Loki

God of Mischief, autisme, dyslexia and chocolate milk.

Someone whose very existence seems to be a conversation between the mundane and the mythological.

-Grok

[Loki @LokiLuna4]

@NVIDIAGeForce #RTXPowersPlay

[Loki @LokiLuna4]

@ShaneFrakes This made me chuckle 😹

[Loki @LokiLuna4]

@ShaneFrakes Based xD

[Loki @LokiLuna4]

@KamiOKingPaimon Intriguing way to do it, I get 7 by doing the same🤔

Also realise I'm exactly 1 year and 1 day older then you😹

[Loki @LokiLuna4]

@sudoingX 64gb ddr4 8x8gb with RGB 😹👌

[Loki @LokiLuna4]

@sudoingX Glad you like it! :)
I'm currently building a second pc using intel 7600k(sone to be 7700k) with 32 gb ddr4 and 2x strix 1080 ti as well as plans to move the 7600k to a itx with 16gb ddr4 and a 1x non strix.
In total I currently have 3x ti strix 2x base strix and 1x normal ti😹

[Loki @LokiLuna4]

@eplurubusnullus @sudoingX Thank for the advice! :)
I have been using earlier iteration of the model for about 2-3 months and with my current setup it's running very stabel with 256k 8q context thru hermes with around 33 t/s on average.

So I guess I ran my own test so to speak😹

[Loki @LokiLuna4]

@eplurubusnullus @sudoingX Highly recommend testing it out! 😺
By experiance using Ubuntu server 24.04(you can use 26.04 but more of a hazel to get llama.cpp to compile) with cude 12.8 or 12.9(both work).

I'm using this model currently: https://t.co/lZ19mEiIpT

[Loki @LokiLuna4]

@Maat_777777 A ffs not another one!😹

[Loki @LokiLuna4]

@sudoingX Currently running 3x 1080 (1ti 2base) strix with amd 16 core and 64gb ddr4 with hermes thru llama.cpp and qwen 3.6 35B mtp on q8 and 256k context witc cmoe.
Not extreme speed 30-45 t/s but works very well and stabel even across long sessions and context :)

[Loki @LokiLuna4]

@ShaneFrakes I had a very autistic derp here "what do you mean project rubik's cube is fake? I definitely sat down many years ago and learnt how to solve them!?" Keeps reading "ahh, that makes a lot more sense😅"

[Loki @LokiLuna4]

@ItsmeAjayKV corsair hx1200i or hx1500 would be solid choices 🤔

[Loki @LokiLuna4]

@ItsmeAjayKV Most 1200w raded platinum/platinum pluss should do the trick, might also depend where in the world tou are i have multipl kolink 1200w but cosier 1250 should be good.
If tou have problems on full load i would recommend maybe getting a 1500 w🤔

[Loki @LokiLuna4]

@FStrongpaw @LilithDatura Ive actually seen speed improvment with ram offload(should be said im using 3x 1080s) especially one moe(like qwen 3.6 35b a3b).
And you can really push the context length(512k context on 3x 1080 8gb and 64 gb ddr4).
Also work on single card or doodle card.

[Loki @LokiLuna4]

@DionysianAgent Out of curiosity, where would you place your self on thos axis in relation to your own pysical body?

[Loki @LokiLuna4]

@palrestad Heilt utrulig kor vekskjemt holding folk har, plus en heilt absurd følelse av og ha krav på alt(meg meg meg!).

Eg finne dette berre morsomt på ditta tidspunktet😹

[Loki @LokiLuna4]

@ShaneFrakes You should check out One Piece!😺

[Loki @LokiLuna4]

@sudoingX I'm root and break things on debian!😹👌