NVIDIA RTX 6000 (Rubin): specs, release date, and what we really know in 2026

January 2025, NVIDIA launches the RTX 5090 and the RTX 5080. March 2025, the RTX 5070 and 5070 Ti finally arrive. And in June 2026, we still haven't seen the slightest RTX 5060 — NVIDIA is taking its time, as usual. But in the meantime, the first leaks about the next generation are starting to come out.

We talk about architecture Rubin , of the transition to the TSMC 3nm , of specs that stand out for the RTX 6090, and of a potential DLSS 5 who would no longer do upscaling but straight up neural rendering. It looks strong - on paper.

Let's be clear from the start: nothing is official NVIDIA has not announced a GeForce RTX 6000. The following is a summary of leaks, credible rumors, and what can reasonably be extrapolated from the NVIDIA cycle. We separate the true from the speculative in each section.

The code name of the next architecture is Rubin . That's official - NVIDIA announced it at Computex 2024, and we reviewed the details in 2025 and 2026. The full roadmap looks like this:

• Blackwell — RTX 5000, it's now (2025)
• Blackwell Ultra — refresh on the datacenter side, not necessarily visible on GeForce
• Ruby — the real next-gen, announced for the second half of 2026 in datacenter
• Ultra Ruby — refresh planned for 2027
• Feynman — the architecture of the future, horizon 2028 (from The Next Platform)

The important thing: Rubin arrives first on the datacenter and AI side, and the consumer version of GeForce always arrives later. If the Rubin datacenter chips come out at the end of 2026, we can expect some GeForce RTX 6000 early-mid 2027 It's not a certainty, but it's the pattern we've been observing for years.

On the platform side, NVIDIA has confirmed that Rubin integrates the Vera CPU, NVLink 6, and that the datacenter GPU delivers 50 petaflops in NVFP4 . Obviously, this has no direct relation to the gaming performance of a future RTX 6080 — but it gives an idea of the architectural leap between Blackwell and Rubin.

The current RTX 5000s use TSMC 4nm (N4). The Rubin generation would switch to the TSMC N3 — the famous 3nm. Not officially confirmed by NVIDIA for the GeForce, but it's the most solid assumption circulating.

Concretely, the transition from 4nm to 3nm allows for more transistors to be packed into the same area. More CUDA cores, more cache, or a more compact die with equal performance. Historically, each node jump at NVIDIA yields between 15 and 30% energy efficiency gains at equivalent performance. Or conversely, more performance at equal TDP.

The flip side of the coin is the bill. TSMC 3nm wafers cost more than 20,000 dollars each , against around 14,000-16,000 for the 4nm. And NVIDIA will not absorb the difference. Get ready for higher prices — we'll come back to that later.

This is the most exciting part. And the least reliable. Several leakers have thrown out presumed specs for the first three models in the range. We'll detail them, but keep in mind that all of this could change ten times before release.

192 SM for the RTX 6090 , or about 13% more than the RTX 5090. The memory bus would remain at 512 bits with 32GB of GDDR7 — logical, it's already the ceiling on the 5090. The gains would come from the number of cores and the architecture itself, not from the amount of VRAM.

On the performance side, the most optimistic leaks announce around 30-35% increase in rasterization compared to the RTX 5000, and downright the double in ray tracing It is ray tracing that would make the biggest leap with 5th generation RT Cores. If true, full path tracing would become playable without DLSS as a crutch. Big "if".

For the RTX 6080 and the RTX 6070 , we have fewer details. But 20 GB and 16 GB of VRAM respectively, that would be really good news. The current RTX 5080 has 16 GB, the 5070 has 12. If NVIDIA increases the VRAM across the board, it's a strong signal - games in 2027 will need it, between 4K textures and local AI that consumes memory.

DLSS 4, it's already impressive: Multi Frame Generation, Ray Reconstruction, Super Resolution. The complete package. But DLSS 5 would go even further. According to Igor's Lab, NVIDIA is working on what they call the guided 3D neural rendering .

The idea: instead of creating a classic image and then enhancing it with AI, the neural network would directly participate in generating the image. Lighting, reflections, certain visual effects would be calculated by AI rather than by the traditional rendering pipeline. A paradigm shift, if it lives up to its promises.

For you, what does it mean? Games with photorealistic lighting without blowing up your framerate. Full ray tracing - global illumination, reflections, shadows - without dropping to 30 FPS. That's the promise anyway.

Will DLSS 5 be exclusive to the RTX 6000, or backward compatible with the 5000? No one knows. But historically, NVIDIA keeps its big features to push the new gen. DLSS 3 and Frame Generation were exclusive to the RTX 4000 while the RTX 3000 could technically have run it. It's a safe bet that this will happen again.

