• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Nvidia RTX 6000 series (codename Rubin)

Soldato
Joined
6 Feb 2019
Posts
20,975
Let's rumours begin

Jensen showed this at CES this year and said it's his shield but I think it's actually the RTX 6090




The RTX 6000 generation will likely be manufactured on the TSMC 3nm node and once again feature a heavy focus on AI. Rubin based products for data centres will release some months before consumer parts, with the first R100 Rubin based parts rumoured to be arriving by the end of 2025 or early 2026, followed by consumer parts in the 2nd half of 2026
 
Last edited:
Realistically the 6090 will be the only card to beat the 5090. The 6080 will be around half the 6090 in price and spec. Vram will be the minimum amount for all but the top end sku. The rest of the stack will get a 10-15% bump gen on gen.

There is just no incentive for Nvidia to compete with itself.

AMD is coming back next year, with a new UDNA GPU architecture on 3nm. People hope it will better than RDNA GPU were
 
Last edited:
I wonder what DLSS 5 will look like.
I think it'll be some form of adaptive/variable multi-frame generation, where you set a target framerate and it'll generate only what's needed to reach it. Currently implementations of framegen (DLSS 3 and 4) is always exactly 1 or 2 or 3 extra frames generated.
Like for example Stalker 2 can't quite reach 120fps with one extra frame generated, but using MFG to generate two extra frames would be completely overkill and will push the base framerate down to fit the cap, resulting in sluggish input latency.
DLSS 5 will probably introduce a method where it fills in the small gaps with variable amounts of generated frames instead of exactly one or two or three extra generated frames all the time.
That's my guess anyway.


X16 frame generation

 
Jensen announces TSMC has taped out six new Nvidia chips based on Rubin architecture and Jensen is visiting TSMC to inspect the these next generation chips. Jensen also said buying TSMC shares is smart

 
Last edited:
Rubin will still use monolithic die design and surprised Rubin CPX GPU used GDDR7 memory instead of HBM4, maybe Nvidia will launch Rubin gaming GPUs at CES in Jan 2026 as AMD confirmed Lisa Su will be at CES in Jan 2026 likely to announce next generation Zen 6 CPUs and RDNA5 GPUs.

As per Nvidia's roadmap, this first Rubin GPU uses GDDR7 but there are also Rubin GPU's coming that use HBM4e
 
Saw a video last night saying the 6090 will have 30% more cuda and rops then a 5090 so with out any other boost should be 30% there and the node is shrinking 40% what would you say the gains there will be ?

I'm hoping 55% faster then 5090 minimum and 100% over 4090 l. Would make a nice upgrade

30% more cores going to 3nm seems like a let down

They'll push clocks with the power savings, it will probably still be a 600w GPU

I think it will close to 100% faster than 4090

Still not fast enough to run Borderlands 4 at native 4k 60fps
 
Last edited:
  • Haha
Reactions: P.B
What's the bus width got to do with it though, by itself it doesn't increase performance, memory bandwidth it just acts as a bottleneck on the GPU if the GPU is powerful enough, which you can get around anyway by adding more cache or using HBM. In case faster gddr7 is available, you don't need higher bus width to get more gddr7 bandwidth
 
Last edited:
the first R100 Rubin based parts rumoured to be arriving by the end of 2025 or early 2026, followed by consumer parts in the 2nd half of 2026


Timeline update for a delay unfortunately. When I made this thread in January 2025 the belief was that Rubin server would launch very early 2026 and then based on the 6 month Nvidia standard cadence between server and desktop, desktop RTX6000 would launch late 2026.

However Jensen now says Rubin server enters mass production this time next year - meaning late 2026. So now we're looking at desktop RTX6000 cards only launching mid to late 2027, meaning we're still 18 to 22 months away from new Nvidia GPUs

Kind of annoying for me as I wanted to build a new PC next year and now I have to wait for 2027

 
Last edited:
Means I get a bigger chunk of time with my 5090 which is why I got in as early as possible. Bring on the 6090.

Sort of the same thing here, I got in early with my 4090 purchase and the longer we still have to wait for the RTX6000 launch, the better value my original purchase becomes. 4090 launched at $1600 in October 2022, I ordered but only received mine in November. Fast forward, now I might only get to replace it in mid 2027 with a 6090 - about 4 and a half years later. That works to under $1 a day to own what will still be the second fastest gaming gpu available before RTX6000 launches and its cheaper than renting hardware from Nvidia's streaming service as well.

Never the less I'm still sad, because there are other components in my build I want to replace and prefer just doing a total new build at once. Some of my parts are barely hanging on, my case fans for example are now dying one after another
 
Last edited:
I don't tend to follow the server stuff but is it always server GPU's first and then desktop GPU's ?


I don't think it was always the case but it has been for the two or three generations at least and I don't see it changing because the profit margin is much higher on server products and server products make up something like 90% of their revenue now
 
Last edited:
Back
Top Bottom