building my 1st pc

[topaz solstice]

still fustrated that nvidia didn't just stay with the standard mini fit and not go with nano/micro-fit connectors

matters more for a few things that's no use to you tbh

[arctic lotus]

like what

[topaz solstice]

there was a gen 5 ssd that goes into pcie slot, can find the image any more

found it

enterprise/server drive

like i said, something that's no use to you

[arctic lotus]

those are ssds tho

im talking about psus

[topaz solstice]

12vhpwr/pcie 5.0 power connector

[arctic lotus]

what is the significance of it on a psu

[topaz solstice]

none for the better part, but it makes it easier and nicer when using a nvidia gpu

[arctic lotus]

why easier

[topaz solstice]

in your case, it means not using an adapter but a direct cable that uses the same fittings bothe side of it

[naive robin]

Do you want to route and plug 4 cables in, or just 1

[steel sapphire]

less clutter

[desert mist]

Personally I’d trust the 4 more as many do but. Defo would be less to squeeze in a case with only 1 connector

[topaz solstice]

i'd prefer the corsair one personally if i had to

@arctic lotus

@naive robin just found this

[arctic lotus]

so is nano fit not the best one? ik you were talking about how you didnt like them

[pale tulip]

realistically theyre not that different

its just nitpicking

[topaz solstice]

to mean the minifit was fine, it can handle the higher wattage much better

all they had to do was use a 12 pin (2x6) minifit instead

[arctic lotus]

i like the idea of them getting smaller.

[topaz solstice]

it would have negated the issue the current fitting has

[arctic lotus]

whats the current issue

[topaz solstice]

and the moment that over loads, it melts

[arctic lotus]

(pun not intended)

[topaz solstice]

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

https://www.youtube.com/watch?v=ig2px7ofKhQ

https://www.youtube.com/watch?v=mKqC2wJIeEI

cablemod sent several 4090's from customers, even buying the card from the customers and sending a brand new one out

to northbridgefix, as even cable mod adapters melted due to the 4090

as the cable mod adapters fully connect in and don't pull out

it might only be a small amount of them that has melted it's still a design issue

the melting issue is quite obviously to much wattage going through the cable and connector itself

16 gauge wire + minifit is much safer due to handling a much higher wattage overall

nano/microfit = 12v 8.5a = 102w on each pair of pins

minifit = 12v 13a = 156w on each pair of pins

the nano/microfit hit failure point faster than a minifit

https://videocardz.com/newz/pcie-gen5-12vhpwr-power-connector-to-feature-150w-300w-450w-and-600w-settings

https://www.kitguru.net/components/graphic-cards/joao-silva/pci-sig-reportedly-working-on-a-replacement-for-12vhpwr-cable/

https://www.igorslab.de/en/rest-in-peace-12vhpwr-connector-welcome-12v-2x6-connector/2/

https://www.igorslab.de/en/nvidia-geforce-rtx-4090-and-a-design-problem-on-the-board-12vhpwr-connector-could-clearly-be-cool-pad-mod/ igor's lab offers a potential 'fix' that can offset some of the issue in the design flaw itself

http://jongerow.com/12VHPWR/

the dumbest bit about the 12vhpwr cable, is you can easily modify this bit on the cabe and just ground the pins out

[arctic lotus]

interesting

so then do i not get the 4090

or do i not get atx 3.0 lol

on a side note ive memorized san pan lan can man etc

[runic bloom]

huh that was not the conclusing rauko was getting at

[arctic lotus]

i still dont know what ran or ban is. or nano tho

[topaz solstice]

wouldn't worry about them

[naive robin]

You get a 4090 and an ATX 3.0 PSU

[sharp harness]

So you don’t have to route all of these to the adapter.

[arctic lotus]

are you saying that with pcie 5 on a psu which is just that 12vhpwr or basically atx 3.0

i only have to plug in one of those cords into the psu instead of all of those ones. (ik other parts still have power cords.)

but it allows you to consolidate all those into 1 cord?

all those as in the ones in the video

[sharp harness]

Yes.

What I'm showing you in this video are 3 8-pin PCI-E cables being put into a PCI-E to 12VHPWR adapter.

But with a PSU with native support for 12VHPWR you won't have to do this.

[arctic lotus]

why not just plug the three 8 pin pcie cables instead

instead of plugging them into an adapter that turns them into 12vhpwr

[naive robin]

The graphics card doesn't have 3 8 pins

It has the 1 hpwr connector

[arctic lotus]

so atx 3.0 is only important for graphics cards

[naive robin]

Also for a 4090, it's 4 8 pins

Yes

[arctic lotus]

okay cool

is there a chart for all the atx standards

like how usb standards have a chart

[lusty tiger]

Wdym “all the atx standards”

[arctic lotus]

like atx 2, atx 3

[steel sapphire]

that is a very very deep rabbithole that basically includes standards for every single pc part from cases to mobos to psus

[arctic lotus]

i just mean something like this

but for atx3

[desert mist]

probably not

[pale tulip]

its very irrelevant

honestly man, like every psu at the highest end is realistically the same

[topaz solstice]

no, they'll be separate pdf documents for each atx specifications

[topaz solstice]

which to be fair goes into enough detail it's more for an EE to understand

[arctic lotus]

i will probably throw out all my ram research because of the LTT testing stuff

level1 (the optane guy) showed up in the ltt video today. so that was cool.

[lusty tiger]

What?

[robust finch]

lol trusting ltt research lol

[topaz solstice]

well that would be less than smart

[sharp harness]

👆

[topaz solstice]

any of the LTT "tests" needs to be taken with 3 barge loads of salt

[lusty tiger]

Any tests should be cross checked with other reviewers’

[arctic lotus]

at the time, i couldnt find any good ram benchmarks. i have since found another, but my ram choice is based almost 100% off LTT

[lusty tiger]

why would you look for ram benchmarks

[arctic lotus]

but ive been told to mostly do this. and after watching everything. i think i will take that advice.

[lusty tiger]

am5 basically cant do more than 6000 cl30 anyway

[pale tulip]

anyway, ram speeds dont affect x3d chips as much

so buy whatever

you will still be happy

[arctic lotus]

because ram speed isnt the most important thing

[pale tulip]

take LTT benchmarks with a grain of salt

[arctic lotus]

my cpu and gpu selections were multi sourced.

storage as well

[lusty tiger]

https://www.youtube.com/watch?v=c9JqPNOpLFM&pp=ygUMamF5enR3b2NlbnRz

[topaz solstice]

not without some extra tuning to be had

[lusty tiger]

yeah

[pale tulip]

nah but buying that hynix kit for $80 and tuning it might be fun

[topaz solstice]

their tests are not fully done properly, they've had backlash over it at how rushed they are so they do not give accurate enough a result

and if you want to get anything for ram, gamernexus/ hardwareunboxed/techspot

[topaz solstice]

it would but never expect someone new to pc building to understand tuning ram

[pale tulip]

ok but its like the best place to start having fun

[arctic lotus]

i would say am5's biggest problem is ram

[pale tulip]

meh

[arctic lotus]

honestly, it was outright intentional. but i personally dont care cause i havent made any purchases yet.

[topaz solstice]

eh, not really, you can just use some 6000mhz cl30-32 straight away and have no issues tbh

[arctic lotus]

yeah but intel benefits way more from ram

not just cause the number is bigger

[topaz solstice]

both benefit from ram, especially on timings

[lusty tiger]

am4 definitely benefited more from faster ram, idk about am5

[arctic lotus]

i would say am5 benefits more from timings

[topaz solstice]

7000 series is a bit like 2000 series in regards to ram currently

[arctic lotus]

the difference in 5600, 6000, or 6400 speed is negligible.

[topaz solstice]

you wouldn't use lower than 5600, and that's mainly if it was 4 sticks of ram

[arctic lotus]

yeah i dont think i would use lower.

