Are AMD GPUs good now?
They're the only ones anyone can afford, so sure. What's the competition, 10k euro 5090s and 2k 5080s?
I hope it will be much better at rendering in Blender than RDNA 4.
>AMD's coming back to the high-endGood, it's about time. I thought it was a mistake for them to skip it this time. They should go in the next generation swinging hard.
>>110189301Amd should do something like artcraft to adobe, or apple silicone to jewtel, and basically come up with a new way that fuck nvidia and intel till they cum without touching their dicks, just pure prostate fucking and sell those cards for $600 each or something
I'd want them to release another strong iGPU, one that can be acquired outside prebuilts. With shared system memory it could be a banger.
>>110189301Worthless without CUDA
>>110189517Does ROCm not work in 2026?
>>110189530The issue with AMD is how fast they abandon their GPUs hardware/software support instead of building on it for a long period.It makes people less likely to buy their stuff out of worry that it may get dropped all of a sudden and tanking its value.It would be a different thing if their went the same approach they do with their CPUs.
It's not worth getting an AMD GPU cause their driver team is small, filled with boomers, assholes with massive egos who don't share any knowledge cause of fear or layoffs and they literally don't have enough people to handle the changes required to ensure your games work. Nvidia also has superior gpu crash debugging tools so every single artist and developer at game companies is using nvidia gpus so they can debug their games which means all optimizations and fixes are done on an nvidia gpu during production. AMD hardware is good and game companies utilize it to the fullest extent but there just isn't enough love for them on PC.
>>110189640Fuck I meant console developers utilize it to the fullest extent, not game companies.
>>110189301If you don't mind the high power draw they are good for gaming performance per dollar. And win over Nvidia in that.If you want to do AI shit, it's a decent budget option.>>110189530>>110189585I do AI shit on AMD's stuff. It's is janky, requires plenty of flags, and can be a headache. But their drivers really do just keep getting better over time, and honestly at this point it's completely workable if you set the right flags on a llama server or something.And considering the cost of Ngreedia shit it's worth it to me.
>>110189301>prolly 10k buckerinosunless it's <3k i'm keking.
Board manufacturers and amd themselves have experimented with attaching more storage to gpus like m.2. Using a p2p storage. We need heterogenous storage solution to survive rampocalypse
>>110189759Latency?
>>110189309Can we really expect them to not match nvidia's prices?
>10900they need a new name scheme before they reach 100k
>>110189301Does it exist yet? If not, no.
>>110189345Wasn't the cope that RX 9000 series were a stopgap before they would release UDNA shortly after? If AMD released a proper counterpart at the 80 & 90 range they could have smoked Njewdia and actually taken up some much needed market share.
>>110189301
>>110189779If they do, they won't sell any cards.
>>110189846Gods I hated kids spamming this retarded inane shit for years, same for all you base and desu. Definition of a forced meme.
>>110189906yeah seemed fitting though
>>110189585>>110189640AI tools already help in driver development and this releases late 2027.
>>110189906I didn't. It was a better time.
>>110189301>10900XTWhat the fuck are they doing?
>>110189301$10,900 msrp
>>110189906It doesn't make a difference, guy. The balls are inert.
>>110189980You were one of the spammers weren't you?
>>110189585Depends on what you mean by abandon. Just because they don't add the latest and greatest to an older card right out of the gate doesn't mean they completely abandon it. FSR 4.1 on RX 6000 is an example. Also, it doesn't mean their cards magically stop working. I was running LLMs and image gens on a Vega 56 until earlier this year when I finally upgraded to a 7900 xt.
>>110189773Yes
They are available. Nvidia is slowly exiting the consumer gaymer gpu market. 6000 series will probably be the last of GeForce before that brand is killed.
Are AMD GPUs good at path tracing yet?
>>110189301>AMD aims to take Nvidias place as the top consumer graphics cardAnd other hilarious jokes you can tell yourself
>>110189318>make RDNA4 works faster with Blenders, make no mistake, gosimple as
>>110190135>full total meltdownwhy are nvidiots like this
>>110190188AMD/Ati have said this for over two decades. They never actually pulled it off.I've owned 3 Ati/AMD graphics cards over the last 20 years and I was always glad to be back on team green. Its not even the card itself that is the issue. Its the drivers in most cases.
>>110190204>over the last 20 yearsthat's turbo-faggotry reaching
>>110190204and the last decent GPU that nvidia made is 1080ti
>>110190233What about the 3090?
The only reason for gamers to upgrade from a 9070xt from this point on is if their card melts down
>>110190233>decent>1080tiare you high?1080ti was and still is the best GPU ever made>>110190219zoomie spotted
>>110189309>>110189779>>110189862What in the fuck are you retards even on about? Their MSRP for equivalent cards are the same.
