I can't justify building a desktop just for video games. Post cool things I can do with a powerful computer that aren't video games.
I can't justify building a desktop just for video games.
I can't justify building a desktop just for video games. Post cool things I can do with a powerful computer that aren't video games.
Porn upscaling
mine crypto
run AI simulations
donate the power for protein folding
>Post cool things I can do with a powerful computer that aren't video games.
Opening directories
IDLing in front of a monitor
Jerking off to the sound of your overpriced cooling system
Getting epilepsy from RGB lights
Fisting yourself with an overclocked geforce 9999 XT Nitro Ultra Pro Ti
Making some bacon & eggs on your i9 heat spreader.
that image is funny
oh man my sides
what a fricking dick. op legit makes a good post, this b***h ass Black person, says the dumbest shit?
anyways, i use my radeon with davinci to edit videos from my sony camera. shit is accelerated. kinda dope. back in my day we didnt even have this kinda shit so..
Buy midrange components
The midrange is for the midwitted.
>consoooome 4090
Of course not. That would be totally idiotic.
I was referring to the 7900XTX.
The ultimate cope.
>running open source LLMs
>automate generating images with stablediffusion
>transcode videos for a movie library, realtime and for storage
That's what I do on my 1080ti. I can only do one at a time because of vram limitations but the fact I can do all this with a 6 year old gpu is amazing to me.
Honestly it's the cost of electricity that stops me from buying more compute and running it 24/7. Otherwise I'd have sillytavern, piper/bark and SD running all the time.
How much would you feel is worth spending on a computer to do these things, if you started from zero and had to buy it all today?
- 7950X = 550
- 4090 = 1700 = 2250
- 128gb ddr5 6000/30 ram = $400 = 2650
- motherboard that will support all of it without problems = 200 = 2850
- 1000w psu = 150 = 3000
- adequate case = 80 = 3080
- 2tb fast nvme = 100 = 3180
- good heatsink = 35 = 3215
computer itself = $3,215
- 4k/144hz monitor = 400 = 3615
- decent audio = 200 = 3815
- keyboard, mouse, webcam/microphone or whatever = 100 = 3915
total computer setup = $3,915
chair and desk can be anything from 50 to 1000+
>- 7950X = 550
replace with 7800X3D for $300
>- 4090 = 1700 = 2250
replace with 7900XTX for $850
>- 128gb ddr5 6000/30 ram = $400 = 2650
go down to 32GB for $100
>- motherboard that will support all of it without problems = 200 = 2850
>- 1000w psu = 150 = 3000
>- adequate case = 80 = 3080
>- 2tb fast nvme = 100 = 3180
>- good heatsink = 35 = 3215
>computer itself = $3,215
>- 4k/144hz monitor = 400 = 3615
>- decent audio = 200 = 3815
>- keyboard, mouse, webcam/microphone or whatever = 100 = 3915
>total computer setup = $3,915
>chair and desk can be anything from 50 to 1000+
down to $2,515 with no appreciable difference beyond getting rid of Nvidia and increasing game performance with X3D part, and let's be frank, 99% of people aren't utilizing a 7950X.
you can do everything listed but LLMs on a fairly modest computer (<$1000), you don't need much to play games/run stable diffusion/transcode video with acceptable performance
for LLMs you need at minimum a 3090 to have a good time, preferably two
I'm willing to just skip LLMs. They seem bad unless you go the whole way with 2x24GB 3090 or 4090 systems.
>for LLMs you need at minimum a 3090 to have a good time, preferably two
7b LLMs are all the rage now beating out some 13b models and even reaching 30b quality. It runs ok with cpu+ram but can hit above reading speeds with <10GB vram. There has never been a better time for small models.
Dual gpu setups are for merging your own models or running 70b+ size. Unnecessary unless you're using it for freelance work.
Small models are fricking horrendous and frick you for pretending they're remotely comparable.