[topaz solstice]

2 sticks you can get it to do 6000-6400, though amd themself said that 6000mhz is the sweet spot for 7000 series ryzens

they quite clearly stated it as it is

[arctic lotus]

the benches from my other source are showing a lot of lower capacity ram performing better similar to lower capacity storage drives.

so a lot of the 16 GB sticks are doing better than 32 GB sticks

[pale tulip]

2x16 the goat fr

[arctic lotus]

i wonder why

[topaz solstice]

not for ddr5 if you mean 16vs 32 gb kits

[pale tulip]

interesting source..

[topaz solstice]

32gb kit on ddr5 is like doing 16gb kit on ddr4

[arctic lotus]

also i know now that obviously ram has read and write speeds too

but whats the difference?

[lusty tiger]

dont worry about it

[arctic lotus]

what does a ram read speed do thats different from a storage read speed

or is it the same thing but faster

ik the order of caching

[topaz solstice]

though tbh ddr5 really makes a difference when it comes to memory intensive programs

[topaz solstice]

that if i remember correctly is due to the timings

[arctic lotus]

yeah which is why i dont want to go dual 16 GB tbh. i actually see a bigger benefit from going as much as possible.

[pale tulip]

in which program bro 😭

[arctic lotus]

so whats better, faster individual sticks of ram. or higher threaded ram with more capacity.

[lusty tiger]

dream, either get 2x16, or 2x32

[topaz solstice]

won't make any difference apart from compiling for programming AI in your case

[arctic lotus]

i mean, that's what we are discussing.

[pale tulip]

if u run out of ram, more ram
if you dont run out of ram, faster speed

[topaz solstice]

a kit, you never do single stick

[arctic lotus]

okay i understand now.

[topaz solstice]

or combine the 2 and get it to peak

[pale tulip]

get the epyc with 8 sticks of 512gb

[arctic lotus]

more ram is always better unless you dont need it. in which case the speed is better.

in tech its easier to make lower capacity go faster too.

[topaz solstice]

i.e. 64gb kit of 6000 cl30-32 ram

[pale tulip]

yeah this might be the play

[topaz solstice]

that depends on the dies(ic's) that makes up the ram

[arctic lotus]

oh i didnt mean 1 stick. i meant how some sticks seem to prioritize speed over capacity. but that's weird with ram

[pale tulip]

you can have both

[topaz solstice]

you can have both, the ones that prioritize speed have better ic's that OC better to allow for that

[arctic lotus]

sodimms are laptops only right

[topaz solstice]

yes

[arctic lotus]

okay ty

[topaz solstice]

there are adapters but i'd never trust them

[arctic lotus]

also thought it was cool knowing now that ram have notches in different places like m.2s

and pcie cards

[runic bloom]

they look funny

[topaz solstice]

eh, those notches don't work the same as it does for m.2's

[arctic lotus]

yep child ram lol

[topaz solstice]

also risks killing the pc

1 bad adapter = fried ram/cpu

i'd rather not kill the pc the moment after it's built

[pale tulip]

thats the funny thing to do

[topaz solstice]

nope, remotely doing it if the person who had you make the pc, tries to screw you over

doing 4 sticks of ram to make up 64gb or higher capacity, you'll end up running at 5600mhz

meanwhile keeping it to 2 sticks you'll get the full speed

the other downside to doing 4 sticks is that you might only see half the total capacity

with ddr5

[sturdy anchor]

Lmao “STD”

[pale tulip]

be mature

[sturdy anchor]

No 🍞

[topaz solstice]

that memorybenchmark.net one is a clone of userbenchmark @arctic lotus do not use it

really sad that ubm owner cloned their site but added gpu/storage/memory to the front of it

[pale tulip]

https://tenor.com/view/daireb-gif-26213999

[topaz solstice]

they have a big hardon against anything amd related

[arctic lotus]

no its passmark

which has been regarded as a good benchmark site

also no, i love amd.

[arctic lotus]

#1 6400 16GB sodimm❌
this is high threaded but it's for laptops only.
https://www.memorybenchmark.net/ram.php?ram=Gold+Key+Technology+Co+Ltd+NMUD516F81-6400F+16GB&id=20432
#2 5600 16GB✅
high threaded but it's only 16 GB.
https://www.memorybenchmark.net/ram.php?ram=Patriot+Memory+(PDP+Systems)+PSD516G560081+16GB&id=21827
#3 4800 16GB green stick ❌
This is high threaded but it's only ddr5 4800, and 16 GB and its a green stick.
https://www.memorybenchmark.net/ram.php?ram=Kingston+KF568C36-16+16GB&id=20835
#4 5600 2x16GB (32GB) sodimm❌
this is high threaded but its for laptops only.
https://www.memorybenchmark.net/ram.php?ram=SK+Hynix+HMCG88AGBUA084N+32GB&id=22267
#5 #6 ❓
two more ram sticks that could be #5 or #6 but i can't find the speed for them. idk anything about apacer.
https://www.memorybenchmark.net/ram.php?ram=Apacer+Technology+GD2.45312H.002+16GB&id=22652
https://www.memorybenchmark.net/ram.php?ram=Apacer+Technology+GD2.42312H.002+16GB&id=21981
#7 6000 2x16 (32GB) ✅
high threaded. i didnt know thermaltake makes ram. it is either #5 #6 or #7, but it has a speed of ddr5 6000 cl is probably 36 but it's only 2x16GB.
https://www.memorybenchmark.net/ram.php?ram=Thermaltake+Technology+Co+Ltd+RG33D516GX2-6000C36B+16GB&id=22063
#8 6400 2x48 (96GB) ✅
high threaded
https://www.memorybenchmark.net/ram.php?ram=Corsair+CMK96GX5M2B6400C32+48GB&id=22085

https://www.memorybenchmark.net/threaded_ddr5.html

heres the full benchmark

[gentle elm]

We back to ram?

[arctic lotus]

these are all high threaded

[gentle elm]

What all is decided so far

[arctic lotus]

but most of these are not useable for me.

because they are laptop only

or because i dont want 16 GB

[lusty tiger]

32GB
https://www.newegg.com/g-skill-32gb-288-pin-ddr5-sdram/p/N82E16820374377

64GB
https://www.newegg.com/g-skill-64gb-288-pin-ddr5-sdram/p/N82E16820374383

[arctic lotus]

heres where most of the ones that i might be able to use start

[topaz solstice]

i'd go for the ripjaw ones

[lusty tiger]

so this?
https://www.newegg.com/g-skill-32gb-288-pin-ddr5-sdram/p/N82E16820374369

[topaz solstice]

yeah

[arctic lotus]

why ripjaws over trident

[gentle elm]

I prefer tridents

[lusty tiger]

afaik, the only difference is esthetics and, by extension, price

[topaz solstice]

tridents don't perform better than the ripjaws, and the ddr5 ripjaws give the build a classy look

[pale tulip]

flare at least

[topaz solstice]

for 32gb yeah, for 64gb it's ripjaws

[arctic lotus]

i would go trident just for the rgb because i think its pretty

[pale tulip]

ok imo this is useless
if you care about performance you manually tune

[topaz solstice]

if you wan't rgb you can go better and tbh just get some rgb strips instead

[arctic lotus]

also the tridents i have chosen are the same price as the ripjaws equivalent right now

[arctic lotus]

what are rgb strips

[lusty tiger]

what have you chosen

[lusty tiger]

holy shit

[topaz solstice]

https://www.newegg.com/g-skill-64gb/p/N82E16820374468 <- cheaper than tridents

[arctic lotus]

is that for pcs?

[topaz solstice]

https://www.newegg.com/phanteks-ph-neledkt-m5-rgb/p/N82E16811984033

[arctic lotus]

ohh okay

[topaz solstice]

the second one is better due to the diffusion built in

[arctic lotus]

well its not on the ram tho

requires some diy too

[lusty tiger]

generally, but you can use them for whatever the hell you want

[lusty tiger]

you wouldnt buy an rgb strip to put on ram

[topaz solstice]

better lighting via doing it this way as it'll act as accents

it also doesn't result in performance loss due to having rgb built into components

[pale tulip]

should have done that in a lancool 3

[lusty tiger]

....