>>110190255Terrible value that was never truly available at MSRP. There was a brief moment where it was available for a decent sum on the used market back in 2025 but due to the AI scam its back to inflated scalper prices. RTX 40 and RTX 50 series are ticking time bombs with that faulty connector. Nvidia is simply not an option if you have any semblance of dignity
>>110190233>>110190323It's not the card. You're confusing hardware capability with developers being more capable of optimizing, or video game development processes was at a sweet spot of maturity that standard practice at the time was to optimize. More than anything, drivers and developers optimizing for your cards specifically matters most. That isn't something most people think about because NVIDIA has owned like 80% of consumer GPU market share for decades.
>>110190391>literally the only player in town>something something not an optionYou're a joke.
>>110190391>Nvidia is simply not an option if you have any semblance of dignityThe most powerful AMD GPU on the market loses to a 4070 in Blender.
>>110190419yeah but i'm not putting my gpu in a blender so what does it matter
>>110189301Not a single one of these will ever hit store shelves. Haven't you faggots heard AMD is now desperately trying to do the same circular financing bullshit Nvidia does but with Fermi of all companies? What did you really think they were your friends? L M A O
>>110189640Playing games is for children, you should just grow up.
>>110189530AMD is retardedROCm HIP works, and it works for some timeYou need to be CEO CTO type of retard to understand the stunt they doing with itBy default wrapper/emulator of some tech cannot run as fast as native shit. And fucking geniueses at AMD decided, 100% because of optics and nothing else, that they will semi officially be supporting itAMD decided that YOU must buy NVIDIA for 90% of anything GPU work related because some marketing cuck said to even bigger CWTFO cuck that AMD will look bad in HIP vs CUDA comparisonsYou just cannot have it cause its slowIts just fucking works, and you cant have it work in any shape, cause its slower than CUDAAbsolute degens
>>110189640What you talking bout willis? Game devs target AMD hardware with Xbox, PS and steam machines being just reskinned AMD boxes.
If all you care about is video games, then yeah. I'd even say that AMD is better in that regard especially with everything using d3d12/vulkan now.
>>110190204I've never really had any issues with AMD cards until I got this 7900xtx and had to roll back drivers a couple of times. I get the impression they forgot about the 7900 and only give a shit about the 9000 series now, even though the 7900 is superior.
>>110190448yes yes I only use my computer as a Pornograph machine because games are for children and I am a grownup who likes grownup stuff like boobs and spreadsheets
>>110189318Blender is for faggots and wanna be "artists " The real question is can it run larger local models and 4k gaming.
>>110190014Mind sharing your setup? I'm still on Vega 56 and I've tried stuff like LM Studio with the Vulkan backend and it's a bit too slow to be useful IMO, even with 7b 4 bit quantized models.
>>110190869Use ROCm like a straight white man and stop being a homosexual
CUDA is better than ROCm because it flows off the tongue much better.
>>1101904199070XT is trailing the 5080, and obliterates 5070Ti in gaming. you're clueless
>>110190880No support for Vega 56, otherwise I obviously would.
>>110190837>I get the impression they forgot about the 7900 and only give a shit about the 9000 series nowthey adapted FSR4 to work with RDNA2 what are you on about?
Only cucks buy AMD. That shit is so behind cuda it's worthless for anything with AI in it (basically everything nowadays) they should stick with CPUs and give back the silicon to nvidia
>>110190918>for anything with AIsimply ask the AI in duckduckgo and get over it. GPUs are for gaming. if you say otherwise you're a plantation slave and stop talking to me
>>110190176and yet, where is it?AMD is just a controlled opposition at this point
>>110190918Since ROCm 10 dropped it's been a much different story. For a cost concious person AMD is a genuine alternative to anything Nvidia has. At best you have to do a bit of tinkering that anybody with a working brain can do.
>>110190973And hell I'd argue the tinkering and flag setting I've had to do over the past year has helped me learn far more about LLM architecture because of it.
>>110190908I've had two of their updates fuck up my 7900, while my wife sitting next to me with her 9070xt has had absolutely no issues whatsoever
>>110191033>my wife GPU cucked methat's a new one
>>110191033What about your wifes boyfrind bvll with his 5090 32 gigs of girthy vram?
>>110189517Good, maybe that will keep the prices sane.
>>110189906I hate dragonballz, but that post is alright.
>>110190869Sorry, I don't remember my exact setup. It was fairly basic. Ryzen 1700 with 32GB, Vega 56 8GB, Linux Mint, Koboldcpp for LLM and vladmandic sdnext for image gen. I know I was testing Zluda but can't remember if I got it to work or not. It wasn't fast obviously but usable. The HBM seemed to make the difference. Really wished I would have gone with a 16GB Vega.
