Nvidia GeForce RTX 50 graphics cards will soon be available, starting with the RTX 5090 and RTX 5080. Price, release date, technical characteristics, performance, Blackwell architecture, DLSS, ray tracing, new features and improvements, here's everything that we know about the next generation of GPUs from Nvidia.
The long-awaited Nvidia GeForce RTX 50 series of graphics cards is about to be revealed. These new GPUs are based on the new Blackwell architecture, which replaces Ada Lovelace from the previous generation. With this renewal of the range, we expect an increase in power and performance of course, but we also hope for better energy efficiency. As for the prices, they look particularly steep.
When will RTX 50 graphics cards be available?
At the end of 2022, Nvidia launched its first RTX 40 graphics cards. These were followed by the RTX 40 Super from the beginning of 2024. The Greens are not disrupting their schedule since the first GPUs of the RTX 50 generation are expected at the start of year 2025. We know that Nvidia has stopped production of certain RTX 40 GPUs in anticipation of the new models, which will be ready to take over as soon as the stock of RTX 40 is sold out.
According to the information we have available, the RTX 5080 and RTX 5090 will be revealed during CES 2025 in Las Vegas, which is being held from January 7 to 11, 2025. Their official release should therefore be scheduled for the end of January 2025. Nvidia always launches its most high-end references first. You generally have to wait a few weeks or months before seeing the more modest GPUs arrive. The RTX 5070 would however be presented during the month of January 2025, for marketing in February.
For the RTX 5060, we currently have no reliable indication, and for good reason, it may not be made available until much later, in spring 2025. The existence of a desktop RTX 5050 is uncertain, but a version for laptop PC should see the light of day, perhaps around June 2025.
How much do RTX 50 graphics cards cost?
Until Nvidia communicates the price of the RTX 50, we can only rely on guesses and leaks. And again, even when Nvidia announces the prices, we know that Founders Edition graphics cards are rare, and that we often have to turn to models customized by third-party manufacturers like Gigabyte, Asus, PNY or MSI to equip ourselves , which often charge higher prices than Nvidia.
On premium GPUs, you have to prepare for a significant price increase. An indiscretion suggests that the RTX 5090 could be sold from between 1,899 and 1,999 US dollars, while the RTX 4090 was marketed at 1,599 US dollars. The RTX 4090 being sold for 1,949 euros in France, it is possible that the RTX 5090 will approach 2,500 euros here if these prices are confirmed.
It is also suspected that the RTX 5080 would also be more expensive than the RTX 4080. According to marketing materials sent by Nvidia to its customers, the RTX 5090 and RTX 5080 must be presented to the public as products “intended for professionals”. To justify a high price, the RTX 5080 would therefore not be sold with the image of a GPU for gamers. These would rather be oriented by Nvidia towards the RTX 5070 as a mid-range option, or towards previous generation models, such as the RTX 4080 or the RTX 4080 Super.
What is Blackwell?
The Blackwell architecture was announced at the end of 2023 to Nvidia investors and was made official in March 2024 to the general public. It was presented as the company's new GPU architecture for data centers (in particular for servers dedicated to machine learning and generative AI calculation, replacing the Hopper architecture. Surprise, Blackwell will also be the architecture used for consumer graphics cards, thus succeeding Ada Lovelace. On the other hand, we distinguish the GB100 chips for data centers. GB200 chips for RTX 50 graphics cards.
The Blackwell architecture, and therefore the GB200 GPU chips of the RTX 50, is based on TSMC's 4NP manufacturing process. We are therefore moving to a 4 nm node on this generation (compared to 5 nm on the RTX 40), while TSMC has already developed 3 nm engraving. Nvidia preferred to start with an improved version of an already well-mastered process than to start with a completely new technology. It is not known whether this choice was made for reasons of performance, reliability, costs, production capacity, or a combination of several of these factors.
What is the performance of the RTX 50s?
While waiting for the benchmarks and tests, we can get a first idea of the performance of the GeForce RTX 50 graphics cards based on their supposed technical characteristics and the various leaks. The leaker Moore’s Law Is Dead believes that the jump in performance between the RTX 40 and the RTX 50 is likely to be less significant than that experienced between the RTX 30 and the RTX 40. AMD having withdrawn from the ultra high-end GPU segment, Nvidia no longer has to wage a power race as intense as before.
