>>108651121
besides leddit post
xmp is pretty much a retarded concept
i've seen modules spontaneously die from higher config voltage and especially DDR5 with non reporting ODECC silently masking degradation i would not use that on anything serious

>>108651129
XMP is with in warranty.

>>108651129

Manual tuning is far better in terms of performance and it's also tends to be significantly safer.

XMP/EXPO profiles doesn't even guarantee short-term stability. It's borderline useless to be honest.

You are better off trying to copy overclocking settings from random redditors who don't know what they are doing.

>>108651121
>buys 9950X
>not even X3D
>cucks it with slow RAM
>having bought a Assrock mobo in the first place
Hahahahahahaha

>>108651138
for gaming, maybe but i dont see benefit of dram OC outweighing stability cost of stepping out of JEDEC specs for any serious workload

>>108651151
How about running your entire OS and programs from RAM. EVERYTHING becomes snappier and more responsive. Unused RAM is wasted RAM

>>108651151

You can get tuned ram as stable as you wish. You just have to test it properly (which most people don't know how to do).

Ram bandwidth has a massive impact on productivity performance. If you are not doing any productivity why buy 9950x ?

>>108651121
My PC is 13 years old and I've never had to worry about the ram controller exploding because it's too high strung, factory overclocking is the computer equivalent of direct fuel injection and weak ass 1.5L inline 4 engines with turbochargers, they're extremely high strung and basically not built to last.

>>108651189
that is the delusional part, that you think you can fuzz out ram specs

>>108651143
>Hahahahahahaha
they took our gpus and ram, and you are laughing

>>108651199
Only if you're a poorfag.

>>108651190
my ram overclock is lower than the xmp profile with the same voltage

>>108651209

my XMP is 6000 CL32 @ 1.4v

I am running 6800 CL32 @ 1.395v

I could probably had gotten it properly stable with less voltage (such as 1.38v) but i was getting fed up with tuning it and 1.395v should be safe with A-die anyway (i have a fan actively cooling it).

>>108651121
do PCfags really? my macbook just runs it at 9600MT/s

>>108651224

Apple have control over all hardware aspects allowing them to push for those high speeds.

When you build your own PC it comes down to the individual parts working properly with each other even though they were never tested together to begin with.

>>108651221
I'm happy for you
i'm running manual 3600 1.35V ddr4

XMP was a mistake.

>>108651262

It's there because customers want it but they don't actually make those profiles worth using. People as a whole would have been better off if no such profiles existed.

It's a nice concept in theory but the actual reality if it is pretty bad.

>what's jedec
JEDEC DEEZ NUTS AHAHAHAHAHAHAHAHhaa

File: maxresdefault.jpg (62.8 KB)

62.8 KB JPG

I run mine below JEDEC

>>108651193
5ghz is a lot less than 7ghz.
The change will make bandwidth bound tasks significantly faster.
Though if you really care about bandwidth you should use a platform allowing more than 2 memory channels.

I can't get my DDR5 RAM to XMP. I don't know if it's the motherboard's fault or the tendency of DDR5.

It just crashes the system if I enable XMP.

>>108651121
I run it at JEDEC because it takes two fucking minutes to "train" the RAM every time I boot my PC.

>>108651136
Warrenty doesn't make up for an unreliable PC
There is a reason XMP/EXPO is very much limited to gayming PCs.

Feel free to run it, it's your PC but I'd wish people quit with this delusion that the standard of reliability given to OCed RAM is somehow just as good as standard JEDEC

i even use ECC ram, i care about stability

>>108651662

The problem is typically the motherboard or the CPU memory controller.

https://vintologi.com/threads/ddr5-overclocking-nightmare.1229/

>>108651121
Why would I not run JEDEC? The question isn't even regarding stability or whatever, I just want to use my devices, PC parts and even software as the manufacturer/developer intended me to use it, not in some special snowflake custom way.

File: 6800 11 10.png (506.8 KB)

506.8 KB PNG

>>108651696

You can typically get DDR5 to pass 48 hours of any stresstest fairly easily if you stick to the (slow) XMP settings.

>>108651824

Because performance is way higher if you do a proper manual tune (while being stable to the point where no stresstest find anything wrong even if you run it for 48 hours).

>>108651843
I don't want to learn how to tune anything or spend time running stress tests

>>108651907

you likely would want run memtest three clean runs

>>108651907

Even running at JEDEC does not fully guarantee stability though. You can still get issues due to faulty hardware.

>>108651136
XMP voids the CPU's warranty

>>108651824
>as the manufacturer/developer intended me to use it
Which would be the XMP/EXPO profile.

Note that all the speeds listen on the packaging, on the webshops and in reviews all use XMP/EXPO, nobody ever uses JEDEC.

It's like buying a 200HP car that says 200HP in the brochure, says 200HP at the dealership, all the car reviewers say is 200HP and test it at 200HP but actually it's only 100HP or you're being a special snow flake car tuner who just voided his warranty.