>>110191067>I hate dragonballzfuck you
>>110190000>90000Wait a second, I think we're jumping the gun a bit. Have you checked your other pants pocket?
>>110189906I'd rather have a million 20 year old memes than a single frog or jak posted.
>>110190596 Console development and PC is not the same. If you straight port a console game to pc it'll run at like 3fps. That's what happened to RDR2 (or was it GTA V? I forget, maybe someone else from r* or amd can chime in here lol, I know you guys read this shit) and the rockstar devs were like yeah that's cause pc sucks lol and then AMD had to help them get it to run faster.
>>110189517CUDA is becoming irrelevant in the general compute space. Nvidia is even moving away from as being a centerpiece.
>>110191308>CUDA is becoming irrelevant in the general compute space. Nvidia is even moving away from as being a centerpieceSaid no one ever.
>>110191240DragonCuckZ was awesome when you were 5. It's not an anime any grown ass man should be watching.
>>110191291>AMD had to help them get it to run fasterBut I thought AMD was filled with egotistic boomers that don't share knowledge and don't have enough devs to make things work?
>>110189301Data centers are going to bot and buy all their stock.
>>110189301and where will radeon get the gddr7 vram from
>>110189585The way they decide which GPUs to stop supporting is very strange. I wanted to run a moss model on my 6800 but apparenty that needs hipBLASLt support. Vega has hipBLASt support and the RX 7000 series also has support but the 6800 does not. I got sandwiched by support but but missed itEven weirder that apparently on windows hipBLASLt is supported on my GPU with pytorch 10 but ROCM is not. I dont know what the fuck the driver team is doing
>>110191463found the onepiece faggot
>>110191463Dragon Ball is so popular in South America that cartel activity drops when a new episode airs.
>speculative msrp of $999Dayum
>>110191308Imagine being this retarded.
>>110191793But nshitia does the same except worse because the jeetware includes spyware...
>>110189517>Worth lessA good thing in this economy.
>>110189301They said the 9070XT was going to be 32gb but then chickened out and released it as the R9700 with a huge AI markup so I'm guessing it will be the same and we will be lucky to get 20~24gb instead.
>>110189301Will hiring people who know how to write Windows drivers be part of this effort to take over the consumer graphics card market?
>>110189301Have never bought Ngreedia because whenever I look at buying a new GPU, I discover AMD gives me 95-100% of the performance of the Ngreedia offering at 50-60% of the price. This has been true for the last 15 years. Also, AMD keep improving the performance with driver releases, while Ngreedia cripples their GPUs to make you buy the new model. RX 580 was slower than GTX 1060 at launch, now it's faster. RX 9070XT was beaten by 5070Ti at launch, now it's matching the 5080 in some games.>>110192877I have not had driver issues since 2012. Funnily enough, Nvidia are the ones that have had driver issues fairly recently.
>>110191308>>110191451>Nvidiots and baby ducks getting this mad that times are changing
>>110191793>using Windows in 20-fucking-26
>>110191793>have to install 5gb software suite just to install the gpu driversClearly an NVIDIA astroturfing jeet
I love my bigass 9070xt. I don't think I'll be upgrading until 32gb gddr7 is affordable, which will be never
>>110189301I thought that was coming in 2028? It would be a nice change of pace to have a high-end xx90 Radeon after so long. 7900xtx didn't count as it barely matched the rtx4080
>>110190864nobody uses graphics cards for games anymore
>>110189517why?
>>110191793amdgpu driver + firmware on my gentoo system is 112MiB total, and that includes firmware for my previous 3 cards which i haven't bothered filtering out since they're tiny
>>110189954The ultimate irony, Nvidia GPUs writing better AMD drivers than AMD themselves.
>>110189530It works the same way it worked in 2020 -like shit
>>110192684>36GB VRAM>presumably a sizable die on a newer, supply-limited process node (wafers that could otherwise be allocated for more datacenter/enterprise products)>a mere $999 MSRPguess again
>>110189773Plenty.
>>110189530ROCm is deprecated by Vulkan
>>110189999$10900 in late 2027 will be worth as much as $1090 is worth today.I keep telling you guys ,computer parts arent actually more expensive, your fiat currency is just worth less.A flagship GPU costs about the same as it always did, if you price it in ounces of gold, or McDoubles, or gallons of diesel, instead of hyperinflationary USD.
>>110193871I have yet to see any AI program in the wild actually using Vulkan.
