/biz/ - Business & Finance

>>54228235
Kaspy?

>>54228235
I was thinking about running an uncensored ai chat model and pimping it out for money. Seems sketchy though.

>>54228235
just get a 4070ti it runs all games un ultra at 100+ fps

>>54228348
neo gpt is a steaming pile of shite though. any other alternatrives that jew can run at home?

>>54228500
The Llama models are pretty good. 30B has shown pretty incredible results although I still have to tinker around with it for a bit.

>>54228537
thx anon gongacheckitout

>>54228235
How would you sell it? Just Meeting dudes in an alley on craigslist?

>>54228662
np. check out /aicg/ on /g/ and /aids/ on /vg/. youre going to want to run the models with ooba/textgenui.

>>54228235
I have a 3090 and my PC crashes every 5/10 games.

I have a 1060 from 7 years ago and it plays all modern games just fine. Don't fall for the meme. You're doing the equivalent of buying a Ferrari to drive to work and back, except you can't even brag about it. No game exists that requires a card from the 40 series.

>>54228837
fucking legend. thx!

>>54228235
I have a 2080 and it still runs every modern game I throw at it. The only reason why you'd want a top end card like that is for AI. And no, you're not going to make enough money off selling your chatbots to randoms as you'll be undercut by every datacenter
But as always, people love chasing bigger numbers for various reasons, so do what you will

>>54228235
If you have to ask you definitely shouln't buy it

>>54228879
That's because you cheaped out on your RAM. Tick the timing down a notch in BIOS.

>>54228917
It depends on your monitors resolution, a 1060 can't even play games from 7 years ago on 4k.

>>54228235
use the money to buy NVDA calls

>>54228235
I bought a 3090 for $3.3k in May 2021

>>54229684

worst part about the 3080/3090 is that the VRAM would get so fucking hot, literally 120*C (while mining ETH) it would throttle the card and most certainly lower its lifespan. I had to open the card up and manually replace the pads on all the memory modules, which took 3 tries or so to get the right sized pads for my particular manufacturer (NO consistency in pad thickness between manufacturers). I had to buy a Zotac card because that's all that was available on the market at the time. Wanted an EVGA because they had the best performance/build quality based on test results but they remained on backorder for another fucking year after I got the Zotac.

Do plenty of research before you buy a top end card model. There is a lot more variation between manufacturers than you might expect, and the difference between a good vs bad manufacturer is the difference between whether or not you will need to crack open the fucking card and LARP as a hardware engineer. Also you may just get a model which, regardless of manufacturer, has some new gen components in it (le latest and greatest!) which have yet to be battle tested. That was the case with the 3080/3090 which had next gen VRAM modules that overheat like a motherfucker. Kinda like the early AMD Ryzen chips, sure they are fast but they will suicide themselves in the process if you let them. Would have been better off getting a 3070 and not having to worry about my card setting my house on fire in the first place.

Higher specs, not always better.

>>54228537
I'm waiting on my card to get in to try to get alpaca 30b running. Hows llama been compared to chat gpt 4? /g/ is so filled with angry anime trannies right now it's hard to have a sensible discussion on it

>>54229684
was virtual girlfriend worth it?

>>54229684
i bought a 3080 for $1.7K in Feb 2022
literally less than a month b4 the crash

>>54229684
>>54230097

>>54230094
pls no bully

>>54229710
>>54230523
bonus round: around the same time, I also bought a network NAS, filled it with 5 10TB drives for a total of ~$2.6k and mined Chia for all of 2 months before it crashed so hard it became unprofitable.

That NAS hasn't been turned on in almost two years now and the hard drives are worth half what I paid during peak shortage.

>>54228235
yes, when shit hits the fan and the internet goes out then you can still produce infinite porn with these things.

>>54228235
gpus are no longer money printing machines, but the price they're sold for essentially assumes they are
I want a new gpu, but there's very little point in getting one for how expensive they are right now
there's really no new games worth the price and getting more fps in old games isn't really worth it either

>>54228348
even then wouldn't you be better off renting gpu usage? then it's scalable and your users won't lose service if your internet goes down

>>54228235
>want to get a good gaming rig
>have some sort of epilepsy that makes me retarded when looking at certain computer screens and computer monitors
>also happens when I’m exposed to fluorescent lights
Jesus fucking Christ just put a bullet in my head

>>54230039
Haven't used GPT 4 yet to be honest.
>so filled with angry anime trannies right now it's hard to have a sensible discussion on it
I was under the impression that people had all the Llama shit figured out until I had to figure it out myself. There's maybe like a handful of people who actually know what they're doing in those threads with the rest populated by normies. Makes sense I guess as not everyone has expensive hardware to run their own models.

In regards to 30Bs quality, it's really really impressive sometimes. Other times it needs to be wrangled to give you the outputs you want. I'm still trying to figure out the correct generation parameters as well as how to properly format the context to get the type of outputs I want. Of course /aicg/ and /aids/ have been useless as they're all busy cooming. Like I've said, there's probably like a handful of people who actually have this up and running.

In regards to alpaca, I haven't used it yet. Been too far into figuring out 30b and not enough time to implement alpaca.

>>54230993
Thanks for the rundown fren. Yeah I am kind of waiting for 30b alpaca and the 4090 to get in, but I think the real game changing aspect of alpaca is just the lora training process. Theres gonna be all sorts of different llama forks that are fine tuned for different shit, I think as long as llama 30b is showing good potential they should be able to work the kinks out. But I'm basically retarded and am just now learning about all this myself, it's kind of astonishing the rate at which things are developing now.

HOW THE FUCK DO I MAKE MONEY WITH THIS THING WHY DID I BUY IT CRYPTO MINING DOESN'T WORK ANYMORE WHY DID I SPEND 5K ON THIS RIG I COULD HAVE GOTTEN 5 ETHS INSTEAD IM MISSING THE ENTIRE PUMP AGAIN
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAHHHHHHHHHHHHHHHHHHHHHHHHHHHHHHHHHHHH

>>54228235
I bought a 3090 for $3200 and it was worth it