Nvidia announces next-gen RTX 5090 and RTX 5080 GPUs

Similar CUDA core counts for most SKUs compared to last gen (except in the 5090 vs. 4090 comparison). Similar clock speeds compared to the 40-series.

The 5090 just has way more CUDA cores and uses proportionally more power compared to the 4090, when going by CUDA core comparisons and clock speed alone.

All of the "massive gains" were comparing DLSS and other optimization strategies to standard hardware rendering.

Something tells me Nvidia made next to no gains for this generation.

This is maybe a dumb question, but why is it so hard to buy Nvidia GPUs?

I can understand lack of supply, but why can't I go on nvidia.com and buy something the same way I go on apple.com and buy hardware?

I'm looking for GPUs and navigating all these different resellers with wildly different prices and confusing names (on top of the already confusing set of available cards).

The increasing TDP trend is going crazy for the top-tier consumer cards:

3090 - 350W

3090 Ti - 450W

4090 - 450W

5090 - 575W

3x3090 (1050W) is less than 2x5090 (1150W), plus you get 72GB of VRAM instead of 64GB, if you can find a motherboard that supports 3 massive cards or good enough risers (apparently near impossible?).

Pretty interesting watching their tech explainers on YouTube about the changes in their AI solutions. Apparently they switched from CNNs to transformers for upscaling (with ray tracing support) if I understood correctly though for frame generation makes even more sense to me.

32 GB VRAM on the highest end GPU seems almost small after running LLMs with 128 GB RAM on the M3 Max, but the speed will most likely more than make up for it. I do wonder when we’ll see bigger jumps in VRAM though, now that the need for running multiple AI models at once seems like a realistic use case (their tech explainers also mentions they already do this for games).

Even though they are all marketed as gaming cards, Nvidia is now very clearly differentiating between 5070/5070 Ti/5080 for mid-high end gaming and 5090 for consumer/entry-level AI. The gap between xx80 and xx90 is going to be too wide for regular gamers to cross this generation.

32GB of GDDR7 at 1.8TB/sec for $2000, best of luck to the gamers trying to buy one of those while AI people are buying them by the truckload.

Presumably the pro hardware based on the same silicon will have 64GB, they usually double whatever the gaming cards have.

The most interesting news is that the 5090 Founders' Edition is a 2-slot card according to Nvidia's website:

https://www.nvidia.com/en-us/geforce/graphics-cards/50-serie...

When was the last time Nvidia made a high end GeForce card use only 2 slots?

> will be two times faster [...] thanks to DLSS 4

Translation: No significant actual upgrade.

Sounds like we're continuing the trend of newer generations being beaten on fps/$ by the previous generations while hardly pushing the envelope at the top end.

A 3090 is $1000 right now.

> GeForce RTX 5070 Ti: 2X Faster Than The GeForce RTX 4070 Ti

2x faster in DLSS. If we look at the 1:1 resolution performance, the increase is likely 1.2x.

Let's see the new version of frame generation. I enabled DLSS frame generation on Diablo 4 using my 4060 and I was very disappointed with the results. Graphical glitches and partial flickering made the game a lot less enjoyable than good old 60fps with vsync.

I have a serious question about the term "AI TOPS". I find many conflicting definitions while others say nothing. A meaningful metric should at least be well defined on its own term, like in "TOPS" or expanded "Tera Operations Per Second", what operation it will measure?

Seemingly NVIDIA is just playing number games, like wow 3352 is a huge leap compared to 1321 right? But how does it really help us in LLMs, diffusion models and so on?

I will be astonished if I'll be able to get a 5090 due to availability. The 5080's comparative lack of memory is a buzzkill -- 16 GB seems like it's going to be a limiting factor for 4k gaming.

Does anyone know what these might cost in the US after the rumored tariffs?

Looks like a bit dud, though given their competition and where their focus is right now maybe expected.

Going from 60 to 120fps is cool. Going from 120fps to 240fps is in the realm of diminishing returns, especially because the added latency makes it a non starter for fast paced multiplayer games.

12GB VRAM for over $500 is an absolute travesty. Even today cards with 12GB struggle in some games. 16GB is fine right now, but I'm pretty certain it's going to be an issue in a few years and is kind of insane at $1000. The amount of VRAM should really be double of what it is across the board.

