>>108563931
they probably forgot to set their cheapo 3200cl16 kit to 3600cl16

>>108563851
>3200MHz
lmao

>>108563948
>HDD (SATA)

File: A.jpg (4 KB, 187x124)

4 KB JPG

>>108563851
>he think his ramlet setup is special
I regret not going threadripper for more upgrade headroom now I'm stuck

oh well upgrading right now is too expensive anyway

>>108563991
Obviously storage, his OS drive is SSD.

>>108563994
>>108563851
I have more VRAM than you guys have RAM stop being cringe

>>108564253
>I have over 4 dead GPUS with collective VRAM over your active RAM
>just stop being cringe bro

Prove you have over 128 on a single machine, pa/g/eet. Sounds like delusional Indian gloating to me.

>>108564253
So does other people with more everything then you
my point is OP is underage

>>108563944
>>108563948
NTA but I bought one 16GB stick like three years ago. It was cheap and crap, so now the speed is 2667 MHz.
I literally didn't notice a difference. Local AI runs the same, performance is the same.
So yeah, it's a meme, you're a meme, and thank you for coming to my TED talk.

>>108565309
this is a lie. memory speed matters.

File: NOOooooOOOoo NOT MY HECKE(...).jpg (18 KB, 220x336)

18 KB JPG

>>108565463
Whatever gives you a false sense of superiority, I guess.

>>108565531
i've spent the last week trying to eek out every tenth of a token i can for local llm. tuning my memory had real gains.

>>108565309
lmao it is not
huge ram+ok vram allows (albeit slowly) 70B tier MoE to run
I do it when I need full or extended context
fitting vram is when you wanted speed
with no ram you simply dont have that option

>>108563851
Renting is cheaper. Just have good spending limits set to detect runouts.

>>108565723
>>108565309
oops misread nvm
ram speed doesnt matter
compared to vram/unified memory ram is slow as shit anyway

>>108565737
>trusting other people with my data
no thanks. AI companies have already demonstrated that they don't care about the law regarding data

>>108563851
>>108563994
>>108565531
That's cool and all, but do you actually do something productive with that other than post it on a dying imageboard moderated by pedophiles?

>>108565862
Hope you get kneecapped AND cancer.

>>108566041
Wtf? That is so mean...

>>108566053
post shit, get shit

File: 1760819891076137.png (163 KB, 718x133)

163 KB PNG

I regularly use 40-42GB of RAM when playing games (while watching youtube and torrenting in the background).


I can't imagine anyone building today with 16GB or 32GB only.

>>108566077
I am crying now.

>>108563851
ramlet

$ doas lshw -class memory -short
H/W path              Device          Class          Description
================================================================
/0/0                                  memory         64KiB BIOS
/0/b                                  memory         1MiB L1 cache
/0/c                                  memory         16MiB L2 cache
/0/d                                  memory         64MiB L3 cache
/0/11                                 memory         64GiB System Memory
/0/11/0                               memory         [empty]
/0/11/1                               memory         32GiB DIMM Synchronous Unbuffered (Unregistered) 6000 MHz (0.2 ns)
/0/11/2                               memory         [empty]
/0/11/3                               memory         32GiB DIMM Synchronous Unbuffered (Unregistered) 6000 MHz (0.2 ns)

>>108566104
h-hot

File: file.png (225 KB, 1105x358)

225 KB PNG

>>108564462
not him, but here

>>108566299
based
I'm waiting for yall to dump it cheap when it finally pops

>>108565531
>3 of 4

>>108566639
nah. i am planning on buying more. just bought a 512gb kit of ram despite the insane prices.

>>108566647
It's a secret and you shouldn't do it because of timing mismatching but you still get dual channel if channel A has a 16GB stick and channel B has two 8GB sticks for example, as long as there is equal memory in both channels.

>>108566746
It's honestly not even all that bad, the biggest problem is stability, you're not going to be OCing that kind of setup much if at all.

It also puts extra strain on the IMC, but it's generally not significant.

The performance is within margin of error when compared to 2x16GB of the same speed/latency.

>>108563851
>3200/mt
lol

File: 1x1_1200x1200_highres-lin(...).jpg (113 KB, 1200x1200)

113 KB JPG

You kids better be using ECC