Nvidia RTX 50 Series Blackwell: all GPUs, specs, prices, and Super rumors in 2026

Nvidia took its time. Announced at the end of 2024, gradually deployed since January 2025, the family GeForce RTX 50 Series is now complete by June 2026. Six models, from the RTX 5090 to the RTX 5060, all based on the architecture Blackwell . And already, there is talk of a Super refresh for the end of the year.

We're not going to give you a master class on each chip. The idea here is to give you the real specs, the prices we observe in France - not the fantasized MSRP - and to sort out what's worth it and what's just marketing. Because spending 4,000 bucks on an RTX 5090, that's something to think about.

Before we get into each card, two minutes on what Blackwell brings compared to Ada Lovelace (RTX 40 Series). Because it's not just "more CUDA cores" this time.

The big piece is the DLSS 4 with Multi Frame Generation . Where the RTX 40 Series generated a single additional frame between each native frame, Blackwell generates three . Total: 4 frames displayed for 1 frame calculated by the GPU. In January 2026 at CES, Nvidia even announced the DLSS 4.5 with a Dynamic mode that goes up to 6 frames. Concrete result: full ray tracing games running at 180+ FPS in 1440p. Cyberpunk 2077 with full path tracing, smooth like an indie game. It's pretty crazy.

Behind, there are the 5th generation Tensor Cores who now use transformer models instead of the old CNNs. The AI model runs 40% faster, consumes 30% less VRAM, and runs only once per native frame — even when generating multiple frames. 4th generation RT Cores follow the same improvement logic.

And then there is the Flip Metering hardware. A very technical thing but which changes everything for frame pacing: frame timing management is directly integrated into the GPU's display engine, no longer managed by the CPU. As a result, timing variability is reduced by 5 to 10 times. In concrete terms, this means zero micro-stutter even with Multi Frame Generation at maximum. Combined with Reflex 2 who warps the image at the last moment to reduce latency, it's a real leap in input lag.

The entire range uses the GDDR7 — a first. Faster, more efficient. But it is also one of the reasons for the high prices at the moment.

21,760 CUDA cores. 32GB of GDDR7 on a 512-bit bus. 575W TDP. The RTX 5090 is a monster, and so is its price.

The official MSRP is at 2,349 euros Except that in reality, good luck finding one at this price. On LDLC, custom models start around 3,700 euros for an ASUS ROG Astral and it goes beyond 5,000 euros for the WATERFORCE versions or limited editions. On the secondary market, some listings have reached 10,000 euros. Yes, ten thousand.

The reason? The AI demand. The guys running local LLMs, image generation, fine-tuning — they want the 32GB GDDR7 VRAM. Gaming almost takes a back seat for this card. If you just want to play, even in 4K Ultra, the 5080 does the job for a lot less.

• GPU: GB202 — 21,760 CUDA cores, 680 Tensor Cores, 170 RT Cores
• Video RAM: 32 GB GDDR7, 512-bit bus
• TDP: 575 W - 1000 W power supply recommended
• Perf: ~104.8 TFLOPS FP32
• Actual price France: 3,700 to 5,500 euros (LDLC, Materiel.net, Alternate)

The card that most wealthy gamers should aim for. 10,752 CUDA cores, 16 GB of GDDR7, 360 W TDP. It's exactly half the CUDA cores of the 5090, but in real life, the difference in pure gaming is rarely just double.

The MSRP was announced at 1 179 euros . In practice, models can be found starting at 1,250 euros at Cybertek and around 1 380 euros at LDLC for a Gainward Phoenix. High-end models like ASUS ROG Astral or AORUS XTREME WATERFORCE are around 1,800-2,000 euros. It's expensive, but it's a whole different world from the outrageous prices of the 5090.

In 4K Ultra without DLSS, the 5080 delivers very comfortable framerates in almost all games. With DLSS 4 enabled, you exceed 120 FPS even in the most demanding titles. And honestly, if you play in 1440p, it's almost overkill — but at least you're set for 4-5 years.