[pale tulip]

lancool 3 ❤️

[lusty tiger]

esthetics are so subjective

[topaz solstice]

https://www.newegg.com/g-skill-32gb/p/N82E16820374419 <- the flares

basically ripjaws but with expo spec

[lusty tiger]

also cl36 not cl30

[topaz solstice]

you can get it as cl30 still

[arctic lotus]

https://pcpartpicker.com/product/yPBG3C/gskill-64-gb-2-x-32-gb-ddr5-5600-cl28-memory-f5-5600j2834f32gx2-tz5rk

this is what i have selected so far

[lusty tiger]

idk how 5600 cl28 compares to 6000 cl30. im guessing theyre basically identical

in which case, thats a perfectly fine kit

[pale tulip]

eh, kinda

[topaz solstice]

basically a 10ns latency on both but more bandwidth on the 6000 kit

[lusty tiger]

so 6000 cl30 is better

[topaz solstice]

yup

[pale tulip]

its better but like the difference inst the end of the world

[topaz solstice]

this for the better part

it's not like where on ddr4 3600 cl18 = 3200 cl16

i need to find a chart like the one for ddr4 that shows their latencies

[arctic lotus]

i basically made this choice because of LTT's video but then that whole situation came up

with gamernexus

[lusty tiger]

either that or this is fine
https://www.newegg.com/g-skill-64gb-288-pin-ddr5-sdram/p/N82E16820374383

[arctic lotus]

https://www.youtube.com/watch?v=b-WFetQjifc&pp=ygUIYmVzdCByYW0%3D

based off this video i chose the 5600

[topaz solstice]

i've never made a purchase based on anything posted by ltt tbh

[terse pecan]

Lol

[arctic lotus]

well thats why im changing it

at the time it was different

[topaz solstice]

6000 cl30-32

if you get a cl32 kit, they can be found with better timings that can be tuned even more later on

still in the end be a better choice in comparison

especially when amd themself tells you that 6000mhz is their sweet spot currently for 7000 series ryzens

which i did state 2-3 threads earlier tbh

next gen the sweet spot will probably be 6400mhz knowing how it was on the am4 cpu's

3000 series with 3200-3600 mhz, 5000 series with 3200-3800 mhz

[arctic lotus]

in that video the 5600 out performs the 6000

[pale tulip]

uh, LTT testing

[arctic lotus]

and is on par with the 6400

[topaz solstice]

yeah, no

[lusty tiger]

nah just get 6000 cl30. thats the best speed for ryzen 7000

assuming you want 64GB:
with RGB-
https://www.newegg.com/g-skill-64gb-288-pin-ddr5-sdram/p/N82E16820374383

without RGB-
https://www.newegg.com/g-skill-64gb/p/N82E16820374468

[topaz solstice]

https://youtu.be/X7bGhv5rJyU?t=67 basically a 7.5 tonner

[pale tulip]

why is that legal

[topaz solstice]

because it's decomissioned

for £20k i could get a decomissioned scorpion scout tank

[pale tulip]

still weird

[desert mist]

Ram benchmarks don’t make a lot of sense tbh. Cause one ram kit will use Samsung and the next will use SK Hynix. At least some brands use whatever anyway

Well that and the fact am5 doesn’t really run super fast ram stable anyway

JayzTwoCentz made a rant about it but I didn’t really understand why cause am4 never ran super fast ram stable either sooo, especially not if you run quad channel

[topaz solstice]

you'd never be doing quad channel on any consumer setup

though his rant is more from an OC'r point of view

[barren trench]

How is this pc not done yet

[desert mist]

Either or he can’t really be mad it’s known that amd doesn’t do so well with fast ram speeds.

Tbh when Intel starts stacking cache on their chips I believe it’s gonna hurt AMD bad if Intel stays at the same price point per chip

[topaz solstice]

lets be fair, intel has issues at times with faster ram also, but after a certain point it's diminishing returns in regards to performance

with ddr4 it was after 4000mhz

with ddr5, it might be at a much greater speed over 8000mhz

[desert mist]

Some have mentioned ddr6 being around 12000MHz +

The leaks I’ve seen anyway like bruh

Why

[topaz solstice]

well ddr5 already can hit 11,000mhz tbh as it's been shown with XOC

[desert mist]

I mean with xmp profiles

[pale tulip]

pure speculation

might be possible though

[topaz solstice]

we're at what? 8000mhz with xmp currently?

[desert mist]

I’ll probably be on ddr4 for the next 3-4 years anyway so it’s meh to me

[hazy hedge]

Anything purchased yet ?

[gentle elm]

No

We went back to ram

[hazy hedge]

Ok I’ll check back in a week might still be on ram tho

Or back to storage

[gentle elm]

Prob back to storage

[lusty tiger]

Hell nah

[gentle elm]

?

[lusty tiger]

Not back to storage

[gentle elm]

She went back to ram

Next step to back to the start fully

[arctic lotus]

i do. but im okay with 96 gb too

[arctic lotus]

why does xmp also have gens?? i saw xmp 3.0 etc

[desert mist]

Just means it’s improved over the years

[arctic lotus]

im not back to ram, just talking about maybe switching it

since ik a lot of people want me to switch it too

[desert mist]

Just like any other thing that has multiple versions it’s new and improved software and hardware

[arctic lotus]

but it would seem a lot of 5600 modules are outperforming the 6000 modules too on this benchmark too. so that makes 2

[desert mist]

Like I said it’s sorta a lottery lol

[arctic lotus]

what about expo

[arctic lotus]

lottery implies the die

[desert mist]

A lot of brands use a mix chips from various manufacturers

[arctic lotus]

whats the difference between the die and the controller

[desert mist]

No it also implies the types of chips used on the ram

Some use Hynix some use micron some use Samsung

No way to tell until you buy it

[desert mist]

Same scenario, just improves with the version number

[arctic lotus]

that corsair vengence tho has 2x48 modules that outperform the 2x16 and 2x32 modules

[desert mist]

Not all of them will tho

There’s Samsung b die and m die

Corsair uses both

[arctic lotus]

whats the highest gen for xmp and for expo resepectively?

[desert mist]

I think it’s called m die anyway

[pale tulip]

3

[desert mist]

Urm XMP I think 3.0 and idk for expo

[pale tulip]

ignore that really

its marketing

[desert mist]

Yea pretty much

[arctic lotus]

do you know for expo woopy

[pale tulip]

it doesnt matter

[arctic lotus]

so you dont know

[desert mist]

One set of Corsair vengeance will not always perform the same as the next exact same kit

[pale tulip]

ig its just expo

[desert mist]

Because of the type of chips throw on the ram kits

Yea it used to be called DOCP and EOCP

Guess expo is the new thing idk

For amd anyway

[pale tulip]

isnt docp the asus name

ahh its all sillyness anyway

[desert mist]

Maybe idk they all have their own dumb versions of it but the do the exact same things

[arctic lotus]

okay i found it. its SDK 47

SDK 45 is the old one

[desert mist]

When buying ram just look for low CAS latency and decent ram speeds that the community says run stable. Cause not every single ram kit will perform exact the same even if it’s the same model number as another @arctic lotus

[arctic lotus]

the 5600 has the best latency

[pale tulip]

according to LTT*

[desert mist]

It’s pretty stable from what I hear on am5 too

Butttt

[arctic lotus]

well also according to passmark

[desert mist]

The listing will literally show you the latency

[arctic lotus]

and it does have a lower CL

[desert mist]

Cas Latency that is

[pale tulip]

its the same thing anyway

[arctic lotus]

which is more of a multiplier instead of additive speed.

especially on am5

[desert mist]

Cl30 is pretty solid

[arctic lotus]

which has a weaker memory controller

[arctic lotus]

this one is cl28

[desert mist]

Seems pretty solid

[pale tulip]

(CL doesnt mean much for performance)

[arctic lotus]

they all share a first word of 10ns

[desert mist]

Depending on the chip lottery you might be able to overclock it a little too

Maybe up to 5800-6000

[pale tulip]

sure dream

not like secondaries or tertiaries matter more

[arctic lotus]

yeah ive seen those

which is why im trying to look into this more

[pale tulip]

anyway, as i said previously

if you want best performance, buy any kit and manually tune it

otherwise, simply dont care about it too much

[arctic lotus]

corsair seems to be a better brand than gskill tho

p on par tho

[pale tulip]

uh

[naive robin]

Wont make a difference

[pale tulip]

guilty mentioned before a bunch of issues with their DDR5 fwiw

[naive robin]

The RAM ICs are made by Samsung, Micron, SK Hynix and a few others

Corsair or G.Skill slap a label on it and sell it

[arctic lotus]

what does fwiw mean

[pale tulip]

for whatever its worht

[naive robin]

For what it's worth

[topaz solstice]

you be joking right?

