The manufacturer has already presented this Blackwell architecture during its GTC in March 2024, with its B200 GPU for artificial intelligence in data centers. The RTX 50 series uses a Blackwell variant, including 92 billion transistors compared to 76 billion for the RTX 40 GPU and 2x 104 billion for the B200. The latter, as a reminder, uses two dies on the same chip.
Tensor, Ray Tracing, PCIe, GDDR… : les RTX 50 font « + 1 »
In any case, four RTX 50 models are announced to start: the RTX 5070, 5070 Ti, 5080 and 5090. Broadly speaking, the generations of the different technologies are “+1” compared to the RTX 40 series. .
The manufacturer offers a summary table on this page, with the same for RTXs of previous generations.
Depending on the model, the number of CUDA Cores varies between 6,144 and 21,760, with a frequency in Boost mode from 2.41 to 2.62 GHz. NVIDIA insists on the performance of GDDR7 capable of reaching up to 1,792 GB/s on the RTX 5090, compared to 1,008 GB/s on the RTX 4090 in GDDR6x. The DisplayPort 2.1b UHBR20 video output allows for 8K at 165 Hz.
The new RTX 50s thus use 4th generation Ray Tracing Cores, 5th generation Tensor Cores for AI (with FP4 support), PCIe 5.0, a 9th generation NVENC encoder, a 6th generation NVDEC decoder , DLSS 4, GDDR7, etc. On the RTX 40 we had 3rd generation RT Cores, 4th generation Tensor Cores, PCIe 4.0, etc.
Regarding 4th generation Ray Tracing Cores, NVIDIA announces “ twice the intersection rate between a ray and a triangle compared to the previous generation and improved compression to reduce memory footprint ».
Here is DLSS 4 with multi-image generation and Reflex 2
A dedicated blog post looks at the operation and performance of DLSS 4. According to NVIDIA, this version represents “ the biggest upgrade to its AI models since the release of DLSS 2.0 in 2020 “. Multi-image generation with DLSS (an RTX 50 exclusive) “ accelerates frame rate by using AI to generate up to three frames per frame for display “, explains NVIDIA
As an example, NVIDIA claims that on the GeForce RTX 5090, “ DLSS 4 with Multi Frame Generation [génération multi-images dans la langue de Molière, ndlr] increases performance by more than 8X compared to traditional rendering in this Cyberpunk 2077 scene ».
NVIDIA is taking the opportunity to present Reflex 2 to reduce latency by up to 75%, compared to 50% with first generation Reflex, thanks to Frame Warp technology, “ which further reduces latency based on last mouse click ».
This feature will arrive soon, but will not be exclusive to the RTX 50, it will also be found on the RTX 40, RTX 30 and RTX 20.
Power requirement 650 to 1000 watts
GPU consumption (Total Graphics Power) is 575 watts for the RTX 5090, 360 watts for the RTX 5080, 300 watts for the RTX 5070 Ti and finally 250 watts for the RTX 5070.
On the power side, 1,000 watts of power will be required for the RTX 5090 according to NVIDIA. We go to 850 watts for the RTX 5080, then 750 watts for the RTX 5070 Ti and finally 650 watts for the basic RTX 5070.
Plan for four 8-pin PCIe power cables or one 600-watt PCIe Gen 5 cable for the RTX 5090, versus two 8-pin cables or one 300-watt PCIe Gen 5 for the RTX 5070 Ti).
RTX 50 vs RTX 40 : les performances font « x2 »
NVIDIA announces that its GeForce RTX 5090 is “ twice as fast as the GeForce RTX 4090 », a promise that we also find on the RTX 5080 compared to the RTX 4080, as well as the RTX 5070 (Ti) vs the RTX 4070 (Ti).
The manufacturer offers numerous graphs on in-house tests here. A quick glance shows that we should not take these doubled performances as a formal promise but rather as the optimal scenario.
The RTX 5080 and 5090 will be available on January 30
Founders Edition (NVIDIA’s in-house cards) will be offered for all versions, except the RTX 5070 Ti. The RTX 5090 is announced at 1,999 dollars or 2,349 euros with availability until January 30, while the RTX 4090 was at 1,599 dollars (1,949 euros) at its launch.
The Founders Edition version of the RTX 5090 measures 304 x 137 mm and occupies two PCIe slots (compared to three for the RTX 4090). She is “SFF-Ready”, SFF for Small Form Factor. This certification was launched at Computex last year and ensures compatibility between graphics cards and SFF boxes.
The RTX 5080 will be sold for 999 dollars or 1,179 euros, also starting January 30. The dimensions of the Founders Edition are the same as those of the RTX 5090, therefore with SFF certification.
You will have to wait until February for the RTX 5070 and 5070 Ti, which will respectively be offered from 549 and 749 dollars, or 649 and 884 euros according to the French NVIDIA site. The RTX 5070 Founder Edition measures 242 x 112mm and also takes up two slots.
Of course, NVIDIA’s many partners will not hesitate to offer in-house models with more or less extravagant coolers and more or less marked overclocking. The waltz of announcements has already started. TechPowerUp includes manufacturer press releases on this subject.
Four RTX 50s for laptops in March
On laptops, NVIDIA is also announcing RTX 50, and explains that its Max-Q technology benefits from new features such as Advanced Power Gating, Low Latency Sleep and Accelerated Frequency Switching which “ increase battery life by up to 40% compared to previous generation ».
We find an RTX 5090 with 10,496 CUDA Cores and a computing power of 1,824 TOPS for AI, with 24 GB of GDDR7. With the RTX 5080 we go to 7,680 CUDA cores, 1,334 TOPS and 16 GB of GDDR7 respectively.
Two GeForce RTX 5070 also complete the range. The RTX 5070 Ti with 5,888 CUDA cores, 992 TOPS, and 12 GB of GDDR7 as well as the RTX 5070 with 4,608 CUDA cores, 798 TOPS and 8 GB of GDDR7. You will need to pay attention to the frequencies which vary depending on the laptop models and the performance of its cooling system.
All RTX 50 series mobile chips will be available in March.