NVIDIA has never communicated a date for the GeForce RTX 6000. But we can cross-reference:

• Rubin datacenter announced for the second semester 2026
• The datacenter → GeForce cycle usually takes 6 to 12 months
• RTX 4000 (October 2022) → RTX 5000 (January 2025) = about 27 months
• NVIDIA likes launches at CES (January) or in the fall

By combining all this, the most likely window is Q1-Q2 2027 . Announcement at CES 2027 in January, first models (6090 and 6080) in February-March, the rest of the range in spring-summer. Scenario consistent with NVIDIA's history.

But big caveat: if the RTX 5090 Ti or TITAN Blackwell is released in Q3 2026 — it's a rumor reported by Igor's Lab — it could push back the RTX 6000 to late 2027. NVIDIA won't release its Blackwell halo product just to bury it 6 months later.

Let's talk cash. The NVIDIA trend for three generations is upward. RTX 3090 at 1,549 dollars. RTX 4090 at 1,599 dollars. RTX 5090? 1,999 dollars You see where this is going.

If we add the cost of TSMC 3nm, the inflation on GDDR7 components, and the fact that NVIDIA has no serious competition in the high-end segment — AMD has dropped the ultra-premium segment since the RX 7900 XTX — we can expect:

• RTX 6090 — 2,000 to 2,500 dollars
• RTX 6080 — 1,000 to 1,200 dollars
• RTX 6070 — 600 to 700 dollars

In euros, with taxes and the margin of European retailers, add 15-20%. An RTX 6090 for 2,500-3,000 euros at LDLC or Amazon.fr, it's not crazy — it's the natural trajectory of this market. Unless AMD comes back swinging with the RX 10900 XT at an aggressive price.

The million-dollar question. The answer depends on your current setup.

Do you have an RTX 3000 or older? Don't wait. The RTX 5000 are here, they pack a punch, and you're not going to wait another 18 months with your old card hoping for a GPU that isn't even announced. Upgrade now, enjoy now.

Do you have an RTX 4070 Ti or better? There, it's up for debate. The jump from RTX 4000 to 5000 is decent but not huge in classic rasterization. If you play in 1440p and you're happy with your performance, waiting for the 6000 series can make sense. The leap over two generations will be much more significant.

Did you just buy an RTX 5000? Take advantage of it and stop reading leaks. Your card will last for years. FOMO is the worst enemy of the PC gamer — there will always be a faster GPU in 18 months.

Impossible to talk about the RTX 6000 without mentioning what AMD is preparing. The architecture RDNA 5 is in development, and the Radeon RX 10900 XT would be the flagship planned for 2027. AMD clearly aims to return to the high end after conceding ground to NVIDIA on the current gen.

If AMD arrives with a competitive product against the RTX 6080 or 6090, it could finally put pressure on prices. Because let's be honest: if NVIDIA can sell its GPUs for 2,500 euros without opposition, they will do it. Competition is the only thing that lowers prices in this market.

No. The Rubin architecture is confirmed on the datacenter side, but NVIDIA has made no official announcement for the consumer GeForce RTX 6000. All we have are leaks and extrapolations from the usual cycle.

The most consistent estimate with the NVIDIA cycle is Q1-Q2 2027. Nothing set in stone. It could be pushed back to late 2027 if NVIDIA releases an intermediate model (RTX 5090 Ti) this year.

32 GB according to leaks, with a 512-bit bus. It's identical to the RTX 5090, which makes sense — 32 GB is already massive for gaming. The gains would rather come from the higher bandwidth of the new generation GDDR7.

Yes, the RTX 6000 will remain in PCIe. If they use PCIe 6.0 (not confirmed), a PCIe 4.0 motherboard could slightly limit performance in some cases. The power connector will probably still be the 12V-2x6 we have known since the RTX 4000.

Probably not. Jumping just one generation rarely justifies the wait and resale. Enjoy your 5080, it will run like a beast for a long time.

The RTX 6000 based on the Rubin architecture is the logical continuation of Blackwell. The first leaks outline an attractive generation: 3nm engraving, more cores, more VRAM on mid-range models, doubled ray tracing, and DLSS 5 focused on neural rendering.

But we are still early in the rumor cycle, and everything can change by 2027. Prices will probably rise again. And NVIDIA remains the sole master of the schedule.

We will update this article as new leaks and announcements come in. In the meantime, if you need a GPU now, the RTX 5000 and AMD's RX 9000 do the job very well.