>>108652006
Most people don't redline their shit anyway because they think it hurts the car

File: 20260105_175044.jpg (810.9 KB)

810.9 KB JPG

>have 6000mhz 32x2 gskill ripjaws ram
>supposedly is EXPO compatible
>everytime i try to activate EXPO the motherboard wont post for 5+ minutes, then it restarts, and resets to default JEDEC
>i set it to XMP, works everytime after training the ram.
It just works. I wont OC it as the testing process isn't great for the sticks.

>>108652017
it does, unless its just after a fresh oil change, and you put high octane fuel in there, and you're going at a decent speed so the radiator can work under ideal conditions. Its also good for clearing out the carbon buildup in the pistons in older cars.

>>108652006
this, im kinda stunned to see anyone in /g/ of all places, admit to boomer-tier old wives tales about ram specs of all things. This is something i'd expect my dad to say.

>>108651907
apple has the perfect product lineup for you

>>108651121
I have had two xmp kits and one mobo die to physical memory failures meaning that after they went shit they wouldn't work even at jedec specs, so now im running basic ass kingston valueram at jedec spec and it has lasted longer than either xmp kits did. Only retarded gaymurs think they need more.

>>108652006
>Which would be the XMP/EXPO profile.
Those profiles are made by the DIMM packager, who may not even be the DRAM chip manufacturer, and they certainly are not the CPU memory controller manufacturer.
I can put tires rated for 180mph on my car, that doesnt mean my car can do 180 now.

File: hgktg.jpg (15.2 KB)

15.2 KB JPG

i combined a faster and slower kit at the same time so i can't mess with the timings i think

>>108652006
>It's like buying a 200HP car that says 200HP in the brochure, says 200HP at the dealership, all the car reviewers say is 200HP and test it at 200HP
This analogy only works if you have a prebuilt PC or laptop. Ford tested every component of my car. Intel did not test my AsRock motherboard, and neither Intel nor AsRock tested my DDR4 kit.

>>108652231
>tested
validated is a more fitting word

>>108652225
There is a relatively high chance the memory kits are identical aside from the sticker.

>>108652271
no they're completely different brands

>>108652225
you're right
you have low agency so you can't do anything

>>108652085

You can reduce the voltage if you are worried about that and still probably exceed the XMP/EXPO performace by a large margin.

>>108652281
Brand is just a sticker. The companies don't actually make the ram, they just ask the manufacturer to put their logo on it.

File: 1761763097666851.png (1.4 MB)

1.4 MB PNG

"Hardware" enthusiasts have progressively gotten more and more insane as time has gone on. I'm reading through this thread as someone that CADs circuitboards and programs PCIE FPGAs and I have no idea what the fuck you people are smoking.
>Your RAM isn't overclocked? Tsch, basically you're a fucking idiot.

>>108652004
not in white countries
if you show your product being capable of running at those speeds in your material, and that you have sold the product with that feature encouraged, you have to warrant that use case.
you cannot show all this XMP/PBO/OC potential in your official material then void the customer’s warranty when they follow your advice.

>>108652770
shouldn't be hard for you to understand at all then? what the spec demands and what the hardware is actually capable of is two different things

>>108652845
https://www.amazon.co.uk/dp/B07RW6Z692
Show us in the product description where it says
>up to
>capable of
>that it is designed for JEDEC and XMP is OC
please

>>108652869
i don't care about XMP profiles
any "auto overclocking" features are inevitably going to be dogshit
you're gonna have to find someone else to get mad at

>>108652880
so the answer is they’re falsely advertising
thanks

>>108652893
ok

File: XMP-vs-manual-tune.png (145 KB)

145 KB PNG

>>108652770

>>108652845
No actually, it is in fact very hard for me to understand why you are obsessing about overclocking RAM on a desktop system. I mean I understand on some level it's just retarded benchmark autism, but the amount of flaming here is genuinely asinine.

>>108652682
so what are you implying? i am mixing 2666hz with 3200hz. i know i can technically change the lowest one to go higher, but it's a cheap kit so i don't think it will be stable enough to reach anywhere near the faster sticks

>>108653016
The first step is to find out which "ic" you have. These are the black square chips on the memory sticks. You can use a program called Thaiphoon Burner to read this information from the ram stick. This requires windows operating system.

>have to run ram+cpu+mobo outside of specs and void the warranty to get best performance
it's a scam all the way down

>>108653026
what would i do with that info?

>>108653040
then we can decide if it's worth trying to set all your sticks to 3200

>>108652231
thats what the QVL is for

>>108653011
it's fine if you think it's a waste of time, you can do whatever you want with your system, but you made it sound like the concept itself was impossible to understand

(and you haven't actually tested it so you have no idea how much of a difference it does or doesn't make)

>>108653051
can i look at online pictures of the sticks? i am using troonix

>>108653073
No, I didn't. I was quite clear about finding this thread and modern hardware enthusiasts to be insane. You're definitely not beating the allegations by interpreting "You are all out of your minds" as "I don't understand what you are talking about".

