NVIDIA® RTX™ 50: expectations and reality
The highlight of CES 2025 was NVIDIA® CEO Jensen Huang’s speech. The revelation of new GPU specifications within minutes caught many off guard. In this article, we’ll examine how expert predictions matched the actual announcements.
Let’s look at the lineup first. The RTX™ 40 series launched with 6 models, ranging from the RTX™ 4060 to the RTX™ 4090. While many expected a similar range for the RTX™ 50 series, that didn’t happen. Instead, the RTX™ 50 family includes just 4 models: RTX™ 5070, RTX™ 5070 Ti, RTX™ 5080, and RTX™ 5090. We may see both the RTX™ 5050 and RTX™ 5060 in the future, but no official sources have verified these graphics cards yet.
Technological process
Moore’s law, the empirical observation that “the number of transistors in an integrated circuit doubles about every two years”, is often said to be no longer relevant to chip performance. Since 2022, Jensen Huang has repeatedly declared Moore’s law dead. Instead he proposed a new concept that emphasizes the simultaneous development of architecture, microchips, software libraries and algorithms.
Together, this shift allows us to focus on overall system performance rather than just transistors count. The concept of computing efficiency has sparked ongoing discussions in the tech community. While views on this topic vary, the industry clearly faces both physical and economic barriers to further miniaturization.
Let’s take a look at the new generation GPU process technology. The presentation didn’t specifically mention this, but all previous generation cards were built on the 4N process. The RTX™ 50 series uses a different 4NP process technology. At the same time, it’s important to understand that 4N and 4NP are just marketing names. The transistors themselves remain 5 nm in size.
The improved 4NP process technology primarily enables higher transistor density on the chip and faster clock speeds. While experts predicted that the RTX™ 50 would use the same process technology as the RTX™ 40, they were technically incorrect, though not by much, since the transistor size remains unchanged and TSMC continues as the manufacturer.
Number of cores
Prior to the RTX™ 50 series release, numerous data leaks revealed the GPU’s basic characteristics. Initial insider reports from July 2024 suggested the flagship would feature 24,576 cores, 192 Ray-tracing cores, and 768 Tensor cores. However, subsequent leaks adjusted these numbers to more realistic values.
The final RTX™ 5090 shipped with 21,760 CUDA® cores (up from the RTX™ 4090’s 16,384), 170 Ray-tracing cores, and 680 Tensor cores. This aligns with the company’s recent strategy of boosting performance not just through increased transistor count, but through comprehensive architectural optimization.
Memory
The new GPUs use of GDDR7 memory came as no surprise. Industry experts had predicted this move in 2024 after the three major manufacturers (Samsung, Micron, and SK hynix) showcased their GDDR7 prototypes in succession. NVIDIA® was generous with memory distribution: the base RTX™ 5070 model features 12 GB GDDR7 on a 192-bit bus, while the RTX™ 5070 Ti and RTX™ 5080 both carry 16 GB GDDR7 on a 256-bit bus. At the top end, the flagship RTX™ 5090 comes with a massive 32 GB GDDR7 on a 512-bit bus.
Experts initially predicted that the maximum throughput of this memory configuration would be 1,5 Tbps. However, reality surpassed these expectations, achieving a throughput of 1,7 Tbps. This dramatic improvement primarily benefits the GPU’s AI processing capabilities rather than gaming performance. The new generation’s combination of high capacity and fast memory is particularly valuable for large language models and generative neural networks.
Technologies
For gamers
Real-time ray tracing has become one of the most revolutionary GPU technologies, marking the beginning of the RTX™ line. For many consumers, this feature has been a key factor in their purchase decisions. In RTX™ 50 series cards, DLSS (Deep Learning Super Sampling) version 4 may play an equally important role. This technology significantly boosts GPU performance in games through its hybrid frame rendering approach.
With DLSS enabled, instead of rendering every frame conventionally, some frames are generated in real time using AI. While early versions of this technology could only upscale frames to higher resolutions, DLSS 3 introduced a more advanced capability: for every conventionally rendered frame, it can generate an additional AI-created frame.
DLSS 4 will generate three AI-powered frames for every traditionally rendered frame. This significantly increases the frame per second (FPS) without putting heavy load on the GPU. The AI analyzes object and scene movement to ensure the generated frames closely match conventionally rendered ones.
This raises an important question: how do we handle input lag? Since frame generation takes time, each iteration adds to the response time. A smooth picture with slow response to player actions can severely impact the gaming experience. To address this, NVIDIA® has improved their Reflex 2 technology alongside DLSS to minimize latency.
Specifically, Frame Warp was integrated into the system. This technology reduces game latency by updating rendered frames with the latest mouse input just before display. It enhances both multiplayer competition and single-player responsiveness.
For content creators
The RTX™ 50 series isn’t just for gaming. Video content creators will find significant value in these new GPUs. The flagship RTX™ 5090 model comes equipped with 3 encoders and 2 decoders, compared to the RTX™ 4090’s 2 encoders and 1 decoder. These components have been enhanced through collaborative development with industry leaders: Adobe, Blackmagic Design, ByteDance, and Wondershare. As a result, the RTX™ 5090 renders video 60% faster than the RTX™ 4090 and four times faster than the RTX™ 3090.
Beyond raw speed improvements, the quality has also been enhanced. The 9th generation NVENC encoder delivers 5% better quality in HEVC and AV1 tasks. The AV1 Ultra Quality mode achieves better data compression while maintaining image quality, reducing file sizes by 5%. This means faster video rendering on the RTX™ 5090, and decreasing the time between editing and production.
Conclusion
Looking back six months, the experts’ predictions and expectations proved overly optimistic. As the release date approached, it became evident that the new GPUs would offer more than just additional computing units. The key innovation would be new optimization and AI technologies enhancing existing frame rendering systems.
At CES 2025, during the GPU 50 series presentation, a new AI era was unveiled. This vision portrayed a world where digital assistants and robots handle complex tasks. At its core would be an ecosystem combining supercomputers for AI training, affordable inference accelerators for consumer devices, and versatile software operating both locally and in the cloud. While the full extent of this future remains uncertain, one thing is clear - we stand at the threshold of turning science fiction into reality.
LeaderGPU remains committed to providing reliable access to these cutting-edge technologies. Order your first GPU server today and begin transforming your ideas into reality.
See also:
Updated: 12.08.2025
Published: 23.01.2025
