what idiot would buy a 4090?
 It's All Fucked Shirt $22.14 |
 Nothing Ever Happens Shirt $21.68 |
It's All Fucked Shirt $22.14 |
what idiot would buy a 4090?
what idiot would buy a 4090?
It's All Fucked Shirt $22.14 |
Nothing Ever Happens Shirt $21.68 |
It's All Fucked Shirt $22.14 |
a rich one
Someone who needs the VRAM.
What is that supposed to show? Rasterization performance or what?
Truth to be told, the 24 GB fill up quickly, Stable diffusion is perfect for the RTX 4090, LLMs are not unless you don't go beyond 13B, in such case, an A6000 or similar models with 48GB of VRAM or more are recommended, BUT, a macbook pro with 96/128 GB is a good choice too. You get around 70% the performance with huge VRAM to fill demanding models with way less TDP, it is a way cheaper option than getting A6000's or H100's plus huge TDP consumption. There's no correct answers here.
Also, Intel is changing really fast, they have been releasing a lot of updates on their software stack, not only gpu drivers.
I really like the idea of buying a macbook with 48gb with a m3 ultra / max for AI, however it costs the same as buying 2 4090 (ideally 5090's), and a single mac ship is only able to barely match the performance of one 4090 in theoretical numbers.
And last time I checked, apple AI still struggles with AI software support (only gguf works, I could buy a 24gb tesla p40 for $200 and it runs gguf, it would run like shit, but it would still run at a decent enough pace to fap to).
>I really like the idea of buying a macbook with 48gb with a m3 ultra / max for AI, however it costs the same as buying 2 4090 (ideally 5090's),
>let me just get the aluminum paperweight for a thread demanding task when I could get the graphics cards and have infinite more threads available to not melt a hole in my system
34b works, just quant it. I make my own MoE blends with 7b finetunes, and zipper stack layers for frankenmerges. Mixtral 8x7b and its family's q5 quants run about as fast as chatgpt4 on a slowish day, if openai is down I just load one of those up. Llama3 drops this week so there's a whole new generation for 4090's to play with, gonna try finetuning one of those. Today I'm running some benchmarks on 70b q2/q3, q4 works too slow at 2-3t/s - but I haven't bothered to optimize beyond layer offloading. It's a solid card and ready for fp4 libraries whenever those become common. Plus it plays games, macbook is the lazy option if you have a couple grand to burn and don't want to set up a decent computer.
>Stable diffusion is perfect for the RTX 4090
I don't find myself needing more than 12 GB of VRAM when using Forge.
why do these low effort monkey bait threads exist?
You know GPUs aren't used just for gaming anymore, right?
Gayming GPU SLUs are used for gayming and media consumption. Nvidia gimps them insufficient VRAM capacity for any useful work.
If the 4090 had a price cut I would have bought it a long time ago.
>what is social status
people buy 4090s for clout and to show off
to who? i can literally say that out of all close people to me none care or even know the difference. somewhat distant friends i see 1 time per year has a few people who may notice it if we have a conversation about gpus.... and then i guess i could make like a reddit thread about it or something as this is the only place where someone would even acknowledge it.
What social status are you talking about? if we assume that women really care what car you have or what phone you have, i cannot imagine a woman knowing the difference between 1080 and 4090. Who are you going to impress? your teammates in online games? the game scores do not change much even if you have steady 200 fps on a 4k 200 hz display. only in sc2 i could imagine you making 1-2 more kills, but if i activate my hacks i will make 30 more kills than what i usually do.
everyone wants a 4090 but not everyone has the money.
Hourly poorshit cope thread.
Where's my RX 580, what have you done to it !!
it's time for us Polarichads to think about upgrading in the upcoming couple years
>77% better performance (at best)
>for 3x the price
Yeah truly winning moron
>4090: measured in it/s on sdxl
>4070: measured in s/it on sdxl because memory is smol
that's why people buy 4090, thx for coming to my ted talk
>>77% better performance (at best)
Actually, that's an average. Also, the average greatly increases if you enable RT, where the 4090 is well over twice as fast as the gimped pile of shit called the 4070 Super. It's also worth keeping in mind that the 4090 is frequently CPU bottlenecked even at 4K. It is literally too fast for even an overclocked 14900K or 7800X3D to keep up with in many titles.
u can stop advertizing noone will buy bro
Nobody is wasting their time advertising to you broke-ass b***hes and your Sandy Bridge Thinkpads.
ok shitskin
>nooo you have to buy based on price to performance
Frick you I want fast you dumb fricking moron.
thanks for fueling the economy i guess moron
Where's my 3080 Ti?
People who can't afford to wait for the 5090
One with a 4K monitor
Weird comparison though, the 4070 is a dogshit card that's easily mogged by AMD price equivalents. Unless you got the 16GB Ti version I am sorry to tell you that you got scammed.