Looks like most of the improvement is only going to come when DLSS4 is in use and its generating most of the frame for Ray Tracing and then also generating 3 predicted frames. When you use all that AI hardware then its maybe 2x, but I do wonder how much fundamental rasterisation + shaders performance gain there is in this generation in practice on the majority of actual games.

I have trained transformers on a 4090 (not language models). Here’s a few notes.

You can try out pretty much all GPUs on a cloud provider these days. Do it.

VRAM is important for maxing out your batch size. It might make your training go faster, but other hardware matters too.

How much having more VRAM speeds things up also depends on your training code. If your next batch isn’t ready by the time one is finished training, fix that first.

Coil whine is noticeable on my machine. I can hear when the model is training/next batch is loading.

Don’t bother with the founder’s edition.

It's a shame to see they max out at just 32GB, for that price in 2025 you'd be hoping for a lot more, especially with Apple Silicon - while not nearly as fast - being very usable with 128GB+ for LLMs for $6-7k USD (comes with a free laptop too ;))

Ooo, that means its probably time for me to get a used 2080, or maybe even a 3080 if I'm feeling special

The interesting part to me was that Nvidia claim the new 5070 will have 4090 level performance for a much lower price ($549). Less memory however.

If that holds up in the benchmarks, this is a nice jump for a generation. I agree with others that more memory would've been nice, but it's clear Nvidia are trying to segment their SKUs into AI and non-AI models and using RAM to do it.

That might not be such a bad outcome if it means gamers can actually buy GPUs without them being instantly bought by robots like the peak crypto mining era.

As a lifelong nvidia consumer, I think it's a safe bet to ride out the first wave of 5xxx series GPUs and wait for the inevitable 5080/5070 (GT/Ti/Super/whatever) that should release a few months after with similar specs and better performance based on whatever the complaints surrounding the initial GPUs lacked.

I would expect something like the 5080 super will have something like 20/24Gb of VRAM. 16Gb just seems wrong for their "target" consumer GPU.

Any advice on how to buy the founders edition when it launches, possibly from folks who bought the 4090 FE last time around? I have a feeling there will be a lot of demand.

* MegaGeometry (APIs to allow Nanite-like systems for raytracing) - super awesome, I'm super super excited to add this to my existing Nanite-like system, finally allows RT lighting with high density geometry

* Neural texture stuff - also super exciting, big advancement in rendering, I see this being used a lot (and helps to make up for the meh vram blackwell has)

* Neural material stuff - might be neat, Unreal strata materials will like this, but going to be a while until it gets a good amount of adoption

* Neural shader stuff in general - who knows, we'll see how it pans out

* DLSS upscaling/denoising improvements (all GPUs) - Great! More stable upscaling and denoising is very much welcome

* DLSS framegen and reflex improvements - bleh, ok I guess, reflex especially is going to be very niche

* Hardware itself - lower end a lot cheaper than I expected! Memory bandwidth and VRAM is meh, but the perf itself seems good, newer cores, better SER, good stuff for the most part!

Note that the material/texture/BVH/denoising stuff is all research papers nvidia and others have put out over the last few years, just finally getting production-ized. Neural textures and nanite-like RT is stuff I've been hyped for the past ~2 years.

I'm very tempted to upgrade my 3080 (that I bought used for $600 ~2 years ago) to a 5070 ti.

I have a feeling regular consumers will have trouble buying 5090s.

RTX 5090: 32 GB GDDR7, ~1.8 TB/s bandwidth. H100 (SXM5): 80 GB HBM3, ~3+ TB/s bandwidth.

RTX 5090: ~318 TFLOPS in ray tracing, ~3,352 AI TOPS. H100: Optimized for matrix and tensor computations, with ~1,000 TFLOPS for AI workloads (using Tensor Cores).

RTX 5090: 575W, higher for enthusiast-class performance. H100 (PCIe): 350W, efficient for data centers.

RTX 5090: Expected MSRP ~$2,000 (consumer pricing). H100: Pricing starts at ~$15,000–$30,000+ per unit.

Anyone has any info on Node? Can't find anything online. Seems to be 4nm but performance suggest otherwise. Hopefully someone do a deep dive soon.

Does any game need 32gb VRAM. Did they even use the full 24Gb of the 4090s?