I'm running 70bs right now, haven't bothered with the mistral tunes tbqh
but even the best 13bs I've tried required so much tard wrangling that I wouldn't call it an enjoyable experience
What is your workload? I'm tinkering around with 13b-tiefighter and 7b-zephyr and they're both pretty impressive for general chat and light coding like bash scripts.
agents, code completion, story writing and RP (among other things, I integrated it into emacs and tell it to do various things, summarization, refactoring, etc.)
30b is good enough for code, agents can be tricky, but using one of the various logit force-fitting strategies (llama.cpp grammars, guidance, etc.) can make a 13b usable for some tasks (I still wouldn't have it interact with anything important), although I typically use turbo for agents still
and for story writing/RP, anything less than 70b is too dumb for complicated/realistic characters imo
> integrated it into emacs and tell it to do various things, summarization, refactoring, etc
Pretty cool anon, I have to get into more integration stuff. For now all I have is krita+stablediffusion to play around with. I wanted to build a RAG system for my nextcloud/dokuwiki instance but I'm so VRAM limited it might be useless.
>30b is good enough for code
I could barely run 30b llama1 on my 1080ti but coding didn't seem to work well. I gave up on anything useful and just use it as a regex/bash generator.
>for story writing/RP, anything less than 70b is too dumb for complicated/realistic characters imo
I guess I'm not really doing complex rp just testing out <2k context and moving on to something else. For a few lines without complex characters 13b was fine.
>I typically use turbo for agents still
You mean gpt 3.5 turbo or 4 turbo? Isn't 3.5 basically useless now?
>RAG system for my nextcloud/dokuwiki
that's a cool idea, I kinda have something like that in my emacs extension. You can use a vectordb to match files/load those into the llm's context, (still somewhat of a WIP, I'll put it on github eventually), I wrote it because my attention deficient brain will turn off if I have to flip through too many papers, it's really handy for academic writing.
>I could barely run 30b llama1 on my 1080ti but coding didn't seem to work well
I mean 34b for code, some of the tuned llama-code models are pretty good (phind v2 in particular), not perfect, but good enough that it's a help rather than a hindrance
>For a few lines without complex characters 13b was fine.
yeah 13b is good enough for known characters or characters that don't have ulterior motives (which is what I like to do with the RPs)
>turbo
3.5 is still useful for agents. but yeah, it has really gone downhill for everything else, I've been trying out gpt-4-turbo recently, but I haven't used it enough to really have a good idea of how it performs
I'm still using an old 1080ti too, I just also bought a 3090, can get like 8-9 T/s on Q3S 70b and 4-5 on Q4S. It's alright, but I'll probably buy another 3090 once inebriated enough to make poor financial decisions.
>it's really handy for academic writing
I didn't even think of that. I would archive papers if a model could summarize and explain them for me. I'm too moronic to understand them as is.
>some of the tuned llama-code models are pretty good
I check the coding leaderboards every couple weeks just to see if there's anything new. Chat models are good enough for regex and nginx configs but I could see a usecase for making better ansible playbooks using a tuned model.
>ulterior motives
I avoided complex characters because of how much work needs to go in building backstory and your character. I'd rather a bot just spit out smut rather than spend time writing a flawed yet well rounded character for the ai to bounce words off of. I see the appeal but it's too much work for my taste.
>3.5 is still useful for agents
What do you mean by agents?
>4-5 on Q4S
That's around reading speed which is pretty damn impressive for a single consumer card. I too am looking at 3090s (and maybe p40s) but won't pull the trigger until I figure out automation with ComfyUI, home assistant integration and a better tts.
>I would archive papers if a model could summarize and explain them for me.
been meaning to do something like this, but automating scihub retrieval has proven to be a bit of a pain in the ass, and I don't have the space to download everything
> Chat models are good enough for regex and nginx configs but I could see a usecase for making better ansible playbooks using a tuned model.
I'd agree, although even gpt-4 fricks up regexes sometimes. LLMs are great for config files, so long as its well known software
>it's too much work for my taste.
you're not wrong, complete waste of time, but it's addictive
>What do you mean by agents?
anything that interacts with the outside world, like running shell commands, calling APIs, searching for things on the internet, etc.
generally very well defined tasks
I've thought about p40s, but the old pascal architecture isn't great, plus you need a very annoying blower fan mod. not an issue if your server is in a closet somewhere, but in my case, it's next to my desk.
