Casual RTX 5000 and RX 9000 Discussion

[Bibr]

So, CES 2025 is happening. Nvidia and AMD announced their new gpus. What are your thoughts ?

Quick summary for everyone that missed it:

>AMD
AMD mentioned new gpus, rx 9070 (xt) and 9060. There's no info on pricing or performance. A single picture from press briefing shows that rx 9070 xt will be around rx 7900gre / 7900 xt levels. Board partners leaked that both 9070 xt and 9070 will have 16gb of vram, no idea about rx 9060. Grim ngl

>Nvidia
They're cooking.
Announced 4 gpus. Specs below: Model; VRAM; TDP; Price; Release date
RTX 5090; 32gb gddr7; 575w; 2000$; 30.01.2025
RTX 5080; 16gb gddr7; 360w; 1000$; 30.01.2025
RTX 5070ti; 16gb gddr7; 300w; 750$; 02.2025
RTX 5070; 12gb gddr7; 250w; 550$; 02.2025

>wacky features and claims
Nvidia announced DLSS 4 that promises to enable high fps gaming by using AI (of course it's AI) to predict and render up to 3 frames ahead while your gpu is working on another frame. Good ? Bad ? Personally I'm against any software magic that doubles frames or upscales from thin air. I'm curious about graphical artifacts in edge scenarios like background tries or big crowds of people. Will it become a mush that will flicker back to original state every 1/60th of a second, or AI will hold itself together and remain coherent till next full frame will arrive ?

To reiterate. It's not meant to push 15 fps to 60 fps, more like 60 fps to 180-200 fps. Of course there will be some overhead and such techniques work best when fresh source frame appears as often as possible. With original frame gen (one real frame, one predicted) people already complained about input lag. I'm curious how bad it will be when for 3/4th of a second you will be disconnected from a game, watching an AI generated slide show.

This "fps AI magic" led to rather outrageous claim, that was even brought up on Nvidia's conference:
RTX 5070 = RTX 4090

Consult meme below (how fitting) for explanation:


This new 3 frames ahead AI rendering is only avaiable on rtx 5000 gpus. If you have an RTX 4000 GPU and use frame gen, Nvidia announced some optimalizations that can gain up to 10% more fps and up to 5% vram usage reduction.

Also a lot about AI compute. I can make a small write up if someone is interested

 

[Khastle]

All this fancy tech only to bought out in bulk for crypto mining

Both sound cool tho, curious how good the AI will be.

 

[Bibr]

All this fancy tech only to bought out in bulk for crypto mining

Both sound cool tho, curious how good the AI will be.

ngl crypto mining is long gone and busted. I'm curious about availability of rtx 5090. I think that many AI companies will be buying them in bulk just for that sweet 32gb of vram. Earlier you had to pay serious money to nvidia to get a gpu with that much memory

 

[ignika98]

Frame generation and AI upscaling are trash gimmicks that Nvidia are trying to gaslight the public into thinking they need. The truth is GPU tech is long past the point of diminishing returns now. All new GPUs are good for nowadays is to give game devs an excuse to be lazy and not think about optimization. Your game runs below 60 fps at 1080p? Just turn on DLSS bro it'll get you above 100! Don't mind the blurry mess of shit the game just turned into, though. That's normal.

If companies would actually optimize their games you could run 99% of all modern AAA games on a GTX 1080.

 

[Khastle]

Frame generation and AI upscaling are trash gimmicks that Nvidia are trying to gaslight the public into thinking they need. The truth is GPU tech is long past the point of diminishing returns now. All new GPUs are good for nowadays is to give game devs an excuse to be lazy and not think about optimization. Your game runs below 60 fps at 1080p? Just turn on DLSS bro it'll get you above 100! Don't mind the blurry mess of shit the game just turned into, though. That's normal.

If companies would actually optimize their games you could run 99% of all modern AAA games on a GTX 1080.

More I hear about modern game devs being lazy and trash more I remember how Iwata back in the day managed to compress the almost full Gold and Silver Pokémon code so much that they managed to squeeze Kanto into the game afterwards, old devs were just built different and thanks to no DLC they *had* to release a working product to avoid backlash.

 

[Bibr]

More I hear about modern game devs being lazy and trash more I remember how Iwata back in the day managed to compress the almost full Gold and Silver Pokémon code so much that they managed to squeeze Kanto into the game afterwards, old devs were just built different and thanks to no DLC they *had* to release a working product to avoid backlash.