[arctic lotus]

i might switch to the vengence tbh

[pale tulip]

why would you

[topaz solstice]

don't

[lusty tiger]

Don’t

[arctic lotus]

whats the difference between ram read and ssd read?

does ram read do anything different?

[lusty tiger]

They’re different technologies

[arctic lotus]

the dominator looks good too

but i think its only in 2x16 and is only 5200

so probably a no

[naive robin]

That's DDR4

Will not fit in your motherboard

AM5 is DDR5 only

[solar gulch]

Progress update?

[arctic lotus]

yeah ik lol

is the dominator really only ddr4

i thought i saw a ddr5 version

yeah its ddr5

[arctic lotus]

oh i see why you thought ddr4. i meant 5200 not 3200

im gonna check tom's hardware

[naive robin]

RAM benchmarks are best found by looking at CPU benchmarks that use RAM excessively, such as Y-cruncher or Linpack Extreme

Or various youtube videos

I believe Hardware Unboxed did a RAM comparison?

[arctic lotus]

oh so Ycruncher is the pi one

and then Linpack extreme is the successor to prime95 and it is basically just a bunch of math solving equations speed testing

[naive robin]

linpack ex is not a successor to p95

p95 largeffts is still useful for RAM testing but linpack ex and ycruncher are better for RAM testing

[topaz solstice]

#2 is so not true

[topaz solstice]

no, they're totally different in that sense they're not even the same libraries used

[arctic lotus]

and so hardware unboxed. that's a youtube channel?

[topaz solstice]

does not make it the successor though

[arctic lotus]

okay thats fine

[topaz solstice]

latest occt though uses a more modern linpack library it seems

[topaz solstice]

linpack xtreme is a reason why i know my ryzen 5 3600 can pull 130w

when being really stressed as a cpu

it will stress out the cpu to the point of breaking

[lusty tiger]

Don’t waste money on dominator. Imo, it’s ugly as hell

[gentle elm]

Fr

[lusty tiger]

G.skill FTW

he says while using vengeance lpx
Imma get g.skill next tho probably

[topaz solstice]

you'd be doing yourself a favour

[lusty tiger]

I’ve heard that it’s not as good. What makes it worse?

[solar gulch]

ay wsg yall

[orchid oracle]

Wsp

[arctic lotus]

imo its the prettiest ram out there wym

but its also too slow

[sharp harness]

Can confirm. I put 6400 cl 32 in my firends computer when building it hoping that future bios versions fix that problem. (currently can't get it to run more than 5000mhz stable 😭 .)

but that was before i updated the bios and basically every driver

i might be able to get it to 6000 if i try now.

[arctic lotus]

thanks

how would 5600 or 6000 work then

[pale tulip]

work fine

[arctic lotus]

but they couldnt get more than 5000 at first

i dont wanna get 6000 and have it not post

[sharp harness]

it will post

[arctic lotus]

not everyone has spare ram laying around to temporarily post and then update bios after it posts and then instal back the regular ram

[sharp harness]

just at 4800

[topaz solstice]

bottom of the barrel dies, basically worst binned ones that others wouldn't touch

[topaz solstice]

that's not how it works

it posts ate jedec spec first

after that you have to go into bios to enable it's xmp/expo profile so it can run at the speed it's advertised to do

[arctic lotus]

okay good

[topaz solstice]

all ram post at their respective jedec speeds

[naive robin]

You might even be able to do 6400 now

Latest BIOS updates were magic for Ryzen7k

[sharp harness]

perhaps

[arctic lotus]

so if i get a 6400. itll work at first for sure with the jedec

[pale tulip]

yeah

getting it to 6400 might be a bit of a headache

[arctic lotus]

but then after i post first time and bios updates. i can maybe get lucky and get 6400 to work

[topaz solstice]

yes, just you might not have it stable at 6400mhz without some manual tuning

would be better off just getting a kit that does 6000 cl30 though

[arctic lotus]

whats wrong with running 4 sticks of ram in dual channel mode? vs 2 sticks of ram in dual channel mode?

[naive robin]

Harder to run 4 sticks at higher speeds

4 sticks at 6000 is simply not happening on current gen AM5

[topaz solstice]

with ddr5 you won't be reaching their max speed, and there's also it not running at full capacity, both of which i stated before

[topaz solstice]

5600mhz at most if i remember correctly for speed

[hazy hedge]

technically you could if your sticks are rated for like 7600 lol

[lusty tiger]

That’s not a limitation of the ram, that’s a limitation of the cpu

[naive robin]

The CPU does not like it

The IMC cannot handle it

[hazy hedge]

gotcha

[topaz solstice]

nah, it's zen/zen+ issue in a sense but on zen 4

i suspect 8000 series will cure it

@arctic lotus if you could drive a coach/bus you could really deck it out and basically have a full house inside it

[arctic lotus]

one of these

[arctic lotus]

wouldnt 4 sticks at 5600 still be better than 2 sticks at 6000 tho?

for higher core / threaded workloads?

[arctic lotus]

yeah amd has a weak imc. internal memory controller right?

[topaz solstice]

no

[topaz solstice]

not when 4 sticks of 16gb will be seen as 32gb total and not 64gb

remember ddr5 has 2 channels per stick

[arctic lotus]

do you mean dual rank

[topaz solstice]

nope, that is totally different

[arctic lotus]

whats the difference between dual rank and dual channel

i thought dual rank was just on the stick

itself

im p sure dual rank is the double sided dram thing

[naive robin]

That'd be correct

[topaz solstice]

dual rank is basically 2 rows of ic's on a stick, while dual channel is using 2 sticks of ram

[arctic lotus]

im p sure its like 8 on both sides. ecc is 9. (but like i think only on one side its 9)

[naive robin]

It's not necessarily weak

[topaz solstice]

but with ddr5 the sticks themselves have 2 channels

[arctic lotus]

okayy

[naive robin]

The main issue is that it's first gen DDR5 controllers lol

[topaz solstice]

which is why i said it's like zen/zen+ issue tbh

@arctic lotus

[arctic lotus]

so ive heard dual channel must be supported by the cpu.

and i know dual channel must be supported by the mobo.

and ive also heard that dual channel is on the ram itself.

so is it on all 3? and if dual rank has a distinct look. is it possible to visualize dual channel or is it internal only?

awhile back i read a p in depth article on ram. and i remember most of it. but still a little confused

[topaz solstice]

dual rank you'd see pretty much straight away due to it having ic's on both sides of the ram stick, or 2 rows on a single side of the ram stick

[arctic lotus]

oh so double sided doesnt necessarily mean its dual rank then

[topaz solstice]

yup, but that's dependent on the way a certain brand has it set up

[arctic lotus]

so then, whats the difference between zen4 and am5

[naive robin]

AM5 is a socket
Zen 4 is a codename for a CPU microarchitecture

[arctic lotus]

is zen4 amd only?

[naive robin]

Of course, it is their in-house developed architecture

[arctic lotus]

im guessing intel is rocket lake or some shi

[naive robin]

Rocket Lake is 11th gen

[arctic lotus]

ive learned a bit about microarchitecture.