• GPU: GB203 - 10,752 CUDA cores
• Video RAM: 16 GB GDDR7, 256-bit bus
• TDP: 360 W
• Actual price in France: 1,250 to 2,000 euros depending on the model

8,960 CUDA cores, 16GB GDDR7, 300W. The RTX 5070 Ti occupies a strange niche. It costs significantly more than the 5070 for performance gains that vary between 15 and 25% depending on the games.

On the price side, the MSRP was 884 euros . In store, it can be found starting from 898 euros at Idealo/TopAchat for a Gainward, and it quickly climbs above 1,000 euros for MSI or ASUS models. Some MLG versions exceed 1,249 euros.

The thing is, the 5070 Ti shares the same 16GB of GDDR7 as the 5080. If your priority is VRAM buffer for 4K textures and mods, it does the job. But if you just want framerate at 1440p, the standard 5070 often remains the best value for money in the range. Moreover, Asus has already declared the 5060 Ti end-of-life — this kind of card always has a fragile positioning.

• GPU: 8 960 CUDA cores
• Video RAM: 16 GB GDDR7
• TDP: 300 W
• Actual price France: 898 to 1,249 euros

6 144 CUDA cores, 12 GB GDDR7, 250 W. This is the card that interests the most people, and it's normal. The RTX 5070 offers Blackwell technologies — DLSS 4, Multi Frame Generation, 4th gen RT Cores — in a more reasonable format and price.

The MSRP of 649 euros was already not given. But in France, actual prices range between 825 and 969 euros . Yes, up to 50% above the recommended price. Demand remains strong, and stocks are regularly under pressure. On LDLC, a Gainward Phoenix starts at 989 euros.

Despite everything, if you play in 1440p — and that's the majority of gamers in 2026 —, the 5070 delivers. In Ultra without DLSS, you hit 75-90 FPS in recent AAA games. With DLSS 4 enabled, we easily exceed 120 FPS. The only weak point: 12 GB of VRAM. It's enough for now, but in 2-3 years with 4K textures becoming more common, it could become tight. That's why Nvidia is preparing a Super version with 18 GB.

• GPU: 6 144 CUDA cores
• Video RAM: 12 GB GDDR7
• TDP: 250 W
• Actual price France: 825 to 969 euros

Released in April 2026. 4,608 CUDA cores, two variants: 8 GB and 16 GB of GDDR7, 180 W TDP. On paper, the RTX 5060 Ti was supposed to be the affordable card in the Blackwell range. In practice, it's more complicated than that.

The 8GB version starts at 329 euros — that's the entry price at Grosbill or TopAchat for an MSI Ventus. But be careful, 8GB in 2026 is starting to be tight. Some recent games like Indiana Jones and the Dark Quarry are already saturating the buffer in 1080p Ultra. The 16GB version is more serene on the VRAM side, but it hits the From 549 to 779 euros depending on the model. At that price, you are dangerously close to an RTX 5070...

In raw performance, the 5060 Ti 16GB delivers about 20% more FPS than an RTX 4060 Ti at 1440p. With DLSS 4, we reach around 180 FPS. That's decent. However, the 128-bit memory bus remains a limitation at high resolutions, and the 448 GB/s bandwidth is the same between the 8GB and 16GB versions. In short, it's a card for uncompromised 1080p/1440p gaming, not for 4K.

• GPU: 4,608 CUDA cores
• Video RAM: 8 or 16 GB GDDR7, 128-bit bus
• TDP: 180 W
• Actual price in France: 329 euros (8 GB) to 779 euros (16 GB custom)

Latest arrival, May 2026. 3,840 CUDA cores, 8GB GDDR7, 145W. It is the most affordable card in the series, with an MSRP at 299 dollars — or about 280-320 euros in France depending on exchange rates and reseller margins.

The benchmarks are not very exciting. The 5060 is barely 4% better than an RTX 4060 Ti 16GB at 1080p — except it has 8GB of VRAM instead of 16. Ouch. DLSS 4 and Multi Frame Generation save the day in terms of perceived FPS, but in raw GPU performance, it's a timid upgrade.

