default to whatever 7.5ns is?

[south sky]

by spec min is which ever is higher

[orchid flame]

yeah ok perfect

so it'd be 7.5ns

[south sky]

1066 though

;)

[orchid flame]

ah yes

my fave command 🙂

meh
still proves my point

nice

[south sky]

of course

[orchid flame]

wait

but D3>D4

cursed

[south sky]

both 8n

[orchid flame]

true

[south sky]

ddr3+

[orchid flame]

anyways

is it fixed now?

think it is 🤷

[south sky]

it's better

It'd be cool to see you calculate it based off tCCD_S and tCCD_L

[left blade]

alatron978#7416 has been warned

Reason: Bad word usage

[orchid flame]

I can add those

[south sky]

do it

have fun

[orchid flame]

but theres a slight issue

why would I do that

[left blade]

Arshia#2457 has been warned

Reason: Bad word usage

[orchid flame]

how would it be any better?

[south sky]

more realistic

[orchid flame]

oh

wait

you mean for both same bbank and diff bank

i'd need to include BG count

which is a pain

[south sky]

yes

[orchid flame]

oh ffs

[south sky]

and you need to do refreshes for each density

get to work

[orchid flame]

lbjkhivdskjsdvbklsdbvsd

that would be so incredibly conveluded

I like it

[south sky]

ddr3 will be easy it doesn't have bank groups

[orchid flame]

imagine only having banks

😳

wait

[south sky]

that's a good thing

[orchid flame]

so how does it work again with D3?

[south sky]

no bankgroup

just fast

[orchid flame]

so its constantly just tCCD_S

[south sky]

no

tCCD

just tCCD

[orchid flame]

wait

hmm

ok fair enough

[south sky]

_S is for diff bankgroup, ddr3 doesn't have diff bankgroups

[orchid flame]

waht if it wants to access the same bank again

[south sky]

why

[orchid flame]

no reason
but what if it did

[south sky]

tCCD

but yeah, bankgroups only really allow for higher clocks

[orchid flame]

sidenote

bank interleaving

[south sky]

Like ddr1 and ddr2 were 200mhz max, ddr3 was 266mhz max, ddr4 is 400mhz

[orchid flame]

ughhh can I just assume everything has 4BGs?

[south sky]

ddr5 doubles

that's where one of the big efficiency boosts comes in

still not as efficient as ddr3 though

[orchid flame]

i meant D4

[south sky]

screw x16

[orchid flame]

embrace x8

8BG D5 and 4BG D4

hmmm so with bank groups

[south sky]

but yeah doubling the bankgroups with a dumb IMC is like a 7% boost

to bandwidth

[orchid flame]

the probability of accessing same bank group is (3/4)/(BG*4 -1)

since you wont be accessing the same bank

so for that proportion of the time, you'd be running _L timings

and with tCCD_L of 5, 6, 7 or 8 a diff proportion of operational time will be spent on transfer

so with tCCD_L its 80% spent on transfer

correct?

[south sky]

3/15

=20%

banks in same bg/banks

[orchid flame]

yep

(3/4)/(BG*4 -1)

[south sky]

0.75/15?

[orchid flame]

oh im dumb

(3)/(BG*4 -1)

[south sky]

a dumb IMC would still go to the same bank too

so have fun

[orchid flame]

NOOO stop

I beg

[south sky]

doesn't work for ddr5

ddr5 gets another bonus

I can't wait for 24GB sticks

just seems so right

or 12GB sticks but eh x16

[orchid flame]

why doesnt it work? idgaf about 8Gb DDR5?

if I ignore 8Gb DDR5, it works

[south sky]

8Gb ddr5 good

[orchid flame]

cringe

16Gb Master

[south sky]

Listen, Micron will cut 16Gb chips down to 8Gb for XOC

Of course they will

[orchid flame]

sir, Rev A is already pretty pog

[south sky]

heck you could even do it in the SPD

[left blade]

Arshia#2457 has been warned

Reason: Bad word usage

[south sky]

4 bank groups? nope there are 2

[orchid flame]

16Gb Rev A is strong as HELL

[orchid flame]

.... ok so sidenote

based 1024b SPD

next step is gonna be to store doom onto an SPD

[south sky]

no

[orchid flame]

wdym no

[orchid flame]

@south sky it has been done

this is how I factored it into my equation

% of time tCCD_S is used * 1

% of time tCCD_L is used * (tCCDS/tCCD_L)

[south sky]

@orchid flame awesome, I'll explain how the internal stuff works better another day

[orchid flame]

I’ll factor in command rate and some other stuff tmrw

[modern walrus]

when the numbers are like, for example, 16-20-20-39, do the 2d & 3d #s always have to match? i.e. would 16-19-29-39 be possible?

sorry I meant would 16-19-20-39 be possible rather

or perhaps better stated: must tRCD & tRP be the same value?