According to RedGamingTechthe RTX 5080 and RTX 5070 could offer just under 30% higher performance than the RTX 4080 Super and RTX 4070 Super. The RTX 5090 would be almost 50% more powerful than the RTX 4090. On this generation, Nvidia would have focused in particular on performance in ray tracing and path tracing, doubling the performance of the GPUs in these two aspects. As a reminder, path tracing, which Nvidia renames Full RTX or Full Ray Tracing for its products, is an evolution of ray tracing, further improving the calculation of rays of light for even more realistic graphics rendering. For the moment, few games are compatible (Indiana Jones and the Ancient Circle is among the recent releases) and Full RTX impacts fluidity even more than RTX, it will be interesting to see if Nvidia manages to find a compromise balanced between performance and quality with Blackwell GPUs.
A way to offer path tracing without seeing too great a drop in frames per second could be through an improvement in DLSS. This image scaling technology makes it possible to calculate a rendering in a definition lower than that intended, then to complete the image with the missing pixels by artificial intelligence in order to reconstruct it in the desired quality. This technique significantly boosts performance, with almost no deterioration in quality. DLSS 4 could land on the RTX 50 to follow DLSS 3.5, but we don't know much about it at the moment.
What are the technical characteristics of the RTX 50?
One of the big new features brought by the RTX 50 must be the transition to GDDR7 memory, while the RTX 40 remained in GDDR6 or GDDR6X. Pushed to its limits, GDDR7 can theoretically reach speeds of 36 Gb/s, but it is unlikely that Nvidia will go that far. The RTX 5090 could embark 32 GB of GDDR7 VRAM at 28 Gb/s for a 512-bit buswhich would result in an exceptional bandwidth of approximately 1.8 TB/s. Clearly, this would be a product which is no longer intended for video games at all, but for professional uses requiring more resources.
The RTX 5080 would benefit from 16 GB of GDDR7 VRAM and a 256-bit bus. To partially compensate for these less generous characteristics than on the RTX 5090, Nvidia could integrate a slightly faster memory (32 Gb/s) on the RTX 5080, allowing it to be granted a bandwidth around 1 TB/s. It is rumored that the RTX 5080 could be available in two versions, as Nvidia has already done in the past, one of them benefiting from 24 GB of VRAM instead of 16 GB. For the RTX 5070, we are talking about 12 GB of VRAM at 28 Gb/s for a 192-bit bus.
The GB202 GPU of the RTX 5090 would have 170 SM (Streaming Multiprocessors), 21,760 CUDA cores, 680 Tensor cores and 170 RT (ray tracing) cores. The TGP of the graphics card would reach 600 W, which is much higher consumption than that of the RTX 4090 and its 450 W thermal envelope.
The RTX 5080 would be equipped with the GB203 GPU, which should offer 84 SM, 10,752 CUDA cores, 336 Tensor cores and 84 RT cores, for a TGP around 400 W. The RTX 5070 and its GB205 GPU would benefit from 50 SM, 6 400 CUDA cores, 200 Tensor cores and 50 RT cores, with a TGP of 225 W.
The clock frequency of these GPUs could be around 2.5 GHz base, reaching the boost speed of previous models. In boost, the RTX 5090 could target 2.9 GHzand even beyond 3 GHz on overclocked versions customized by third-party manufacturers.
What connections for the RTX 50?
Goodbye PCIe 4, the RTX 50 are compatible with latest PCIe 5 hardware interface standardwith 16-pin connector. Don't panic if your motherboard is not very recent, backward compatibility with the PCIe 4 standard is assured. The RTX 5090 could be better equipped with pins than its counterparts if the energy consumption of 600 W were to be confirmed.
To connect display devices to your machine, the RTX 50 will of course offer HDMI 2.1 connectivity. Multiple DisplayPort ports will also be available, and Nvidia will likely upgrade to DisplayPort 2.1 this generation. The RTX 40s are stuck at DisplayPort 1.4, while AMD's Radeon RX 7000s were entitled to the famous DisplayPort 2.1. The latter grants the possibility of transmitting a stream 4K UHD at 480 Hz and 8K at 165 Hz with DP80 certified cables, which support UHBR20 (Ultra-High Bit Rate) at 80 Gb/s bandwidth.