Is this rasterization-only or RT?
it looks like the plain ultra quality setting without RT, DLSS or FSR. If it had RT the 4090 would be around 40fps and the 7900XTX would be around 6fps
jesus 4k is a meme
I use my 4070 super at 1920x1440 or 1600x1200 depending on the game and I get alot more then 60fps at ultra settings
So paying for a better image quality is a meme? Why didn't you stick to running games with low settings at 720p then?
2k lines is absolutely a meme for most people.
I do not understand the push to 4k at all, only makes sense in very select cirumstances.
jump from 720 lines to 1080 or 1440 is massive, jump to 4k isnt that big but costs massive amounts of processing power
same except a few days ago.
the card seems to be pretty good value for your money, I know AI monkeys scream about the memory, but 12GB still seems plenty
>jump to 4k isnt that big
Yes it is. It's 4x 1080p screens.
>It's 4x 1080p
1080p x 4 = 4320p
Yeah, it seemed the best bang for the buck. It's goning to be leaps better than this 1660 SUPER
i upgraded from 4GB rx570.
old PC is gonna be my new media rig now, because my current media rig is so old it struggles with high bitrate video.
4070 super was 600 euro, 4070ti super would have been over 800 and i am not paying that much more for a measly jump in performance and a few more gigs of VRAM
doesnt matter to me either way because im not buying a new monitor.
1920x1440 it is for the coming decades
You appear to be suffering with a severe case of "What I have is all you need, anything less is shit and anything more is pointless" syndrome, probably caused by being a third worlder
what i have is more then anyone needs.
and the 4090 isnt pointless per se, its just very bad value for your money.
The high end GPU market is trash value
As for monitors, yes I know for a fact anything less then 1200p is trash anything more then 1440p is pointless.
seen it for myself.
Same with 4k tvs, nonsense, 1080p high bitrate looks perfect for movies and shows
4090 has massive value for AI, there is a reason why chinese people are buying mrsp 4090's from america and reselling it in china.
the problem of 4090's is that they eat a lot of power compared to a legit AI card, but the legit AI cards don't have the shader cores needed to run 4k video games.
4090's at their current price are quite poorly valued for gaming, however they are still competitive in AI, and if you bought a 4090 at msrp, it would be reasonably high spec for gaming. Gamers are much more sensitive to performance per dollar, and that's the fault of people watching reviews to decide which part to buy. Spending $1600 on a gpu is not as much as what people spend on a bicycle, car, or TV.
The value of a 4090 is genuinely real and due to the fact that the size of a transistor is getting very close to the size of a single atom of silicon (0.2 nano meters), it's unlikely that waiting will give you a better gpu than the 4090 (I still have hope the 5090 might be better than the 4090, I think they canceled 4090 TI because they predict the 5090 will sell more, but if you ask me it's very likely going to be 30% more expensive for 30% more performance with the same vram :skulls_emoji:).
the 4090 is overkill for gaming however, I use a 12100f and 1660 ti and it's hard to find a game that won't run my my specs above 120fps on 1080p lowest settings, and the game being genuinely fun and interesting.
I don't see any value in AI.
>more then anyone needs
false, commie
>very bad value
to you
to someone else, it might be extremely good value
to yet another one, a mediocre value
>more of the same ME ME ME ME I I I, therefore FACT
you're not the arbiter of truth, neither the star of the show, my guy
why do you commies always insist on your omnipotence when you're clearly flawed?
>blah blah blah
we used to kill israelites for this stuff
>implying
pic related, false flagging commie
>One with a 4K monitor
I have a 4K monitor but I play at 1080, because the difference is barely visible in motion. 4K is for work and reading.
The 4090 is a really great GPU for Blender.
I like my 4090
Just bought a 4070 SUPER yesterday.
I _could_ have bought a 4090, but I didn't.
has anyone done something similar like pic related for the new generation? the reddit user that did this one hasn't made anything newer
tomshardware.com/reviews/gpu-hierarchy,4388.html
the image was incorrect so I fixed it.
I kinda regret going with a 6800xt instead of a 7800xt. idk
awesome thanks
4090 - $2000
4080 - $1000
4070 Ti Super - $800
I think I'd be happy with the 4070 ti
>1080p medium
That's a dogshit way to compare across such a vast generation count.
You can visibly see how the CPU bottleneck is crushing the results of the current and 1-2 gen old GPUs.
It's like going back to te year 2000, turning off T&L running at 640x480 and saying a Rage 128 is only half the speed of a Geforce 256.
Dogshit.
AI
however the 5090 is coming out and they are limiting 4090 production so the 4090 will not get cheaper any time soon.
