Skip to main content

RTX 5090 and RTX 5080 Final Specs Revealed at CES 2025

·570 words·3 mins
NVIDIA GeForce RTX 5090 GeForce RTX 5080 Blackwell CES 2025 GPU
Table of Contents

CES 2025 has officially wrapped up, and NVIDIA CEO Jensen Huang has confirmed the long-rumored details of the Blackwell consumer GPU lineup. As expected, the GeForce RTX 5090 claims the crown as the most powerful consumer graphics card ever released, while the RTX 5080 introduces next-generation GDDR7 memory to the high-end segment—albeit with a historically large gap between the two tiers.


🚀 RTX 5090: NVIDIA’s First True 8K Gaming Flagship
#

Built around the massive GB202 die, the RTX 5090 represents an unprecedented scale-up in both silicon size and capability.

  • Transistors: ~92 billion
  • CUDA Cores: 21,760, a dramatic increase over the RTX 4090’s 16,384
  • Memory: 32GB GDDR7
  • Memory Interface: 512-bit
  • Bandwidth: Nearly 1.8 TB/s, setting a new consumer GPU record
  • Total Graphics Power (TGP): 575W
  • Power Requirements: High-quality 1000W+ ATX 3.1 PSU strongly recommended

RTX 5090 Spec

Pricing and Availability
#

  • MSRP: $1,999 USD
  • China Market (FE estimate): ~14,999–15,999 RMB
  • Premium AIB Models: May exceed 18,999 RMB
  • Launch Date: January 30, 2025

RTX 5090 Price

In practical terms, RTX 5090 is positioned not merely as a faster 4090 replacement, but as a card designed explicitly for 8K gaming, heavy ray tracing, and AI-accelerated workloads without compromise.


⚡ RTX 5080: High-End, Strategically Limited
#

The RTX 5080 is based on the smaller GB203 die and targets a very different performance envelope. While still extremely powerful, it marks the widest performance separation NVIDIA has ever created between its flagship and second-tier GeForce models.

  • CUDA Cores: 10,752
  • Memory: 16GB GDDR7
  • Memory Speed: 32 Gbps
  • Memory Interface: 256-bit
  • TGP: 360W

RTX 5080

Pricing and Availability
#

  • MSRP: $999 USD
  • Expected China Pricing: Starting around 9,999 RMB
  • Launch Date: January 21, 2025

Despite being the fastest 16GB card on the market, the RTX 5080 reportedly trails the RTX 5090 by approximately 45–55% in raster, ray tracing, and AI workloads—an unusually large generational stratification.


📊 Blackwell Core Specification Comparison
#

Feature GeForce RTX 5080 GeForce RTX 5090
Architecture Blackwell (GB203) Blackwell (GB202)
CUDA Cores 10,752 21,760
VRAM 16GB GDDR7 32GB GDDR7
Memory Bus 256-bit 512-bit
AI Performance ~1,800 TOPS ~3,352 TOPS
MSRP $999 $1,999

This table makes NVIDIA’s positioning strategy clear: the RTX 5090 stands alone at the top, while the RTX 5080 anchors the “accessible” high-end.


🌏 Special Editions and Collector Releases
#

RTX 5090D (China-Specific Model)
#

To comply with export and compute regulations:

  • Retains 32GB GDDR7
  • Adjusted CUDA core counts and clocks
  • Designed to stay within the TPP 4800 compute threshold
  • Launches alongside the global RTX 5090 in late January / early February 2025

Golden Signature Series
#

To commemorate the Blackwell launch, NVIDIA revealed a limited collector initiative:

  • Five historically significant GPUs
    • GeForce 256
    • GeForce 8800 Ultra
    • GeForce GTX 1080 Ti
    • GeForce RTX 2080 Ti
    • GeForce RTX 3080
  • Each unit personally signed in gold by Jensen Huang

🧠 New Blackwell-Exclusive Software Features
#

The RTX 50 series is not just about raw hardware—software innovation was a major CES highlight.

DLSS 4: Multi-Frame Generation
#

  • Generates multiple intermediate frames per rendered frame
  • Leverages new Blackwell AI hardware
  • NVIDIA claims up to 8× performance gains in supported titles

RTX Neural Faces
#

  • Uses generative AI to synthesize realistic facial skin, hair, and micro-detail
  • Runs in real time
  • Significantly reduces uncanny valley artifacts in human characters

Taken together, the RTX 5090 and RTX 5080 showcase NVIDIA’s clearest generational split yet: a no-compromise flagship built for 8K and AI-heavy workloads, and a deliberately constrained—but still formidable—high-end option designed to anchor the broader Blackwell ecosystem.

Related

NVIDIA RTX 5070 vs 5070 Ti: Blackwell Mid-Range Explained
·491 words·3 mins
NVIDIA RTX 5070 Blackwell GPU DLSS 4
Introduction to NVIDIA CUDA
·793 words·4 mins
CUDA NVIDIA GPU Parallel Computing
Why CUDA Powers Modern Deep Learning
·385 words·2 mins
CUDA Deep Learning GPU AI NVIDIA