& another question because driving you all crazy is my role in this discord: Is tightening timing at the cost of an increased command rate a negative or positive change?

[dull ginkgo]

They don't have to be the same

You could have 14-16-12-14-32

And well, it's all tradeoffs

[modern walrus]

Strange cuz when I put 16-19-20-40 in BIOS HWiNFO reads 16-20-20-40

[dull ginkgo]

Iirc GDM is what forces some of the timings to be even

[south sky]

not that timing

[modern walrus]

well Aorus forces them even from what I can tell so far

[forest radish]

They have to be equal on intel

[modern walrus]

oh okay that explains it. Thank you

[zenith palm]

Generally if command rate is high enough (especially when synced on ryzen) looser timings with higher freq can generally be better yes

Sorry i probs repeated myself in that like twice

[modern walrus]

don't worry I need things repeated 3-5x in a variety of ways to understand better lol

What about with Intel? I've kind of given up OC my Ryzen memory because there doesn't seem to be much headroom

[zenith palm]

Same with intel usually afaik, like there comes a point of diminishing returns in timings, eg 3600 16 16 16 36 would still be beaten by 3800 18 18 18 40 or something iirc

[modern walrus]

I see. The XMP on my 5900x memory sets it at 18-22-22-42 at 1T command rate. I was thinking maybe if I drop to 2T command rate I might be able to tighten the other settings more

cuz as it stands, I can't even get close to stable bumping from xmp's 3800 to 3866

[zenith palm]

Ohh sorry i mistook command rate for frequency idk why lol

[dull ginkgo]

You could try gdm, which is basically 1.5t cr, not quite 1t, but a bit better than 2t, but gdm forces some timings to be even

[modern walrus]

& on my intel I dropped from 3500mHz to 3400mHz (XMP is 3200) & tightened from 16-20-20-40 to 16-19-19-38 & it seems stable (XMP is 16-18-18-38) but I don't know if that's better or worse than XMP tbh

oh okay so 2T forces those to be the same? This is on Intel btw

[zenith palm]

Gdm forces some timings to be even, 2T is supposed to be easier for oc iirc

[dull ginkgo]

GDM forces some timings to be even, I believe tCL has to be even, but not sure on the others

[modern walrus]

Lol ya'll learning from the same textbook?

[zenith palm]

Lol

[dull ginkgo]

kinda

[modern walrus]

So dumb noob question no. 872: is 3200 16-18-18-38 "better" than 3400 16-19-19-38?

[zenith palm]

I think i sent mine about 5 secs before mr1111 too lol

[dull ginkgo]

you can use some benchmarks :)

[zenith palm]

3400 with those timings should be better yes

[modern walrus]

Which benchmark targets RAM like that?

[zenith palm]

And cpu benchmark should show a diff

[dull ginkgo]

geekbench 3 has a specific memory section

some here
https://github.com/integralfx/MemTestHelper/blob/master/DDR4 OC Guide.md#benchmarks

[modern walrus]

CB 23?

okay I'll use that one

[dull ginkgo]

tbh cinebench doesn't really care about memory

[modern walrus]

got it got it

I'll do benchmarks once anta777 finishes & I know this is even stable

[zenith palm]

This should be a pinned tbh

[modern walrus]

It says ignore memtest64. I used memtest64 once before Testmem5 & it crashed

[zenith palm]

Memtest64 is fine for stability afaik

[modern walrus]

I have another dumb question

what's afaik?

[dull ginkgo]

it is pinned

[dull ginkgo]

as far as I know

[modern walrus]

I've never looked at these. What the heck is that computer looking thing with the giant black things in it? lol

[zenith palm]

Ohh i thought it was a slightly different guide I'm dumb

[modern walrus]

I think I can do this. I have it all core 5.1 but I could probably get 5.2 with -1 AVX...

I assume both of these stats are with a -1 AVX?

I have it right now at 5.1 all core with 0 avx & ring ratio at 48

thing is the voltage ends up higher than that for sure