5090 will not give free improvements, but if you were planning on buying a 4090, you are essentially buying the most expensive consumer AI hardware on the market, why not wait a year and get something that will be the best hardware for longer than a year (I bet a 5090 TI or Titan will release, but if it doesn't have more than 24gb of vram, it's not very useful for AI).
I you are shameless with sharing your AI fap sessions with google, you could use google colab and buy the subscription so that you can run a bigger AI on a proper AI card.
Colab is a b***h to work with however.
isnt the 5090 bus confirmed to be below 48 GB of ram? what is the point in buying half again a 4090? either way the value of my 4090 skyrocketed after I bought it so It's hard to be mad about my purchase.
Isn't it 512? the 4090 has 384.
I think the 4090 is either going to be 32gb (or that's the 5090 TI) or they just don't utilize the full bus to give 24gb (they did this for the 8gb 4060 TI, and the 4080 using the 4090 chip with cut off cores, and less chips means less bandwidth) or only give the server tier gpu's use all the chips (since they make a lot more money selling the A6000 ADA which was a 4090 chip but with vram chips with 2x the size, 48gb probably only costs $800, while 24gb vram costs like $200-400, to the consumer, but the a6000 ADA costs like $6000, but realistically only 3D studios buy the a6000, the AI startups are buying the h200 which are like $1 mill for a DGX machine).
you seem to have made some typos there. anyway theres no point in buying a high end GPU for thousands of dollars if it A: doesn't have vlink or B: isn't high enough in vram to run a decent Langauge model.
32 GB of vram is not enough. I need 40-48 minimum. 80GB would be ideal. 48 is the expected gain from my 4090's vram in the first place. if 5090 is launching with 32 GB it's a non starter unless you skipped 4000s
Honestly we just need to wait for benchmarks, it might be possible that 5th gen tensor cores might be interesting for inference because they have more fp8 or fp4 performance than last gen because they divide the fp32 data (but I always hear about tensor cores being bad for inference, I don't even know if my 1660 TI even uses tensor cores, how do I check?).
Also I believe nvlink is only for training, I don't think there is any benefit for inference for nvlink.
oh 1660 ti doesn't have tensor cores, my bad.
If your use case benefits from high vram and a) isn’t bottlenecked by memory bandwidth, or b) you’re afraid of used 3090s
What's the use case of a 4090?
huge anime breasts
All of them
>chart showing it's the best gpu of all those listed
>duhhhh why would anyone buy this?!?!!?
Something needs to be done about this shit. There is no reason moronic question bait threads should be the only type of thread that gets posted.
you are the moron. if you compare the prices of the 2 models the 33% performance increase is a 120% price increase. you are paying quite a premium. this is why the question is given: who has all that spare cash to give for the same fps in games.
>inb4 "i use it for work".
ok, so i hope you make the difference then.
So this is the power of a chart reader.
>the chart visibly shows it way out in front of everything, massively superior than all the rest
>DUHHHHH IT ONLY 33% DAT NOT MUCH IT WIPOFF
I bought it, but I have plenty of money to burn for no reason.
They normally make over priced Iphones or cars for you to play with. What else are we going to do for you with all the money you have filtered to yourself.
>12gb VRAM
Might as well be 512MB, you're not meeting system requirements in any modern software.
i do not understand Vram schizos.
4GB rx570 was enough to pretty much run anything okayish
name 1 piece of software except AI nonsense that wont run on 12GB of vram
a lot of games running 4k use more than that. Far Cry 5 and The Division from memory
im not running 4k so why should I care?
>why does thing do thing?
>because this
>but I don't use that so
don't give a frick you stupid c**t
>is there software that WON'T RUN without X
>well this software will totally run without X but you may have to run it with X if you go out of your way to do Y
The worst part is that even after dumbing things down to your level, there's no guarantee that your illiterate ass will understand.
>b-b-but I can go ultra potato mode on my 20 year old shitbox PC and screen and it'll still run, slowly, and look bad but it's still running, right?
lol the state of cope from thirdies. It's time to let go from GPUs with 9pin dsub connectors on.
I wish 4070 super had VGA out and a decent RAMDAC on board
analog output is the only way to go for high end video cards
Just because a software wont run on your nigtastic 4k ultrashite monitor doesnt mean it wont run you spastic moron
Name 1 software that just wont run with 12GB vram in 1080p
there is none
I don't care about 4k but personally I would go for a OLED 1440p ultrawide monitor for movie watching and because 1080p does not exist in OLED or mini-led.
>inb4 OLED burn in
OLED displays are just nice in a dark room.
I would wait for next gen OLED's however, current ones are fine but I have a feeling that either next gen will be either better, or the older models will get cheaper, or mini-led's finally get competitive.