[naive robin]

Alder Lake is 12th gen
Raptor Lake is 13th gen

[arctic lotus]

so just a newer lake name

but still the microarch

[naive robin]

The naming is completely unrelated to anything to do with the architecture for Intel

For AMD it's just what version of 'Zen' it is

[arctic lotus]

is the microarch apart of the cpus or is it apart of the socket

[naive robin]

CPU

It's the architecture of the CPU

[arctic lotus]

okay cool

is zen4 the newest for amd?

[naive robin]

Yes

[topaz solstice]

always the cpu

[limber dune]

Why is this such a large post??????

[topaz solstice]

read pin, if you're not going to help, take a hike

[limber dune]

Lol

Just curious

Is this the largest post?

[arctic lotus]

okay i looked further into it.

dual rank is ram only. its not on the mobo or the cpu.

dual channel is cpu and the mobo. NOT the ram.

the ram configuration can determine the "channel mode" tho

[topaz solstice]

only for single vs dual channel or if you mixed capacities (doing flex/hybrid mode)

[naive robin]

One of (five?) threads made so far

[arctic lotus]

and the mode just depends on how much data can be accessed at one time.

so like

single channel can access 64b
dual channel can access 128b
quad channel can access 256b

whereas

single rank 64b blocks are on standby
dual rank 128b blocks are on standby
quad rank is 256b blocks are on standby

so then a dual channel configuration would be

2 dual rank modules in 2 slots. (the dual channel can access 128b max, and a dual rank module has 128b ready on each module so it accesses both modules at the same time and draws 128b width of data from each module)

or 4 single rank modules in 4 slots (the dual channel can access 128b max. This doesnt change, but the single rank modules have 64b blocks ready instead of 128b blocks. so it access all 4 modules at the same time despite being called dual channel, but it only draws a 64b width from each modules since that is the max single rank has ready. it is still dual channel because 64b (from 1 module) + 64b (from another module) = 128b (dual channel)

this means dual rank AND dual channel dont determine the number of ram modules that the cpu + mobo can access. dual channel determines how much data can be accessed regardless of ram stick quantity. and dual rank determines how much data is ready to be accessed if it could be accessed, which is dependent on the channel mode.

my guess then, is that

[topaz solstice]

4 in total dude

[topaz solstice]

yes though this is primarily ddr4 and before that

[arctic lotus]

no one does single rank ram for dual channel because dual rank ram is newer and more popular

and because theres probably no point of doing 4 single rank

since it still ends with a dual channel bandwidth

[topaz solstice]

not true, they do but for 16gb or higher sticks they will be dual rank most of the time

an 8 gb stick is more than likely be single rank

a 16gb stick can be single or dual rank

a 32gb stick is dual rank

though on ddr5 each stick is also dual channeled

so you can't treat ddr5 like you would for ddr4

#1153136851431989259 message <- see this

[arctic lotus]

okay yeah i think ive heard of this. its called rank interleaving. it basically allows for 32b of that 64b block to be accessed while giving time for the other 32b on the block to ready itself.

which is ddr5 only i think

so the imc is apart of the cpu?

not the mobo or the slots or the ram

[topaz solstice]

well, cpu and chipset, mainly cpu

[arctic lotus]

does the chipset have its own imc

[topaz solstice]

some do afaik

[arctic lotus]

so how would having 2 imcs work

[topaz solstice]

it would be an interlink/interleave between the 2

[arctic lotus]

ohh okay

[topaz solstice]

which should allow for more stability

but with a newer cpu, and different changes, just focus on the cpu having an imc instead

[arctic lotus]

ohh i get it. mobos used to have imcs in the chipset back in the northbridge days.

[arctic lotus]

so for ram testing you wouldnt look at higher resolution cause its mostly just a cpu cache basically

higher resolution testing is more gpu bound

so i need to see 1080p testing

[topaz solstice]

no, you would need to see the results from the likes of ycruncher or linpack xtreme tbh

the faster it is on those the better

[arctic lotus]

whats wrong with userbenchmark again

[runic bloom]

the fact its wrong

[topaz solstice]

which tbh comes down with tuning the system

[runic bloom]

and the fact they hate amd

for literally no reason

[topaz solstice]

heavily biased against amd, goalpost shifting to make any amd product to look bad even when it's proven to be better

[runic bloom]

this wasnt like an own, its literalyl just incorrect a lot of the time

[topaz solstice]

the fake eFPS they show up for gpu's is super misleading

[runic bloom]

wheres that one screenshot of a 5950x losing to an i3 was it?

[topaz solstice]

threadripper losing to a 9th gen i3

which tbh wouldn't happen

[runic bloom]

even ignoring their amd hateboner it just sucks at comparing intel vs intel

[lusty tiger]

They used to love amd

[runic bloom]

like i think its pretty obvious the 13900ks isnt 60% better

[arctic lotus]

https://youtube.com/shorts/cXEauiizE_Q?si=sHLfnoL_SfinxuO3

so this is userbenchmark vibes af

[gentle elm]

yeah according to them a 7700x is beaten by a 7 year old celeron

like what is this

lmao

[lusty tiger]

nahhh

[gentle elm]

what?

[lusty tiger]

they say the 12600 is better than a 3990x

[gentle elm]

ik

nahh what the hell is this

they say the 4080 beats 7900xtx lmao

@lusty tiger

[lusty tiger]

superseded by the 13600k my ass

[gentle elm]

kf*

[lusty tiger]

no k

[stark wedge]

isnt the 7800x better tho

[sturdy anchor]

7800x doesn’t exist

[arctic lotus]

is it saying the 7950x3d is on par with that??

cause its not lmao

[stark wedge]

x3d..

[lusty tiger]

its saying lies

[sturdy anchor]

Yes, which is wrong

[gentle elm]

12%...............

[arctic lotus]

i just wanted to see the "%" lol

[gentle elm]

yeah thats user benchma5rk

[arctic lotus]

arent kf versions worse tho

like a super miniscule amount tho

[gentle elm]

no they just dont have intergrated graphics

here is there list of "equivlent" cpus to a 7950x3d

12100f 💀

[lusty tiger]

literally none of them are equivalent

3600????

[gentle elm]

userbenchmark is a fucking joke

[lusty tiger]

well, the 13900k, kinda

[lusty tiger]

but yes, it is

[gentle elm]

yeah kinda but a 12100f????????????????????

ima go donate to userbenchmark to support there great work

[lusty tiger]

https://www.youtube.com/watch?v=RQSBj2LKkWg&t=606s

[gentle elm]

$100 sound fair?

[gentle elm]

for this

[lusty tiger]

-$100

[gentle elm]

lmao

[lusty tiger]

they should be paying for their lies

[gentle elm]

woops didnt show percent in that

[lusty tiger]

spits drink out

.....1%

[gentle elm]

what

the

f

is

this

ima go buy a 9600k instead of a 7800x3d cus userbench mark said its better
Okay? @lusty tiger

[lusty tiger]

lmao

[gentle elm]

@lusty tiger

[lusty tiger]

.....

i didnt know it was that bad

[gentle elm]

its way worse

6% better..............

[lusty tiger]

what percentage

[gentle elm]

sry i keep cutting it out

and this is for productivity

[gentle elm]

so ima get the old ass sucky xeon cus it is only 5% worse than 3990x

[lusty tiger]

ew

[gentle elm]

anw ima go play csgo 2 now

[naive robin]

Not anymore

[naive robin]

No such thing existed

There is one IMC per system

[desert mist]

userbenchmark is nvidia and intel biased lol

They literally dog amd on every single review lol

the conclusion for the 3090 ti is "I like it" LOL

[crimson niche]

💀

[lusty tiger]

And the amd one is just wrong. “Experienced builders have no interest in buying amd cards, regardless of price.”

[lusty tiger]

lets just move on from userbenchmark

[solar gulch]

userbenchmark is goofy

[lusty tiger]

ye

[orchid oracle]

Bro even the Intel subreddit banned User mench bark what do you expect lol

they just trolling

[arctic lotus]

yeah i wont use userbenchmark

it's not even a good website for intel lol

[arctic lotus]

whats the highest speed intel can do

ik amd it's
6000 sweetspot
6400 kind of unstable
6400+ unstable
but what would the intel equivalent be?