>>110191793apt install firmware-amd-graphicsincluded by default if the installer detects an amd gpuDebian won, windows lost :^)
>>110193900The rapid, insane rise is pricing is not commensurate with that notion at all. Nice try, but that ain't it.
>>110191793Another c00nsumer filtered. This is how I unironically get good prices on secondhand GPus.
>>110194653i use llama.cpp with vulkan since it's faster than rocm for that for some reason
>>110190135>And other hilarious jokes you can tell yourselfLinux will completely dethone Windows and Microslop will end up making a Linux distro instead
>>110189309I didn't see the point in buying a 5080 when i was building a month ago and went with the 5070ti
>>110193900Anon is on the money, same thing with houses as well. If you compare the value of homes compared to boomer rocks like gold/silver it's actually somewhat in line with where they ought to be.
>>110189517>CUDAoverrated slop
>>110193303this, every good game in the last 10 years runs fine on a pc from 15 years ago
>>110197565>sitting here with 24gb of vram so I can play indie roguelikes with an xbox controller on my TVin hindsight I probably couldve just bought a laptop
>>110193004Jashin-chan is amazing!
I'd honestly would rather get an 8GB Nvidia card than the best AMD has to offer at this point.
>>110198033yeah ok retard
>>110198033"at this point"? now that anything you'd use cuda for also supports rocm/vulkan?years and years ago there was some programs i wanted to play with but couldn't because they were strictly only written for cuda, but these days?
>>110189301lol
when was the last time these leakers got anything right because for like 2 years now everything they've leaked has not happened, has not been released and then they just say "it was shelved" as if there was ever any solid proof of shit coming out for sale.They call him reliable but what was the last thing he said that actually happened.
>>110197565At 720p maybe. A 10 year old system will struggle to run games made 10 years ago. <5 years ago? Be serious.
>>110199229his answer is heavily loaded with the word "good"
>>110199229Depends I guess. Not the anon but other than Stalker 2/gamma pretty much everything Ive played for more than ten minutes lately wouldve probably run on my old Pentium 4. >pathogenic>mewgenics>sulfur>rimworld shit like thatStalker 2 certainly didn't justify spending used car money on my pc ngltbqhwy
>>110189301I really hope they don't fuck this up. They haven't had a position this good in decades.
>>110191793I highly doubt this is real. You don't need the gaymen driver pack. Get the professional one that's just the .inf file. That's all you need.
>>110189517KEKDA will stop being relevant in 2 years due to AI
>>110200033dude's too young to know what "have disk" means
>>110197565As someone on a clusterfuck build of a machine, this is mostly true for the shit I play. Some things will slow down like factorio once the factory gets too big. I haven't tried anything newer than that harry potter game epic gave away a few years back and it ran fine.
>>110189820yeah but then the AI boom happened so they were right
>ayymd
>>110189301and all for the cool low prices of $17,999.99
>>110200191>don't have any cards to use internallythis has got to be bullshit, it's impossible
>>110200532its not, the guy writing it is as professional of an engineer as you could get and did a lot of work on fixing amd's shit before throwing in the towel. or at least it was true at the time of the comment
>>110190451>By default wrapper/emulator of some techThey're both C++ compilers you fucking retards.
>>110190448>having fun is for childrenok mong.
>>110200532If you can't trust shilling fanboy retards slinging turds from their memespite folder, then who can you trust?
>>110191067>>110191463>dragonballzWhy do americans put the Z in the end like it's part of the franchise's name or something?
>>110189301Well they're gonna have to be.It's not like we have options anymore.
>>110191793>>have to install 5gb software suite just to install the gpu driversAnon..... the AMD drivers are built-in into the Linux kernel and work OOTB. Why would you do that?
>>110189301insha'Allah they will do it
>>110202023I really like this creature.
>>110202100Lapwing is a cutie
>>110189301Probably not on windows but my current amd is cozily running by games on bazzite
>>110190892>obliterates 5070TiWhy do you lie? You can just say it's cheaper for similar performance
>>110191049>>my wife GPU cucked me>that's a new oneI gaslit myself into a 5070ti build because I was on a 1060 3gb laptop from 2018 and we were in microcenter and my gf wanted to buy a $4k laptop cuz it had a touchscreen, and the 5070ti it has is a joke compared to desktop. So I built the rig and gave her my old laptop to play sims4 porn mods
>>110202518mobile 5090 = 24gb vram 5070the tdp/clockspeed difference between desktop and mobile blackwell is huge
>>110202504>she believes goyslop generated by the least accurate slop machine out thereIt's not AI psychosis. It's SUPER UNINTENLIGENCE psychosis.
>>110189906All your base are belong to us
>>110191269THIS
>>110202303No u
