NVIDIA has officially revealed key details about its upcoming GeForce RTX 50 series GPUs, including performance gains in native Ray Tracing (RT) scenarios, pricing, and a deep dive into the new Blackwell architecture. The announcements were made at CES 2025 during their Editor's Day.
RTX 50 Performance Uplifts (Native, No DLSS)
NVIDIA has confirmed performance uplifts when using native ray tracing without DLSS:
- RTX 5090: 30% faster than RTX 4090
- RTX 5080: 15% faster than RTX 4080
- RTX 5070 Ti: 20% faster than RTX 4070 Ti
- RTX 5070: 20% faster than RTX 4070
RTX 50 Series Pricing and Availability
The launch dates and pricing for the new cards are:
| GPU | Release Date | MSRP (USD) |
|---|---|---|
| RTX 5090 | January 30th | $1999 |
| RTX 5080 | January 30th | $999 |
| RTX 5070 Ti | February | $749 |
| RTX 5070 | February | $549 |
Blackwell Architecture Highlights
The NVIDIA Blackwell architecture is designed for enhanced AI capabilities and graphical performance. Key features include:
- Based on TSMC's 4nm process with up to 92 billion transistors.
- 5th Gen Tensor Cores with high-speed FP4 compute and up to 4000 AI TOPS.
- 4th Gen RT Cores designed for Mega Geometry and with up to 360 RT TFLOPs.
- GDDR7 memory interface with up to 1.8 TB/s bandwidth.
- New Blackwell SM with 125 TFLOPS peak FP32 compute
- DisplayPort 2.1 (UHBR20), PCIe Gen5 support, and 4K NVDEC/NVENC.
- Double INT32 GPU throughput.
- Improved SER (Shader Execution Reordering).
- New programmable coprocessor known as Amp.
- New power gating modes for better efficiency.
- Improved clock frequency switching capabilities.
DLSS 4: A New Era of AI Upscaling
DLSS 4 introduces a new transformer engine with 2x the parameter size and 4x the compute horsepower. It also introduces Multi-Frame Generation (MFG), running five models per frame instead of two. This generates 15 out of 16 pixels or frames using AI while improving image quality.
DLSS 4 will be available with Day-0 support for 75 games and apps.
Reflex 2: Improving Latency
Reflex 2 introduces Frame Warp technology, which improves system responsiveness by 75% by sampling the mouse position right before the frame is rendered.
RTX AI for Gamers: Neural Shading and Beyond
NVIDIA is shifting towards Neural Shading technologies. Blackwell aims to replace traditional programmable shaders using AI to achieve similar graphical output. This approach includes:
- Neural Textures, Neural Materials, and Neural Volumes to enhance efficiency and optimization.
- RTX Neural Radiance Cache (NRC) for real-time training and indirect lighting.
- RTX Skin for lifelike skin rendering through subsurface scattering.
- RTX Neural Faces for enhancing the photorealism of NPCs.
- RTX Hair and Fur for more efficient rendering of detailed hair and fur.
- RTX Mega Geometry for uncompromised high-fidelity scenes, with efficient compression techniques.
RTX 50 Founders Edition Cooler
The new Founders Edition cooler is called a "Double Flow Through" design and uses:
- A 3D Vapor Chamber that sits over the compact PCB.
- Five heat pipes connected to three separate heatsinks.
- Dual axial fans.
This design reduces the noise output and can maintain around 30-35 dBA with a load of up to 600W.
Source: NVIDIA Official Website