>>108653085
maybe it's helpful. it's best just to try and back off if it's unstable
you can run y-cruncher 1b, 2.5b, 5b to get a quick idea whether it's at least somewhat stable or not
it's linux native

>still arguing about theoreticals instead of just going into the BIOS and setting the speed to 3200
yep, low agency
poke it with a stick, see what happens

>>108653140
I bought a 20 euro 2666 kit for my brother and overclocked to 3600 cl18 in less than 5 minutes
this was just to test the sticks in my pc, they were used

>>108651121
reddit niggers

>>108653167
i needed RAM for a home server so i took an ancient 2400C16 kit and overclocked it to 3200C16
still stable 6 months later because, you know, you can stress test your hardware before you put it into production

>Tuning ram in 2026
>Waste days on instability/getting it right
>Get 3% better perf in retardBench26

ISHYGDDT

>>108653184
how did you do that? can i just change the transfer rate without tinkering with voltage or timings?

>>108653231
only one way to find out

File: XMP-vs-manual-tune.png (144.4 KB)

144.4 KB PNG

>>108653219

You are missing "3".

>>108653231
I think it's just a coincidence it was cl16 in both cases
2400 jedec is cl16, 3200cl16 is a common xmp profile for mystery meat ram

>>108653263
>I think it's just a coincidence it was cl16 in both cases
yeah kinda
the XMP profile was 2400C16 at 1.2 V
i did 3200C16 at 1.35 V
of course if i did 1.45 V it might have been able to do 3200C15, but it was for a server, so no point in wasting those extra watts

>>108651121
is that really a bad call? most people include that op have no legitimate use case for fast ram anyway and it can turn xmp back on whenever ram gets cheap again

>>108653278
1.45V is known to degrade some ics, no reason to run that high outside of suicide runs

>>108653314
it's 8gbit AFR
so 1.45 V would have been fine
and if it was in my main system i would definitely have squeezed every nanosecond out of there

>>108652085
its not the pc components, its your fucking houses electrical wiring and/or psu bro, i've never heard of anyone having more than 1 set of ram die on them, period.

Call up an electrician and ask for a fuse box and home circuitry examination for shorts. Or try it yourself and die to electrical shocks idc.

>t. Had a metal screw sitting behind the fuse box circuitry from a lazy home builder, was causing the occasional power overload and blackout where i had to go outside and turn back on the fuses. Electrician installing outdoor lights saw it and removed it for me for free.

>>108652271
>relatively high chance
*1% chance more like, especially if the poster is as tech-illiterate as i suspect.

>>108653301
turn on XMP retard, its the industry standard, you're a moron for not using it.

>>108653365
It's far higher than 1%, it's more like 20-30%.

>>108653314
thats only true of the older intel platforms and even then I ran a 1.65v kit on X79 for 8 years before I replaced it. The real issue is that all motherboards from on both AMD and Intel tend to ramp up the SOC / VCCSA voltages for RAM "compatibility" based on a logarithmic scale around the set RAM frequency even if you don't use EXPO/XMP.

Auto voltages are very fucking stupid and you need to always rein them in. Just setting the frequency by itself (no EXPO) would raise the SOC on AM5 to unsafe levels to the point where it was physically burning out the early 7000 series chips til AMD put their foot down and told vendors that 1.3v was the max.

Baiting enthusiasts into blowing up their ram sticks with XMP/EXPO is a coordinated attempt to sell more ram at today's inflated prices.

>>108651207
>cope, the post.

>>108653600
i'm talking about ram ics, I have hynix ddr4 that is known to degrade at 1.4V+
I'm aware of skylake imcs degrading with non L ddr3

>>108653916

I have used the default XMP voltage (1.4v) for over 3 years without any sign of degradation.

Currently: 1.395v

Shouldn't be any problem with Hynix A-die.

Also: you can typically resuce the voltage a lot from the XMP default unless you are doing an aggressive overclock (that is way past the XMP/EXPO).

>>108654054
why would he cope if he's not poor?

>>108651129
>i've seen modules spontaneously die from higher config voltage
bullshit
>>108651129
>DDR5 with non reporting ODECC silently masking degradation
nah youre definitely larping

>>108652770
literally since 2018 memory overclocking has become standard

you're probably just another samefag spamming this thread with luddite fearmongery bullshit

hence why some anon keeps spamming his website as well. this thread reeks of dylan patel and reddit hardware

>>108654621
ddr5 actively hides otherwise failed stress tests
all you get is proxies like microstuttering in certain games which imo is a pretty retarded method of measuring stability
iirc ddr6 fixes this by making error stats visible

>>108655136
It doesn't hide them, you can monitor memory faults from the OS, it just doesn't crash.
That's why you monitor faults while mem testing, not just checksums.

>>108655136
>iirc ddr6 fixes this by making error stats visible
You can already though, it's a memory controller thing.