Honestly? If you already have an RTX 4060, upgrading to the 5060 is difficult to justify. The main interest is if you are coming from a GTX 1060 or 1660 and you want to access DLSS 4 and 4th gen ray tracing without blowing your budget. Otherwise, aim for at least the 5060 Ti 16GB.

• GPU: 3,840 CUDA cores
• VRAM: 8 GB GDDR7
• TDP: 145 W
• Estimated price France: ~299 euros

The Super refresh is a classic at Nvidia. And this time, the leaks are quite solid. According to MEGAsizeGPU, TechSpot, and VideoCardz, the RTX 50 Super project is back to active development after being paused in 2025 due to GDDR7 supply issues.

The rumored lineup:

• RTX 5080 Super — 24 GB GDDR7 (instead of 16), full GB203 GPU, ~10,752 CUDA cores, 400+ W. This is the big upgrade: 50% more VRAM.
• RTX 5070 Super — 18 GB GDDR7 (instead of 12), GPU GB205, ~6,400 CUDA cores, 275 W. The sweet spot even sweeter.
• RTX 5060 Super — 12 GB GDDR7, probably GB205 as well, ~250 W. Positioned against AMD's RX 9070 GRE.

The launch window? Q4 2026 , so between October and December. Some leakers even mention early 2027, which would push the RTX 6000 to 2028. Nothing official from Nvidia for now.

If you are hesitating between buying now or waiting, here is the calculation: the Super will mainly address the VRAM issue - 12 GB on the 5070, 8 GB on the 5060, it's a bit stingy. If the 16 GB of your 5070 Ti or 5080 are enough for you, no reason to wait. However, if you were eyeing the 5070 and the 12 GB made you hesitate, the 5070 Super with 18 GB might be worth waiting a few months for.

We're not going to beat around the bush. Here is our opinion by budget range:

Less than 400 euros : RTX 5060 Ti 8 GB. This is the entry ticket for Blackwell. At 1080p, it delivers. But if you can spare 200 euros more, the 16 GB version is much more future-proof.

500-700 euros : RTX 5060 Ti 16 GB. The best compromise in this range. You have DLSS 4, enough VRAM for 2-3 years, and decent performance in 1440p.

800-1 000 euros : RTX 5070. The sweet spot. If you play in 1440p — and that's what we recommend in 2026 —, this is the card. Or a RTX 5070 Ti if you find one around 900 euros.

1,200-2,000 euros : RTX 5080. The high-end that makes sense. 4K Ultra comfortable, 16 GB of VRAM, and you will be set for a good while. It's probably the best pleasure/price ratio in the range if you have the budget.

More than 3,500 euros : RTX 5090. For content creators, AI workloads, or those who simply want the best without looking at the price. In pure gaming, the 5080 offers 80% of the experience for 3 times cheaper.

No. In pure gaming, the difference with the 5080 does not justify the monstrous price difference. The 5090 shines on pro workloads and AI thanks to its 32GB of VRAM. If you just want to play, take the 5080.

If you're aiming for the 5070 and the 12GB of VRAM worry you, waiting for the 5070 Super (18GB, rumored Q4 2026) can be justified. For other models, the current versions are already very good. The 5080 Super will not bring a massive performance gain, just more VRAM.

16 GB. Really. In 2026, several games already saturate 8 GB in 1080p Ultra. The price difference of 150-200 euros is worth it for peace of mind. The 8 GB version is a medium-term trap.

Three reasons: the 20% VAT (the MSRP is excluding US tax), reseller margins that have soared since the shortage, and the AI demand that absorbs part of the high-end stock. The situation is worse on the 5090 than on the mid-range models.

Yes. The Multi Frame Generation with Blackwell's Flip Metering hardware is a real qualitative leap compared to DLSS 3 from the 40 series. The generated frames are more stable, the frame pacing is excellent, and the latency remains contained thanks to Reflex 2. It is the feature that justifies the most the upgrade from an RTX 40.