>it's addictive
I didn't get it until I started using sillytavern, the whole workflow is addictive. Especially when you start using SD to generate character images.
>anything that interacts with the outside world, like running shell commands, calling APIs, searching for things on the internet, etc
Ah that reminds me of the autogpt git repo, I remember that was THE cool ai thing a year ago. I'd like an agent that interacts with my searxng instance, maybe linking it with jellyfin to generate local links of youtube videos I've archived. Not exactly outside world but I'm trying to locally host more.
>p40s
It's only appealing because spending all that money on a 3090 or 4090 hurts too much. I have a laundry list of things I would use gpu compute for but I can't justify the cost unless it pays itself off. And I'm too moronic with CAD, transcoding and ai models to make money off it.
A 3090 is only $600 now.
>only
P40s are like $200 or something and they're very slow
samegay stop spamming this garbage
Nothing because your reasoning is backwards. Have a hobby or interest and invest in it. Don't invest in something and look for ways to use it after.
I bought my 1080ti 6 years ago because I wanted to get into VR and needed a better gpu. I'm still using it despite raytrace and dlss memes because I don't need more for gaming. I'm considering an upgrade because I want more compute. Even though I have an actual reason for one it's not worth the cost for something I mess around with a few hours a week.
Be smarter with your money, or don't I don't give a shit.
meant for
make millions off mining alt coins before people realize they aren't worth it
Crack passwords.
When people say this, are they literally talking about, IDK, stealing from crypto wallets or something?
they mean utilizing a GPU to brute force a password hash, i.e. one acquired from someone's wifi network
not worth the time unless you have a (very expensive) dedicated rig with a bunch of GPUs
but if you can buy such a rig then why do even need to crack someone's wifi password, i mean
like what's the point
it's a dead art
I always assumed it was to steal money, somehow.
>what's the point
getting cool ass tripcodes
I used to do that back in 2015 but then I gave up because being a tripgay sucks
I honestly do not know how people live without a graphics card. My life was so empty without one.
What are you using it for, OTHER than video games, unless the video games are educational or something and would justify themselves beyond entertainment.
generating AI child porn
This is the real reason so many mentally unhinged people worship AAI.
nothing
You can get a usable gayming setup for like $110-130ish with a skylake prebuilt + 1070/470/480 + 6to8pin-adaptor though
Just be careful with how much power you let the gpu have, start with a super low limit and just slowly increase until you're in the ballpark of what the cards performance should be with base clocks
If you smell burning plastic you pushed too far
Same i sold my last two gaming pcs and built a 64gb 136k monster
The games are aweful the only new ones i like are cs2 and vr shit which is all dead mp
You can generate furry girls
How much is that worth, though?
if you're a richgay you'll coom so hard you'll end up dangerously dehydrated. not really worth it otherwise tho unless you play gaymes or need GPGPU.
you decide!
god i want a 4090
I have a 2060, rejoice!
My concern with Nvidia's newest cards is the power consumption. I am not environmentalist homosexual but even with a 3080 when I play at 4K/120fps it sounds like a jet engine. I was playing Lords of the Fallen and it tripped the circuit breaker. These cards feel like overkill. It's a lot of heat and a lot of electricity, raw power output, going into a mere videogame. Seems disproportionate. Something to think about. If there's a way to deliver the same results using less power I think that's gonna be the next big innovation, especially concerning the costs of training AI
Powercolor told me that my 6900XTXH Ultimate required a 900W PSU. I laughed @ them and used a 750W PSU and slapped a TDC limit and now the GPU boosts higher than advertised OC BIOS while using no more than silent BIOS wattage limit.
*plaps*
*impregnates*
I dunno like make a billion dollars?
generate disinformation
It can be about the journey. When I embark on a hardware mission, although its usually frustrating, I learn stuff and there is a sense of fulfillment when I get things back to how they were after I mess everything up.
the most interesting thing I've seen so far is the text to speech and transcription AI stuff.
Pixar tier 3D animations.