>>108655152
>>108655156
i didnt know there was a stat counter for ddr5 ondie ecc that is accessible on os
maybe i should give it a look again

>>108655208
WHEA reports memory controller errors too, HWiNFO works.

>>108653259
I lack the required levels of autism to see whats being illustrated here.

Explain to me like im a normie smoothbrain.

>>108655472

15 to 33 percent performance gain over XMP with manual tuning.

https://vintologi.com/threads/ddr5-overclocking-nightmare.1229/page-2#post-9565

File: 1771375400125080.jpg (130.4 KB)

130.4 KB JPG

4chan does not exist for you to post twitter/reddit/other social media comments you're too afraid to reply to on the site they were originally posted to.

>>108651121
Modern hardware is really unstable. My computer would not post with XMP enabled. I miss the old days when computers had really high overclock range and tolerated it well.

>>108655554

Perhaps you needed to tweak some memory controller voltages or ODT/impedance settings.

>>108651121
No, it is more like you value stability and data integrity = peace of mind. Versus trying to min-max a tiny gain for tons of headaches.

>>108655554
Modern hardware is already tuned to almost max.

>>108655601

I clocked my ram higher than what msi claimed to support with dual rank (was possible thanks to bios updates).

Passed 20 hours of stresstesting.

Ill run mprime large FFT again later to see if it's still properly stable. Ran it for 1.76 hours earlier so it's probably fine but who knows.

>>108655678
Normal stress-testing is irrelevant with DDR5.
Unless you monitor for faults.

It might show it's stable and higher clocks / tighter timings, but actually errors so much the end result is worse.

>>108655678
>Try again in six or so months when the memory controller on the CPU or DIMM chips start getting flaky from the overvoltage/overclock
Memory overclocking is 95% meme and purely for epenis points at of the cost of instability and data corruption.
I have seen and heard so many horror stories from morons who overclock their memory wonder why their systems start to CTD and BSOD/kenrel panic after several months of usage.

>>108655668

My ram wasn't.

XMP: as 6000 CL32 @ 1.4v
Manual tuning: 6800 CL32 @ 1.395v

Could probably have done 7000 CL32 if it wasn't for my memory controller and motherboard holding it back. I probably would have had to losen tRP to get it stable at 7000 MT/s though.

tRP at 32 @ 6800 MT/s is about as tight as my ram could handle (XMP: 38 @ 6000 MT/s).

Then there is the massive performance gain you get basically for free by rasing tREFI.

>>108655689

Performance was fine in stressapptest (95000 MiB/s) and intel MLC.

Tightening tRP did significantly improve performance in stressapptest (gained over 1000 MiB/s by lowering it by 2 to 32).

>>108655699

It's getting close to 2 months now and seems pretty solid.

I run stressapptest for 30 seconds after each linux boot as a safety measure but hasn't found anything yet (which was to be expected).

running beyond the CPU's official memory speed (a measly 5600MT/s) is idiotic

I run on the XMP or EXPO configurations. I'm not going to waste my time hoping I can safely determine a correct and stable configuration for my hardware. Either the OEM did the work correctly and baked it into the XMP/EXPO config or they didn't and it's RMA time. JEDEC timings are fine as well honestly. Most memory OCs don't add huge performance.

>>108655750

Why?

I am running way past that and it's fine as far as i can see.

>>108655786
Shit degrades.
Most of the considerations for speeds and voltages is based on having margins to be stable at the initial performance level from now to many years down the line.

Just saying it works now misses a large point of why the non-overclocked spec is what it is.

>>108655833
It will take like a decade to degrade enough for you to have to dial anything back.
You'll still be running the same hardware? Wait don't answer, most faggots here are poorfags.

>>108655862
>It will take like a decade to degrade enough for you to have to dial anything back.
Are you confident on that timeframe?

>>108655833

Yea 1.25v CPU VDDQ is totally going to cause degradation.

>>108655882
So, de we just forget about the memory here?

>>108655881
Pretty much. My FX-8350 has been running at 5GHz with a 1.47V vcore for over 10 years now, with also overclocked 2600MHz DDR3.

>>108655893

I ran it at 1.4v for over 3 years without any issues. Should be fine with hynix A-die.

I have a fan on so the ram temperature remains below 60 °C.

>>108651121
This is the same as people who pay $300-500 extra for AMD equivalent Nvidia because they're too scared and clueless of PCIE cards

>>108651121

>>108651121
Unreal Editor crashes if XMP is enabled

>>108656837

Then you need to tune it manually.

https://vintologi.com/threads/ddr5-overclocking-nightmare.1229/

>>108656814
You'd be better off investing in new hardware than trying to push that outdated scrap.

>>108651681
Simply don't shut the pc down?

>>108651681

You need to change bios settings then so it doesn't retrain.