[mossy granite]

on avarage, how much does it cost to put together a pc that can run oculus rift

[arctic lotus]

why would i know

[mossy granite]

kk

ty

[arctic lotus]

so it looks like 7200 is the 6000 equivalent for intel

whats the highest intel can go?

[arctic lotus]

so another thing i will say for this. is the single sided vs double sided module has nothing to do with rank actually. it's like the physical side of pcie slots vs electrical bandwidth side of pcie lanes.

the actual rank is what matters more, and how it's sided doesnt matter. there are double sided single rank and double sided quad rank etc.

the rank is just the number of 64b wide blocks.

single rank has 1
dual rank has 2
quad rank has 4

so this makes even more sense now.

it is only really the case for single rank double sided modules tho that could be confusing. but there are only ever going to be 2 sides on a ram stick so it wasn't obvious at the time to me.

[naive robin]

Intel can do up to 8000 or slightly higher in gear 2 but that nukes performance
Afaik Intel can boot and daily 6000-6400 in g1

[arctic lotus]

i also found another good source for ram. and i think i was wrong about ddr5 for amd. it seems to be more speed senstive than i thought. and more speed sensitive than intel.

but intel gets more out of ram since they can utilize 7200

[arctic lotus]

what are gears? is that an intel only thing

[naive robin]

Yes

Equivalent to AMD's 1:1 and 2:1 modes

[arctic lotus]

so 7200 = the sweetspot
7400 = kind of unstable?
7400+ = unstable?

[naive robin]

Also it's more like 6000 is the sweetspot and limit for most CPUs
6200 is rare-ish, 6400 is almost unheard of

[arctic lotus]

yeah for amd

what about for intel

ik about the 6200 status vs 6400 and the lottery

[naive robin]

Intel cares much less about RAM speed than AMD's current architecture

Stick to 7200 or less for Intel

[arctic lotus]

which is why i say this

[arctic lotus]

so 7200 is the 6000 equivalent for intel?

[naive robin]

Maybe? I'm unsure

[arctic lotus]

damn

so i am pretty sure i will be switching to 4x modules

the rank interleaving capabilities and the cpu speed of zen4 should be good enough to benefit from 4x modules.

it wouldnt be much improvement, and only matters for cpu bound operations, but the improvement seems to be worth it.

[topaz solstice]

no it would not

not when you only get to use half of each stick's capacity

[arctic lotus]

in dual channel mode, it should be able to access 128b at a time. if each slot 4x dual rank. i dont think it would make a difference from 4x single rank.

4x single rank would be
4x 64b blocks that dual channel accesses 128b at a time.
4x dual rank would be
8x 64b blocks that dual channel accesses 128b at a time.

[arctic lotus]

tbh i actually do not know why there is a slight performance increase

but it looks consistent, and thats also what my original article i learned from said too

so basically all of my sources say 4 sticks is better contrary to popular belief

but yes, its definitely not quad channel lmao

[topaz solstice]

and a 4x16gb stick will not give you a total of 64gb of capacity with ddr5

it will give you 32gb

[arctic lotus]

my personal take is, i think there is just more headroom and the rank interleaving allows better "juggling" of that data. but tbh i really dont know

[topaz solstice]

for 64gb and higher capacity you have to go 2 sticks of ram

[arctic lotus]

i dont understand this part tbh

[topaz solstice]

ddr5 has 2 channels per stick

[arctic lotus]

are you talking about the 32b rank interleaving thing

[topaz solstice]

and with 2 sticks it gives you a quasi-quad channel

but when you use 4 sticks, half of it is disabled

you're limited to 128gb on the mobo

[arctic lotus]

do you have a source for this.

i just want to read more about this on my own time if you can please.

[topaz solstice]

https://www.reddit.com/r/buildapc/comments/wm0x0r/can_you_get_4_sticks_of_ddr5_to_work_well/ some people discussing about using 4 sticks of ddr5

https://uk.pcmag.com/components/137375/what-is-ddr5-everything-you-need-to-know-about-the-latest-pc-memory-standard

[runic bloom]

Also after making such a big dral about ram speeds now you want to go 4 sticks

[topaz solstice]

https://forum.corsair.com/forums/topic/181631-4x16gb-corsair-dominator-ddr5-5600-only-32-usedreconized/

https://www.anandtech.com/show/17585/amd-zen-4-ryzen-9-7950x-and-ryzen-5-7600x-review-retaking-the-high-end/5

https://forum.level1techs.com/t/ddr5-ram-on-am5-fails-to-post-with-all-4-slots-occupied/198640

forgot this one was a thing also

where using 4 sticks of ram can be that unstable the system won't even post

and no it's not stuck in a long endles loop of memory training due to this

as memory training should take no more than 5 minutes at most

discusses ddr4 with zen2/3 architecture but the info can be also applied to ddr5 with zen4 architecture

[arctic lotus]

so for this one. i am already aware stick number doesnt determine channel. that is just the cpu + mobo combo.

you either have quad channel or you dont. you most likely dont unless you have server level equipment. like i think threadripper is quad channel.

[arctic lotus]

so in this article i read through it. and i did find this

[topaz solstice]

i've seen it also

[arctic lotus]

this is what i mean by rank interleaving

[topaz solstice]

https://www.rambus.com/blogs/get-ready-for-ddr5-dimm-chipsets/

[arctic lotus]

i kind of chop this up to being user error tbh.
but i adore wendell. even though the person having the issue isnt wendell lol.

[topaz solstice]

but still for 64gb or higher you go 2 sticks not 4

[arctic lotus]

i hate that this is calling it channels. ive also heard people call it rank. it's neither lol.

[topaz solstice]

less of a user error on this one, it's mostly bios, and it's someone whow was a n early adopter ofr ddr5 system

[topaz solstice]

nope, it's 2 channels

and it uses dual rank ic's anyways

[arctic lotus]

that just makes it sound like its dual channel tho. or implies ddr4 didnt have dual channel

just bad naming imo

[topaz solstice]

ddr4 doesn't have 2x 32bit bus channeling, it has a single 64bit bus channel

[arctic lotus]

lmao yeah but dont call it 2 channel that's sooooo dumb

not you

[topaz solstice]

it's not

[arctic lotus]

just in general

i mean. its my opinion. its super dumb lmao

idek what else to call it tho tbh

[topaz solstice]

you can treat the channels on the ram as sub channels to the configuration of how you would set it up on the mobo

but if half gets disabled when in use for doing 4 sticks, that's half the ram being electrically disabled

[arctic lotus]

im still calling it rank interleaving. its really just a capability unique to ddr5. its not even about the channels. or maybe just call them subchannels.

[arctic lotus]

channel architecture is a good word

[topaz solstice]

it's not rank interleave, that would be mislabelling what it does

[topaz solstice]

this clearly shows that each channel is using dual rank ic's

but as they split the bus from a single 64bit bus into 2x 32bit bus for better efficiency it means data access is faster and you can go with a higher bandwitt per stick

[arctic lotus]

yeah that picture is dual rank ddr5

[topaz solstice]

which all ddr5's will be for 16gb sticks and over

[arctic lotus]

yeah

[topaz solstice]

using an 8gb ddr5 stick would be pointless

different issues for that i believe

[arctic lotus]

fair enough, but i will still call them subchannels.

[topaz solstice]

still though it results in if you use 4x ddr5 sticks you have to learn to manual tune and OC the ram

[arctic lotus]

really just the 32b channel architecture

[topaz solstice]

that would be fine as you can treat them that way

[arctic lotus]

but that is still fine for 4 sticks

[topaz solstice]

and a first time builder generally has no clue beyond just enabling xmp/docp/expo

as well as the tuning side of ram should be only done after you're used to the system as it is

[arctic lotus]

yeah, but i personally will probably try to learn tuning right after building and posting pictures.