It seems obvious to me that even NVIDIA knows that 5090s and 4090s are used more for AI Workloads than gaming. In my company, every PC has 2 4090s, and 48GB is not enough. 64GB is much better, though I would have preferred if NVIDIA went all in and gave us a 48GB GPU, so that we could have 96GB workstations at this price point without having to spend 6k on an A6000.

Overall I think 5090 is a good addition to the quick experimentation for deep learning market, where all serious training and inference will occur on cloud GPU clusters, but we can still do some experimentation on local compute with the 5090.

Way too little memory. :(

Can anyone suggest a reliable way to procure a GPU at launch (in the EU)?

I always end up late to the party and the prices end up being massively inflated - even now I cant seem to buy a 4090 for anywhere close to the RRP.

575W TDP for the 5090. A buddy has 3x 4090 in a machine with a 32 core AMD cpu must be putting out close to 2000W of heat at peak if he switched to 5090. Uff

You also need to upgrade your air conditioner

One thing I always remember when people say a 2k gpu is insanity. How many people get a 2k ebike. a 100k weekend car. a 15k motorcycle to use once a month. a time share home. Comparatively a gamer using it even a few hours a day for 3k 4090 build is really an amazing return on that investment.

Some of the better video generators with pretty good quality can run on the 32gb version. Expect lots of AI generated videos with this generation of videocards. Price is steep and we need another 9700 ati successtory for some serious nvidia competition. Not going to happen anytime soon I am afraid.

interesting launch but vague in its own way like the one from AMD (less so but in a different way).

it is easy to be carried away with vram size, but keeping in mind that most people with apple silicon (who can enjoy several times more memory) are stuck at inference, while training performance is off the charts through cuda hardware.

the jury is yet to be out on actual ai training performance, but i bet 4090, if sold at 1k or below, would be better value than lower tier 50 series. the "ai tops" of the 50 series is only impressive for the top model, while the rest are either similar or with lower memory bandwidth despite the newer architecture.

i think by now the training is best left on the cloud and overall i'd be happy rather owning a 5070 ti at this rate.

Would have been nice to get double the memory on the 5090 to run those giant models locally. Would've probably upgraded at 64gb but the jump from 24 to 32gb isn't big enough

Gaming performance has been plateaued for some time now, maybe an 8k monitor wave can revive things

Smaller cards with higher power consumption – will GPU water-cooling be cool again?

The most interesting aspect here might be the improved tensor cores for AI workloads - could finally make local LLM inference practical for developers without requiring multiple GPUs.

Mmm i think my wallet Is safe since i only play SNES and old dos games.

AI is going to push the price closer to $3000. See what happened with crypto a couple of years back.

Official release: https://nvidianews.nvidia.com/news/nvidia-blackwell-geforce-...

(https://news.ycombinator.com/item?id=42618849)

Finally I will be able to run Cities Skylines 2 at 60fps!

GPU stands for graphics prediction unit these days

Meh. Feels like astronomical prices for the smallest upgrades they could get away with.

I miss when high-end GPUs were $300-400, and you could get something reasonable for $100-200. I guess that's just integrated graphics these days.

The most I've ever spent on a GPU is ~$300, and I don't really see that changing anytime soon, so it'll be a long time before I'll even consider one of these cards.

I'm really disappointed in all the advancement in frame generation. Game devs will end up relying on it for any decent performance in lieu of actually optimizing anything, which means games will look great and play terribly. It will be 300 fake fps and 30 real fps. Throw latency out the window.

More fake poor frames at less price

It looks like the new cards are NO FASTER than the old cards. So they are hyping the fake frames, fake pixels, fake AI rendering. Anything fake = good, anything real = bad.

This is the same thing they did with the RTX 4000 series. More fake frames, less GPU horsepower, "Moore's Law is Dead", Jensen wrings his hands, "Nothing I can do! Moore's Law is Dead!" which is how Intel has been slacking since 2013.

Do they come with a mini nuclear reactor to power them?

Did they discontinue Titan series for good?

Somewhat related, any recommendations for 'pc builders' where you can configure a PC with the hardware you want, but have it assembled and shipped to you instead of having to build it yourself? With shipping to Canada ideally.

I'm planning to upgrade (prob to a mid-end) as my 5 year old computer is starting to show it's age, and with the new GPUs releasing this might be a good time.