>>108656951

There isn't really anything worthwhile to upgrade to.

i bought 6000 rated ram but that was at xmp, default is 4800. xmp is annoying as fuck because any time the computer shuts down unexpectedly or windows updates it resets back to default then i have to restart to turn it on again.

plus when mh wilds came out it caused the game to crash constantly- but that could have just been capcom shittery.

also didnt ram manufacturers get sued because they listed the xmp speeds as if they were the "Default" speeds? what ever happened with that?

which software actually benefits from faster ram transfer speeds?

>>108652770
As hardware has progressed, the amount of performance you can squeeze out of it with OC has decreased. 15-20 years ago you could get another 50% perf out of your CPU by cooling it properly and OCing. Nowadays you hardly get shit since stock perf is very close to the max the hardware is able to give with non-exotic cooling, so it feels like enthusiasts are getting more and more unhinged and manually tweaking tens of parameters to get an extra 3.5% extra perf in real-world use.

It's kinda sad things have come to this. I used to browse OC forums and shit and I certainly don't remember people doing this kind of thing back when you could actually OC your 2GHz CPU to over 3GHz and such.

>>108657199
It's latency that Ryzen benefits from, specially non-X3D chips.

>>108657199
uhhhhhhhhhh

>>108657199
literally every software ever. at the very least it loads from disk faster.

File: 36598-raiva.jpg (37.8 KB)

37.8 KB JPG

>buy Corsair Hynix DDR5 advertised at 6400mhz
>set to 6400mhz in BIOS
>system crashes every 5 minutes, RAM errors
>set it to XMP
>system crashes every 5 minutes
>set it manually to 6400 and undervolt it
>crashes even more
>contact Corsair, retarded reps go "Is it on the recommended parts list?"
>No, it isn't
>"Well, then you should have consulted the recommended parts list before making your purchase."
>Ask them if they'll take the RAM back and exchange it for an identical model that's on the recommended parts list for my mobo
>"No, sorry, we won't exchange it. You should have consulted the recommended parts list."
>go to look at replacing the RAM
>prices went up 400%
JEDEC IT IS, THEN

I got bored of tuning
I need to do it again and see how much better I can make it
RAM is really annoying to stress test, you need to stress all your components simultaneously due to how thermally sensitive DDR5 is
I'm pretty sure my CPU OC is very slightly unstable too, it froze after 1 month+ of use

>>108657498
>at the very least it loads from disk faster.
with theoretical bests, even DDR4 at 2133MT/s is not overwhelmed by a gen 5 nvme.

File: giqfovxrme5c1.jpg (47.7 KB)

47.7 KB JPG

>>108657635
you wouldn't do something that is not recommended, would you

>>108652786
sure, i hope you paid with a credit card though
its not worth going to court over consumer electronics and retailers know that

>>108657635
they have to tell you about the recommended parts list before your purchase

>>108658837
>>108659125
When I built this system I was upgrading from DDR3, 4790k and a 1070. In those glorious days it didn't matter what fucking specific die of chip your RAM had in it, you just fucking plugged it in and it WERKED. THIS IS BULLSHIT!

File: Screenshot_20260410_210125.png (109.1 KB)

109.1 KB PNG

Rate my timings anons. Been running with this for the last 6 months or so, rock solid.
Room for improvement? I don't really push my system to the limit very often, so I'd prefer stability over another 0.5 ns reduction in latency.

>>108657635

Typically when the XMP/EXPO profile doesn't work it's due to the motherboard or the CPU rather than the ram itself.

So replacing the ram probably wouldn't work in the first place.

AMD CPUs specifically have trouble achieving 6400 MT/s in gear 1 and you often end up having to manually tune bios settings to get it properly stable (and even that isn't guaranteed to work).

>>108659738

You can probably get better timings stable at 6400 MT/s (if your CPU and motherboard can handle that).

>>108659766
I thought the memory controller on Zen 4 worked better with 6000 compared to 6400? Maybe next BIOS update I'll fiddle around with higher speeds.

>>108659890

6400 MT/s is better if you get it stable. Works better with FLCK (seems like 3/2 ratio between them has some performance benefits).

Even better is of course if you get 6600 MT/s (in gear 1) stable with 2200 FLCK but good luck with that (requires both imc luck and manual tuning in addition to having the FLCK be stable that high as well).

>>108651121
>ramchyld
>afraid of xmp
How appropriate.

>>108651121
Wait what? I never heard this. I just heard that if XMP or EXPO don't work, try the other or some shit. So I turned EXPO on and it worked fine?
And now I'm hearing I need to undervolt it?

>>108660007

Lower voltage is safer generally but i don't know if there is actually anything you need to worry about in your case.

But even if there isn't any problem with instability or degradation you are still leaving performance on the table by not manually tuning it.

From what i have seen most people don't bother doing a maxed out manual tune but even a low effort one does tend to significantly outperform the XMP/EXPO performance.

>>108660035
What do I need to do? My vidya seems to run fine enough so I can't imagine it will do that much better.