[topaz solstice]

you will have to be prepared to be resetting the cmos several times while tuning ram

as bad tuning = no post

[arctic lotus]

oh so docp is for asus only

[topaz solstice]

for ddr4 mainly

[arctic lotus]

im okay with that if i get a cmos button etc

[topaz solstice]

you don't for the better part

[arctic lotus]

ik its only on more expensive mobos

[topaz solstice]

but the best way is getting a separate switch that would normally be used as a reset switch and connecting it on the clrcmos header

[topaz solstice]

they all still have a 2 pin clrcmos header

and rather than constantly shorting it/ opening the case up to get at the clrcmos button

do what i just said about using a spare button

i repurposed the one on my case for that as i fitted an extended external power button which has a smaller reset on it for front panel control

and yes, i do have my ram on my setup tuned and oc'd

[arctic lotus]

what is fclk?

[arctic lotus]

okayy

seems like something i could do

just buying a switch basically

[topaz solstice]

fabric clock

it's what the infinity fabric runs at

[topaz solstice]

one of these, attach it to the clrcmos header, and any issues that needs the cmos to be cleared, you just press it

[arctic lotus]

i might have an extra one of those

i have an old atx case

[topaz solstice]

i have more than 1 extra of them tbh but due to using an external power switch like this

i can just use the cases switch instead

the external switch i got for my use case, though it could be useful for yours if you have to put the pc in a hidden spot

[desert mist]

Not really true, because especially with the 13900k vs the 3d chips Intel has much more to gain from ram OC

[arctic lotus]

tuning seems to not really yeild that much on ddr4, but for ddr5 zen4, its a measurable difference.

[desert mist]

Granted yea there’s a point of diminishing returns and it’s probably met well before 7200 but

[arctic lotus]

single rank vs dual rank doesnt seem to be a big factor, and 5600 speed seems to have some timing issues even tho it is cl28.

[desert mist]

When your pushing for super high clocks having faster ram definitely isn’t a down side

As long as it’s stable that is

[topaz solstice]

nope it improves on both generations, just ensuring the better performance by a small amount on each and generally better stability for the ram ic's themself

[desert mist]

Some claim to get 8000MHz stable but idk about all that

[topaz solstice]

buildzoid has, though it took a while for it

[arctic lotus]

i dont think you get perf uplift 100% but yeah there is no reason to not tune your ram

[desert mist]

Yea

Really no point tho

[arctic lotus]

but on zen4, tuning has a bigger impact it seems

[desert mist]

Best thing to do is tighten timings not per say OC

[arctic lotus]

so what is the 6400 for intel?

[desert mist]

Idk some claim much higher stable all over Reddit so

I think I even saw one person say they got 8400MHz stable but that’s pretty steep

[arctic lotus]

tuning also makes speed matter even less

[topaz solstice]

tuning is more to do with stability

[arctic lotus]

a 6000 cl30 is still going to beat out a 5200 cl30 tho.

[arctic lotus]

i agree with this

thats why i would tune personally

[topaz solstice]

unstable ram causes system crashes

[desert mist]

Not necessarily

Depends on the chip manufacturer

[arctic lotus]

i mean maybe

[topaz solstice]

5ns difference tbh

[arctic lotus]

like if you lost on the lottery maybe

[desert mist]

Which isn’t an uncommon thing lol

[topaz solstice]

which while not noticeable still affects the systems own responses

[desert mist]

Exactly no point in chasing fractions of a nano second

[arctic lotus]

still, not a valuable reason to go 5200 over 6000

[desert mist]

Exactly. Just throw in CL30 6000 and call it a day

[arctic lotus]

well, like ive said before

the whole point of this additional research is to supplement ltt. i originally chose the 5600 cl28 based off them.

i have many more sources now

[desert mist]

Yea I’m the past LTT has been wrong ALOT so

[arctic lotus]

yeah well. im not going to repeat myself

i wouldnt be doing this, if i didnt watch those videos

[desert mist]

Just saw the video about their amd card experiences and now they’re keeping them in their personal rigs after they dogged amd so

[arctic lotus]

about ltt vs gamernexus and all that

[desert mist]

Steve at GamerNexus is normally a pretty reputable source. If he is wrong he’s pretty quick to fix it

[topaz solstice]

20 mins of testing vs 2-3 days of testing being the difference

[desert mist]

Aka quality of content

[arctic lotus]

i watched all 4 videos

[topaz solstice]

wouldn't be surprised if gamernexus was more like a week or so of testing especially with gpu's

[arctic lotus]

i wouldnt be doing this if i wasnt convinced lol

basically, straight up lying on some benches

[desert mist]

Yea mentioned spending weeks testing several CPUs so wouldn’t doubt it

[topaz solstice]

yeah for sure

[arctic lotus]

but its fine. i have better sources now.

[desert mist]

He’s my go to for most pc part reviews. Especially cases he has some superb case reviews

[arctic lotus]

so what are the 5th and 6th numbers on timings usually

denoted

ik the first 4 mostly

1st one is CL etc

ik one is row speed

and column speed

im p sure the 4th one is the whole time

or something

but yeah it seems that the 5th and 6th timings usually denoted are tuned timings

so like ddr5 6000 cl30-38-38-38-96-96 is better than ddr5 6000 cl30-36-36-36-76-112 because even though 36 is lower than 38. the first one with 96-96 is tighter than 76-112. but these are still both plug and play examples.

is the server down

[topaz solstice]

It shouldn't be

[arctic lotus]

i typed up a prompt but had some issues sending

my bad

its not the server it was doing it on my server too

[arctic lotus]

i think im back

[naive robin]

tCL - tRCDRD/RCDWR - tRP - tRAS - tRC

None of those timings massively matter for performance btw

Under some testing the difference between C30 and C32 was about 0.6fps at a >120fps average, aka entirely within error

[naive robin]

Discord is having issues right now afaik

[topaz solstice]

Yup got cloudflare warning earlier strangely

[arctic lotus]

same

only mobile works for me right now

discord getting hacked lol

watch dogs
untuned winner = #1 6000 cl30 (96-96)+8fps#2 6000 cl30 (76-112)+2fps #3 5200 cl30 #4 6400 cl32 #5 5600 cl28 #6 6000 cl40 #7 5200 cl36 #8 5200 cl40
tuned winner = #1 5600 cl28 & 6000 cl30+5fps #2 5600 cl28 single rank

Hitman
untuned winner = #1 6000 cl30 (96-96)+7fps #2 5600 cl28 & 6400 cl32
tuned winner = #1 6000 cl30+2fps #2 5600 cl28

zero dawn
untuned winner = #1 6000 cl30 (96-96)+4fps #2 6000 cl30 (76-112)+2fps #3 6400 cl32 & 5200 cl30
tuned winner = #1 5600 cl28+2fps #2 6000 cl30

tomb raider
untuned winner = #1 6000 cl30 (96-96)+5fps #2 6000 cl30 (76-112)+2fps #3 6400 cl32 & 5600 cl28
tuned winner = #1 5600 cl28+2fps #2 6000 cl30

riftbreaker
untuned winner = #1 6000 cl30 (96-96)+5fps #2 6000 cl30 (76-112)+4fps #3 6400 cl32
tuned winner = #1 5600 cl28+5fps #2 6000 cl30

requiem
untuned winner =** #1 6000 cl30 (96-96)+2fps** #2 6000 cl30 (76-112)+2fps #3 5200 cl30 & 6400 cl32
tuned winner = #1 5600 cl28+3fps #2 6000 cl30 & 5600 cl28 single rank

spiderman re
untuned winner = #1 6000 cl30 (96-96)+7fps #2 6000 cl30 (76-112)+6fps #3 6000 cl40 (lmao)+2fs #4 6400 cl32 #5 5600 cl28+2fps
tuned winner = #1 6000 cl30 & 5600 cl28+4fps #2 5600 cl28 single rank

callisto
untuned winner = #1 6000 cl30 (96-96)+11fps #2 6000 cl30 (76-112) & 5200 cl30+2fps #3 6400 cl32 & 5600 cl28
tuned winner = #1 6000 cl30 & 5600 cl28+3fps #2 5600 cl28 single rank

https://youtu.be/MOatIQuQo3s?si=7AYaKNSdrzaT9BGq

so yeah it would look as tho you guys were mostly right on ddr5 6000

discord still going bananas lol

[lusty tiger]

See how all of this only gives you like 3 fps? It shows how much ram matters. Just get 6000 cl30 and call it a day

[arctic lotus]

actually no

6000cl30 outperforms all of the rest with almost always a 10fps lead

the +fps is comparing it to the adjacent placeset

[lusty tiger]

So does that mean you can just pick some ram and move on?