It's not exactly like this. For every "muh this game is a technical marvel because X" there are dozens low budget flops lost to time. SURE standard of gamedev was way higher, but old times aren't a fairy tale land where all games were bug free and finished on release.

If companies would actually optimize their games you could run 99% of all modern AAA games on a GTX 1080.

Pretty much true. Maaaybe get a higher vram card if you want to do 4k gaming. Unfortunately industry is full of Unreal engine slop and """games""" made in unity that are lazy asset flips. We would've been so better off if games were actually written, not made in GUI wizards that implement random ass blocks of code

 

[Khastle]

It's not exactly like this. For every "muh this game is a technical marvel because X" there are dozens low budget flops lost to time. SURE standard of gamedev was way higher, but old times aren't a fairy tale land where all games were bug free and finished on release.

Yeah but they were held to a higner standard, you can quite frankly see it in older games with how much more stuff they tried to pack in and etc, the developers genuinely seemed more passionate about their projects too. Mind you that era is long gone now in mainstream game dev, games are too costly to make now and have to make a profit or bust hence why so many are churned about with bugs and what not, take profit now fix later attitude.

 

[Bibr]

Can't disagree with that. It feels like game industry would've been in so much better place if pretty much every game got extra 2 years of time

Not sure how "biggest fans" would react to this tho

 

[Tabula Rasa]

are the 5000's good or should I skip on them?

 

[Bibr]

are the 5000's good or should I skip on them?

Consult the following

 

[Bibr]

A lot of old news, still gonna make an update

>Nvidia

First cat out of the bag, RTX 5090. Dam that was disapointing.
Reviews show around 30% gain in performance for extra 25% of power. Reference model is impressive technically, but still falls behind AIB cards with fat coolers. Reference being 2 slot design might fit in some smaller builds, but it will be toasty.
RTX 5080 reviews drop this wednesday. I have no hopes for this gpu, most likely it will follow 5090's way of linear performance gain to extra power draw. It's such a dead on arrival nothing burger of a gpu with 16gb of vram. As if Nvidia is already preparing a 5080ti version with more memory.

Still waiting for conclusive tests about dlss4 and that new frame gen. I heard good stuff about the first, latter has as much artifacts as you imagine.

>AMD
Ow boy, the clown industry is in full swing.
rx 9070 (xt) was meant to be released in late january, was posponed to march. Not everyone got the memo and few regional social media accounts started posting about them. Still no info about release price, tho more and more sources talk about 550$ for 9070 xt. Performance is also unknown, leaks vary from rtx 4080 super in raster to a bit below rx 7900 xt. Gotta wait till march IG

>Intel
Honorable mention, cause their latest gpus B580 and B570 are a bit of a paper launch. At least here in europe.
People online rave about arc. Saying that it's a good step and intel will quickly gain gpu marketshare with this one. I'm really sceptical.
In performance Arc B580 is around rtx 3060ti for 250$ msrp and Arc B570 is around rtx 3060 for 220$ (nice 30$ upsell for 20% more fps)

If those gpus were sold at those prices... From what I can see on newegg cheapest B580 is listed for 380$ and B570 for 250$. That's a lot of money for a gpu that goes neck a neck with flagships from 2018....

 

[Khastle]

Seem that new Chinese AI that just dropped? Nvidia is in trouble lol, Deepseek is open source, smarter, and way cheaper too, some hackers have even tried to takedown the site due to the stock market havoc its been causing lol

 

[Bibr]

Seem that new Chinese AI that just dropped? Nvidia is in trouble lol, Deepseek is open source, smarter, and way cheaper too, some hackers have even tried to takedown the site due to the stock market havoc its been causing lol

China won by giving first taste for free. There's still absurd amount of censorship on this model (ask about 1989) and online you interact with is absurdly big. My napkin math said that you need roughtly 4xrtx 4090 (or 3x 5090) to run it in heavily compressed form, which will have gimped logic and low reasoning.

Jews at openai got too greedy. 200$ a month for access to their best model is just absurd. Chinese model often goes out due to high demand and their pricing clearly states that current prices are promotional. I'm curious where this will go. Nvidia is still the only manufacturer that can provide AI companies with gpus, so it's not like they will go under because of one freak project