>>108659360
alright so claim your damages and shut the fuck up

File: 1776620398812971.jpg (142.2 KB)

142.2 KB JPG

>tfw didn't buy ECC memory in 2019 and 2023 because XMP is a dangerous meme

>>108651239
damn I have been running my 3200/cl16 kit at that voltage for years

>>108660115

Depends on your hardware.

https://vintologi.com/threads/ddr5-overclocking-nightmare.1229/#post-7217

You can probably get better performance at lower voltages with manual tuning. The only thing you lose is the time you spend getting it to properly work (especially if you go for proper stability which requires hours/days of stresstesting to confirm).

>>108659744
My system specs:
>Core Ultra 7 265KF
>MSI Z890-P WIFI
>Corsair CMH32GX5M2N6400C36 (16GB x2)
>RTX 4090
I just wonder if there's a way to bring this particular chipset RAM up to 6400 without raping it. The mobo is newer than the RAM, I just assumed that it would naturally be compatible but of course DDR5 rapes me. The system performs fine at JEDEC 4800MHz, perfectly stable, but I feel gypped by this bullshit.

>>108660423

None of that should have any problem running 32 single rank ram at 6400 MT/s. In fact you should be able to go a lot higher than that.

Perhaps you simply got screwed over by corsair.

>>108651138
>Manual tuning is safer than xmp

The whole point of xmp is so you don't have to manual tune

My ram is on the qvl list for my motherboard so

>>108660456
That was my thought exactly. But for some reason this shit refuses to work at any setting other than JEDEC. And given that RAM is twice the price at which I bought it, I can't just replace it willy nilly. Fucking sucks man.

>>108660513
>>108660456
I had to go back and double check my order. This RAM kit I bought was $98 for 32GB DDR5 in November of 2024. The exact same kit today is $469. HOLY FUCK.

>>108660423
>intel
>cant even go higher than 6000
lol i thought you could go as high as like 8k with intel aviv and that was the only thing it had against x3d

anyhow, loosen the timings and/or up the voltage you animal

or settle for 5600/5800/6000

>>108660513

It makes no sense at all.

You bought:

ram rated 6000 CL32
Motherboard claiming support for 6600 (with dual rank)
Raptor lake CPU

I run: 6800 CL32

You bought:

Ram rated at 6400 CL32
Motherboard claiming support for 9200 MT/s (with single rank)
Newer Arrow Lake CPU (should be better for memory overclocking)

You got: 4800 MT/s

Even putting the ram in the wrong slots shouldn't produce such a bad of an outcome.

>>108660572
I tried every BIOS selection between 4800 (JEDEC) and 6400. Only JEDEC was stable. It's fucking bullshit man.
>>108660753
See above, I'm still mad about it even though it's been almost 2 years since I built this rig. Still, 4800 DDR5 is leagues faster than DDR3, but god fucking damnit man. From my research, this particular RAM has an inferior chip (Hynix A? Might be wrong) but I never imagined the particular chip my RAM shipped with would ever have an impact on RUNNING AT THE FUCKING ADVERTISED SPEEDS. OCing I understand, that's fine, but what the fuck.

Should I give it another whack? What are the settings I should go for to ensure this cursed RAM kit runs like it should?

>>108660796

I have hynix A-die and it overclocks just fine.

It should be able to do way higher than 6400 MT/s.

My guess is that one of your ram sticks is faulty. If you can demonstrate that you have a better shot at gettming an RMA through.

Is small claims court an option?

>>108660820
Not really. I exhausted every angle I could with Corsair while my return period was still active (I bought them through Newegg). It's been a year or two but I swear I tested each stick individually and both displayed the same instability at 6400MHz, which led me to that stupid exchange with Corsair support, but I might give it another try.

At the end of the day, I got 4800MHz DDR5 for $98, which is a fucking steal at current prices today, but it's incredibly irritating. There's no reason this shit shouldn't run at advertised speeds. And it's perfectly stable at JEDEC too, with uptimes of over 2-3 weeks.

>>108660853
did you pick the "xmp" option or just set the "speed" to 6400

>>108651121
I have custom ram timings, but I am running at jedec frq at 5600 MHz.
Getting similar latency at 6000mhz or higher requires ridiculous voltages. Somehow 6000mhz expo doesn't even have noticeable performance increase in benchmarks. It does reduce stability though, games very rarely crash at expo settings but memtest passes no problems.

>>108661721
I guess it's the extra heat from your cpu and gpu that makes it unstable

>>108660796
Have you done manual settings? For example, running XMP voltages and frequency with JEDEC timings (assuming they are higher than XMP timings, which they usually are)

>>108660753
It makes sense once you realize RAM can go bad or be miss binned.

>>108655497
and this man spent how many hours manually tuning/stability testing?

However much time was saved in faster memory is absolutely going to be lost in the literal weeks of a system doing jack fucking shit but benchmarking.
If your workload is the sort that will benefit - and you will be performing that workload 24/7 for years, go ahead. Probably worth the time/effort investment.
But if you deploy at scale, or you're just doing some 9-5 work, some very simple multi-task workflow will easily make up the difference, not the least of which is mind share on the tasks where this guy's investment in memory tuning could be spent researching literally anything else?