>ultrawide
thats nice but i have disregarded your opinion on monitors since its clearly worthless
OLED is nice and all but 4k is a meme and most still have bad motion.
I feel like my old trusty crt from 1996 is much better suited for my 4070 super
senpai, your crt is about as expensive as a oled. congrats if you found it for free, but IPS is better in every way except for watching cartoons or games. IPS still has inky blacks in a pitch black room as long as the darkness isn't touches the edges of your monitor, because you only notice the contrast when you compare it to real darkness.
CRT's suck at lit rooms (side by side with a IPS monitor like koorui 24e4 and the dozens of monitors that share the exact same panel), and I have zoomer hearing and the CRT coil whine would make me go insane (I can barely deal with the hum coming from my subwoofer).
crts are still cheap as hell.
And I have about two dozen more or less high end crt monitors lying around my flat so im not too worried about not being able to use them anymore any time soon
even miniLED is not enough for me, and again motion will be dogshit
yeah crts kinda suck in lit rooms so I dont light my room problem solved
I agree OLED is pretty cool but its not worth it to me, especially because I dont want to run 4k and the extreme performance penalties that come with it
do you have any 2000x15000 resolution ones you'd be willing to sell?
I dont sell.
I would give you one for free possibly if you were a close friend, but not one of my favorites.
just look at junkyards and ewaste places
I already did like last time you told me to.
All but one said no, and the last one that didn't literally ghosted me. theres no way to get these in rose city without breaking and entering.
yeah dont ask.
just go and take
im not getting arrested trying to lug a CRT out of portland
>24GB VRAM
wasted vram
twice the memories
Every consumer GPU priced at over $500 is inexcusably overpriced.
>B-b-but I can render the godrays bouncing off my character's nipple hairs when gayming!!!
And? That's worth $2000 to you? Really? Tell you what, if you're so "rich" that you don't know the vaule of a dollar, I have a nice toothpick to sell you for half a mil.
I've got a Nvidia Titan X and i can easily play all games on ultra settings i want on 4K or 2K. Why would i need an overpriced RTX?
No idea. I got a 4080 super for half the price.
theres a less than 1% difference between a 4070 and 4090, morons cant into math i guess
its like how people thought a 1/3rd pounder had less meat than the quarter pounder
Unironically, am I behind the times with a 3070? Anyone upgrade and instantly see that it was worth it?
i have the same fps with 3080 during gaming in my 10 years old games as i did on 1650 super. for me: its all a meme. but i will be called a poorgay if i proposed that using 3070 is enough so there is that. for the games i play even 1080 would be enough to reach my display hz.
>Anyone upgrade and instantly see that it was worth it?
Yes but I'm specifically into VR. Going from a 2080 to a 4090 provided huge immediately noticeable differences in VR, playing a game while streaming it using OBS, and stable diffusion. That doesn't mean it's worth it to you though
1440p I wouldn't bother but if you are pushing 4k on some titles, a 4080 doesn't hold up as well.
I have a 4090. I use it for around four hours a week, because my desktop is strictly for gaming and only gets turned on Friday and/or Saturday night.
It's worth it to me because I have a 49" curved monitor, and want the best performance possible at the best settings possible. It's not about perf/$ or even $/hr of use. Sometimes people simply want the best their money can reasonably buy, and the 4090 is it.
pfft 4060 is only $300 and 4090 is only 2x as fast
gee 2grand for 2x or $300 for 1x idk frick
here's the kicker: 4060 115watts
4090: 450 watts
you got 5x the cores on 4090 and it cant get over 2x performance from the 4060 lmfao
bottomline you're limited by the engine and core clock speed which is essentially the same
>researched GPUs (every gen) since 9700 pro
Isn't a 3090 near double a 4060?
ahh, no.
I'd buy one since i recently upgraded to a neo g9 that's like 2 4k panels glued together but the 5090 is apparently coming soon so i might aswell hold onto my 4070ti a bit more
5000 series is supposed to be coming out towards the end of the year so yeah, you definitely shouldn't upgrade from one Ada card to another at this point.
I'm waiting for next gen to buy 90 class cards for both vendors
NVIDIA shills trying their best to sell out stock before they start shill campaign on hyping up new series of cards by NVIDIA
sorry ,
i am sticking with Pascal
an idiot that has more money than you
technically nividia's stock will go higher if they don't make the 5090 better because it forces AI companies to buy their $10,000+ GPU's.
however if they make the 5090 amazing they would get a lot more domination in the ((TINY)) gamer market at least for AI (but I think game studios will rather buy the 5090 to host servers that gamers need to subscribe to, and technically the $10,000 AI gpu's hold more value as long as Intel, AMD, Nvidia, or any other company doesn't start dumping cheap hardware to reduce the prices of AI cards.
Nvidia has a monopoly on both AI hardware and AI software at the moment.