Will all this UE5 raytrace bullshit, if you have good enough art ,you can pretty much do a pixar movie at a decent speed.
*sigh* and what kind of 3D animations are you making with that...?
It don't need to be porn if you don't want it to.
You can even make something cool or wholesome.
>Radeon
Nothing else LMAO
Did AMD frick with the last rx 580 driver for windows 11?
When i try said driver my card would not clock as well as older drivers.
Just wondering if that's a thing...
Can someone recommend me a good build using a 7900XTX sapphire nitro +?
I mostly play games like squad, hell let loose, and rust. I do some photo/video editing and basic stuff like excel. I would like to experiment with 3d stuff like blender but that's on a backburner.
May as well be getting a 7950X3D at that point, everything else in your build is discretionary. Water cooling the cpu is a must at this level unless you can 3D print custom airflow channels for a case (unlikely you’d be able even with a printer unless you have just the right case for it)
What is the difference between a 7950x and a 7950x3D?
Which AIO would you recommend? Arctic liquid freezer ii 360?
I do have a 3d printer but my skill levels with it are terrible.
How do you like that motherboard?
7950X3D is better, has a bigger cache and less power draw with better performance and thermals. Arctic liquid coolers have never disappointed me, but I also haven’t used a cpu of that level with one.
That's not my actual desktop. That's a good motherboard, though.
I'll post my generic top end (for video games) setup
https://pcpartpicker.com/list/32vQRK
Make deepfakes.
You can make cool videos.
you can make stuff in cad
Am i losing much with 3200mhz cl 16-18-17-17 ram on 5800x3d? i dont want to replace my 4x8 gig sticks for 2x16 3600mhz if it doesnt matter a lot
3200/16 should be all you need, IIRC.
( ͡° ل͜ ͡°)
Random question not really on topic, but I have a 4090 with 3 display ports and 1 hdmi port. My main monitor runs at 2560x1600, and the other two run at 1920x1080. All 3 use the display ports, but I also have a projector that I switch over to for movies and games, but I have to unplug my main and another monitor to plug in the projector because the video card can't support another 1920x1080 resolution, which causes some flickering frickups. Not sure what to look for, but I need something like a switch that will turn off two of the monitors and enable the projector without having to unplug two of the display ports to plug in the hdmi... is there such a thing that anyone knows of?
Researching neural networks, Using that research to control drones, make your workflow easier.
Finetune existing neural networks. Use your workstation as central NAS, have it control your house/ backyard with sensors everywhere.
No regrets about my past spending for computing components and will continue to do so.
You will find some algorithm that you find interesting eventually ; )
Honestly? I upgraded just for my fricking monitor. It can run at 240hz but my old PC couldn't handle that, so I built a new system and now it runs at it's true potential. So getting the most out of your hardware is something to consider
do you play anything other than CS2 with such a refresh rate?
This doesn't make sense though. You basically bought the monitor knowing you'd then need to upgrade your computer. What a clever way to consoom.
>I can't justify building a desktop just for video games.
and yet you really want to. that's why you've made this cope thread.
just do it. treat yourself.
You can run demoscenes in real time on your graphical adapter. They produce amazing visual effects that stress your graphics card and the music also amplifies the experience. They can range from a few kilobytes to a dozen megabytes so they are very portable to run.
https://www.pouet.net/
Some of my personal favorites:
(opengl)
(direct3d9)
(this one is only 64kb)
Real answer?
Streaming
Video editing
Compiling (programming)
I run a plex server on my desktop to easily stream my movies/tv/music collection to the living room.
Every time I uninstall Nvidia's audio driver, it returns, why do companies do this?
That's Windows automatically installing drivers even when you tell it no. You have to unplug the Internet and install your integrated sound/sound card driver first.
translate JAV
You can compile Gentoo way faster than you would on a craptop.
video game development
video editing
3D modeling
technically you don't need a powerful computer but it really helps
i shorted my 4090 last week because i jizzed into my case by accident, apparently cum is super conductive or somthing.
anyway, my jizz leaked into the card and the power supply and now both are fried.