Don't get me wrong, there's plenty room for a hobby, and in this context his hobby WILL benefit him in that 9-5 situation. So if we're talking the difference between memory tuning over playing videogames, there is a clear unequivocal win. But if we're just talking shit posting neets or workaholic efficiency maxers, the former is irrelevant, and the latter has WAY better things to do with their time - again, outside that always-on 24/7 constant workload machine.

considering how much vomit I contributed you can guess which category I fall under. Draw conclusions from there

>>108662804

Maxing out DDR5 generally takes way too much effort but i wanted 6800 MT/s since that was a better fit for the timings.

Had i gotten 6600 CL30 stable at 1.4v (or less) i would have gone with that instead but that required higher voltage for worse performance than 6800 CL32.

You can leave a stresstest running over the night but typically when it's stable it will error out fairly quickly.

mprime large FFT was a bitch. Failed an overclock once after it had gone through 12 hours of stresstesting without errors.

But even a low effort overclock tends to be significantly better than XMP even if you leave like 5% performance on the table.

File: what_the_fuck?.png (1.2 KB)

1.2 KB PNG

>>108651121
you are a homosexual gay lord negro and should go back to israel jeffy

>>108662972
CL30 at 6600 is 9.1ns vs CL32 at 6800Mhz which is 9.4ns so CL30 at 6600 is higher performance in memory latency if the work load benefits from that. It's not worse performance. Only worse bandwidth, specifically.

>>108663345

Latency was basically the same in intel MLC but 6800 CL32 had better bandwidth.

6800 CL32 did need less voltage making it better overall even if it was a nightmare to get properly stable. In fact there is a good chance i could have gotten it stable at 1.38v DRAM VDD/VDDQ

6600 CL30 would have needed a bit more at like 1.42v dram VDD (would probably had been safe).

>>108659104
>its not worth going to court over consumer electronics and retailers know that
thats why they always settle
costs a consumer a mere £80 to file a court claim

>>108663421
i dont live in a country with a functioning legal system

>>108663427
that sucks
i work for a furniture retailer, and any time a consumer does a small court claim we always settle it because the lawyers cost more than a £200 coffee table and we will most likely lose

>>108663439
we only have small court claims available for international purchases
otherwise, chargebacks are often sufficient
they still cost the merchant €30 regardless
i suppose thats punishment enough

>>108663405
Of course 32 at 6800 is going to need less voltage, it's way harder to get latency down than it is to get frequency up. Typical DDR memory CL latency tends to be around 10.5ns. It has been like that ever since DDR entered the market.

You should be easily able to run CL31 at 6600 at similar or lower voltages because that'd translate to same latency as 32 at 6800 unlike 30 which is way lower.

my 9900x died on Asrock mobo. I was on AMD Expo but people were not blaming it.

>>108651138
this is totsl bs lmao

>>108663495

Odd values for primaries didn't seem to work very well (i used gear 2 and command rate 2).

>>108663345
you're forgetting that the memory controller runs faster at 6800

>>108652022
an adult man owns that computer, and he's not unique. if you ever feel sad or angry about yourself just think of that picture and realize that you're actually better than a lot of people.

>>108656951
you felt really cool when you typed that, didn't you? you probably imagined yourself doing a cool pose with your arms crossed or something. i can't believe you did that.

I boughted 64GB Kingston DDR5-6400 + Ultra 265k but running it on XMP at 6400 makes youtube freeze randomly.
So I put it to XMP 6000. Now YouTube works fine but it will still freeze the machine every couple of days. I put it to JEDEC at 4800, and still freezes every couple of das so I might as well leave it at 6000.

I had an 8700k + 16GB DDR4-2133 that worked without issue for almost 10 years.
Modern hardware is just garbage

>>108663862

I was also able to keep most timings as is at 6800 MT/s. I did have to loosen tRCD and tRAS by 2 but i was able to keep tRP at 32 (barely).

I probably could have gotten it stable with tRAS = 30 but i didn't see any performance benefit with lower tRAS.

And of course there was no longer any chance of getting tCL stable at 30 (i think it even failed training last time i tried that).

>>108663903

Perhaps you need to lower CPU VDDQ or something. You can also try manually tuning ODT settings (you could try copy mine).

The raptor lake memory controller also struggled with dual rank ram and i guess the arrow lake one isn't actually much better despite being better on paper.

>>108660796
loosen the timings and/or up the voltage you animal. that means manually, not xmp profiles

>set my RAM to the tiniest overclock possible
>ZFS starts reporting failed checksums during scrubs
Yeah no, fuck you overclocking niggers

>>108664593

Then you bought the wrong hardware.

I gained around 20% performance with my ram and it passe 20 hours of stresstesting.

>>108664614
You're right, I should have bought a winning lottery ticket instead of a losing one. How foolish of me.