[arctic lotus]

i already picked some ram and moved on

i am coming back because you guys didnt like the ram i picked

i picked the ram based off ltts video

then the ltt testing fraud was announced. i finally watched all 4 videos. which were really long.

so i returned to ram. found more sources.

a lot of my research is still true too

like speed is still additive but amd is more sensitive than i thought to speed.

so the relative gaps between 5200 and 6000 are actually significant enough

but also cas is still a much bigger factor as i said before. however it is more scaling.

and amd really seems to like cl30. in fact cl28 actually has stability issues making 5600. the only one that uses cl28, actually not a great speed.

[lusty tiger]

So you’re getting 6000 cl30

[arctic lotus]

so in other words. the 5600 cl28 latency is too low. and cl30 or cl32 etc are fine. but cl36 and cl40 is where you really start to have a deadweight.

this is important because it changes everything.

yes 6000cl30 offers the best all around experience and best plug in play experience.

and there is certain 6000cl30 modules with 96-96 timings that are already tuned extremely well with end users in mind.

however, the 5600cl28 still offers more for oc world record breakers. where a tuned module would basically stabilize it and unlock its cl28 potential

so now the question really is.

how hard is it to tune lol?

cause if its not too hard to manually tune.

then i just might input buildzoids numbers into bios and call it a day.

[lusty tiger]

I doubt that will work

Due to the silicone lottery

[arctic lotus]

yeah the +fps is only like a negligible increase of about 2 from the 6000cl30. buy when comparing it to the rest untuned. its like +15fps

[arctic lotus]

im p sure this is what everyone does.

[limber dune]

Me on 2400mhz

[arctic lotus]

but alr thats fine

so then how do you tune ram?

[naive robin]

Manual tuning is a very long process

Takes a couple days of tweaking and stresstesting to ensure stability

[arctic lotus]

how difficult cause it doesnt seem hard

[naive robin]

It's not difficult, it's just extremely tedious

[limber dune]

Sooooo overclocking?

[arctic lotus]

hmm okay. im strongly considering the 6000 cl30 now. but i want to tune anyway so this will be fun

also these are just speeds

[naive robin]

Well, yeah

Anyway CAS latency and all the other primaries don't matter
You want to tune your secondaries and tertiaries for actual performance gains

[arctic lotus]

how do you compare different ram manufacturers

so like

gskill tridents are like top tier right now

but corsair has the vengence

and the dominator

the dominator open box says only 5200

but ik you can set it to anything

cause jedec is like 4800. and so 5200 means nothing then

since i could just set that to 6000 then

[lusty tiger]

Just grab some g.skill 6000 cl30

[arctic lotus]

well i also need to make sure its not 76-112

which will be hard to find probably

[lusty tiger]

How much of a difference does it make

[naive robin]

Well for DDR5 you want Micron Rev. N i believe it is

If you want to overclock it manually

The actual brand of it doesn't matter one bit
Corsair and G.skill and Patriot and literally everyone else uses Samsung, SK Hynix or Micron dies

[arctic lotus]

lets put it this way. the 6000 performs so well out of the box.

the no other module could compete except itself.

sometimes even the cl40 was better than some other speed

[naive robin]

And there is no surefire way to ensure what die you get lol

[arctic lotus]

in addition to this. the 5600 performs comparitively horrible open box untuned.

so if i got a 5600. i better learn how to tune fast otherwise the ram is going to be suboptimal

[arctic lotus]

is this the die

ive heard of hynix A die i think

ill have to watch more buildzoid for the die

[topaz solstice]

Different results within the same specs and no tunings are the same, and what results buildzoid got wont be the same as yours

[pale tulip]

meh, hynix a die is good enough

[lusty tiger]

Don’t manually tune

Just get some g.skill 6000 cl30 ram

[pale tulip]

worst advice uve given

grab 6000cl30 and tune that

[topaz solstice]

6000 cl32 and tune it to 6000 cl30

better timings from the start

[pale tulip]

whatever

close to the same result in the end

[topaz solstice]

Definitely will be

[orchid oracle]

Minimum Spec:
Processor: Intel i3-6100/AMD Ryzen 3 1200, FX4350 or greater
Graphics Card: NVIDIA GTX 1050 Ti/AMD Radeon RX 470 or greater

Reccomended Spec:
Processor: Intel i5-4590/AMD Ryzen 5 1500X or greater
Graphics Card: NVIDIA GTX 1060/AMD Radeon RX 480 or greater
Memory: 16 GB RAM

But i would reccomend start a new thread for that

Thats for the Rift S btw

[topaz solstice]

Make your own thread

[lusty tiger]

I mean, you can, I don’t think it’s worth it

[pale tulip]

you do it once, get performance gains, then never mess with it again

[topaz solstice]

once tuned right, it'll be more stable than being left at xmp/expo profile and once set it's a leave and forget, while saving to a profile that you duplicate onto a usb stick

that way you don't have to retune it everytime the bios gets updated

[arctic lotus]

tuning seems to produce a very nice end result

[pale tulip]

yeah

the journey is a pain but the end result is alr

[arctic lotus]

if not all at once, then 100% will tune eventually

[arctic lotus]

this is totally something i could do

okay, im gonna look at some 6000 cl30 ram

but still consider the 5600 cl28 ram

6400 is actually garbage tho btw

[pale tulip]

theres a guaranteed hynix a-die for $80

[arctic lotus]

6400 isnt even good if you win the lottery

[pale tulip]

if you want to have some fun manually tuning

[arctic lotus]

im thinking about 4 modules if im going 6000 cl30 tho

[pale tulip]

why

[arctic lotus]

oh wait, i decided against it last minute

sorry im just waking up

i decided against it because of latency.

the dual channel would have to travel through a2 before reading a1 i think in channel a

and b2 before reading b1 in channel b

so im guessing you dont plug in ram in a1 and b1 then?

even though a1 and b1 are closer to cpu

[naive robin]

Not how that works

[arctic lotus]

uh idk

ive also heard of it before

but everytime i research it i stumble upon it

so like constantly affirmed that. that is how it works

[topaz solstice]

this you need to check from the manual itself

[naive robin]

You use slots 2 and 4 due to them being at the end of the daisy chain in a standard motherboard using a daisy chained memory topology

We have had this discussion before

[arctic lotus]

isnt that what i just said lmao

im talking about using 2 ram instead of 4

a2 and b2 are slots 2 and 4 lol

[naive robin]

No, you said whatever the hell this was

[arctic lotus]

thats the same thing

[naive robin]

No?

The CPU and IMC do not access memory slots individually

[arctic lotus]

slots 1 and 3 are closer to the cpu

[topaz solstice]

there are some in which it's slots 1+3

[arctic lotus]

is it cause its moved around on the mobo

[naive robin]

It entirely depends on how the PCB designers lay out the slots

And the 'latency increase' from having 4 sticks instead of 2 is more because you have to run them slower to ensure stability

[arctic lotus]

okay thats fine if i get a weird ass mobo ig lmao thanks for the advice

but p sure im completely right lol

[naive robin]

I'm talking about the weird stuff you said about "travelling through a2" or whatever

Because that is not how that works, nor does it make sense

[topaz solstice]

you would expect it to normally go 1,2,3,4 as the slot arrangement, but there are those that do it as 4,3,2,1 or 1,3,2,4 or even 4,2,3,1

[arctic lotus]

ive heard multiple sources talk about it like this

it can't access the data in a1 or b1 immediately

that latency is what produces that slowed down speed