>>108663439
why do you keep scamming your customers?

File: kekd.jpg (5.9 KB)

5.9 KB JPG

>overclocking memory

>>108664624

You can very easily find ram that is basically guaranteed to be M-die or A-die with DDR5 and then it simply comes down to how good your CPU and motherboard is.

Btw: i have a feeling your ram is unstable even at XMP/EXPO.

>>108652225
>i combined a faster and slower kit at the same time so i can't mess with the timings i think
Based. Next start hot swapping ram dims for improved performance.

>>108664642
not replacing a product 2 years outside of the guarantee is scamming?

>>108664624
the silicon lottery is real but overblown
and statistically speaking your hardware is likely just average

but i guess everything is a lottery if you're retarded

>>108664772

They do a lot of binning which reduces the impact of the silicon lottery.

Main thing that isn't binned is IMC quality where you can be badly screwed over on the intel side.

>>108663989
yeah the point is i dont fucking wanna do that. I just want a computer that is fast and stable. Emphasis on stable/

All of this BS already took way too much of my lifetime especially since a full memtest=86 takes hours and it turns out is also completely useless because your system can pass memtest but still randomly freeze up

>>108651162
>force every worthless piece of shit into ram. ram gets so bloated that changes in application force the whole fucking lot into endless paging
>UNUSED RAM IS WASTED RAM
>tinker tranny the shit out ram, reduce lifetime and stability or outright fry it due to uncontrollable autism demand impercetible changes in frame rate
>UNUSED RAM IS WASTED RAM
>design a web browser that fills up every spare megabyte with shit the user might need some point in the future, possibly, fuck what they actually want to use it for. page files for africa. fry ssd due to constant fucking around between memory and storage.
>UNUSED RAM IS WASTED RAM
it's all so tiresome.

>>108653011
most of these guys don't have anything else. so the persona they've created around irrelevant and pointless fucking about with hardware is how they define themselves as seperate and special. it's probably just arrested developement and a maladapted competitive instinct. what you get when you spend all your time inside.

>>108651151
If you care that much about stability why aren't you using ECC RAM?

>>108651138
ive been using xmp on my 9950X3D its pretty stable it seems. I never really felt the need to manual tweak but i am debating if I should

>>108665154

memtest86+ is mostly a wasteful time.

Only thing i ever found it useful for was learning that i needed to loosen tRAS.

You need to both stresstest the ram itself and the memory controller. mprime large FFT seems to be the most efficient at stressing the memory controller (prime95 large FFT if you test in windows) while testmem5 1usmus5 is good for testing the stability of timings.

Other useful tests are

Y-cruncher (there are different versions that come with different tests)
stressapptest
memtestpro (aka hci memtest)
OCCT (annoying but kinda needed if you only use linux, redundant otherwise).

>>108665264

You could do some low manual tune where you mostly stick to the XMP settings. Might gain you like 15% ram performance.

Seems like a lot of people just raise tREFI to 65535 and then call it a day.

>>108665186
I only use it for my browsers, Mullvad VPN, Jellyfin, keepassXC and qbittorrent

>>108660920
Both, both crashed.
>>108661799
I will try that when I get a chance, thanks anon.
>>108664589
I'm still stuck in the DDR3 age brother. Every set I bought up until this DDR5 bullshit worked flawlessly if you just set it to XMP/advertised speeds in the BIOS. I will do my best and try to learn the ways of RAM-fu

>>108651190
This. My desktop is from 2014 and uses DDR3 RAM. I don't have to "tune" or "train" anything, it just werkz. What are modern PC manufacturers even doing at this point?

File: generic boring pc.jpg (304.5 KB)

304.5 KB JPG

>>108663903
your os is probably corrupted a bit, or your OS storage is malfunctioning. I don't know why you are blaming your ram for this issue, maybe reseat the ram?

>>108663871
>what? you want to add personality and flair to your custom built, high end computer that you use for thousands of hours every year? thats fucking GAY bro, your pc needs to look like this or autists on 4chan will judge you, like me.

>>108666523
It works, just slowly. People have been tuning ram for decades because it makes your computer faster for free.

>>108664823
>IMC quality
people have been talking up intel-avivs IMC quality for years on this general, especially in comparison to AMD, where is your rhetoric coming from?

>>108665264
so long as the memory security thingy that winblows 11 likes having is operating fine, there's no need to switch from XMP to EXPO, it works fine on AMD. I have the same cpu.

Isn't it more related that he doesn't want his proc memory controller be fried by the asrock mb?

>>108667576

You can manually set memory controller voltages to prevent that from happening.

CPU VDDIO and VSOC specifically.

>>108667600
>to prevent that from happening.
But it doesn't prevent it, it's superstition. These asrock boards are am5 killers.

>>108651121
I am on DDR4 with XMP, what do you want from me?

>>108667643

It will prevent it from happening due to the ram.

Might still fry the CPU via some other mechanism in which case running JEDEC isn't going to prevent that from happening either.