NVIDIA’s RTX 5090 is shaping up to be the most powerful consumer GPU ever made. Based on the Blackwell architecture, the flagship card promises a massive leap in rasterization performance, ray tracing, and AI-driven features. Here’s what we know so far.
Blackwell Architecture: What’s New
The RTX 5090 uses the GB202 die, NVIDIA’s largest consumer Blackwell chip. Key architectural improvements over Ada Lovelace (RTX 4090):
- 4nm TSMC process — A node shrink from Ada’s 4nm, but with a redesigned transistor layout that improves power efficiency by 25%
- 32GB GDDR7 — Up from 24GB GDDR6X. Bandwidth reaches 1.8 TB/s, a 52% increase over the 4090
- 24,576 CUDA cores — Up from 16,384 in the 4090. A 50% core count increase
- Fourth-gen RT cores — 2.8x ray tracing throughput improvement
- Fifth-gen Tensor cores — FP4 compute support for next-gen AI workloads
- DLSS 4 — A new multi-frame generation system that can generate up to 3 interpolated frames per rendered frame
Expected Performance
Based on leaked benchmarks and architectural analysis, the RTX 5090 is expected to deliver:
- 4K gaming: 60-80% faster than RTX 4090 in rasterization, 90-120% faster with ray tracing enabled
- 1440p: 50-70% faster than 4090 — though CPU bottlenecks become a factor
- 8K gaming: Finally viable with DLSS 4. The 5090 can reportedly maintain 60fps at 8K with DLSS Performance + Frame Generation
- AI inference: 3x faster than 4090 for LLM inference, thanks to FP4 support and 32GB VRAM
Power and Cooling
The RTX 5090 is a power-hungry beast:
- TGP: 600W — Up from 450W on the 4090. NVIDIA recommends a 1200W PSU minimum
- 12V-2×6 power connector — An updated version of the 12VHPWR connector with better current handling
- Dual-slot cooler — Despite the power draw, the Founders Edition uses a dual-slot design with a vapor chamber and dual-axial fans
- PCIe 5.0 x16 — Full bandwidth for the new generation
Pricing and Availability
The RTX 5090 is expected to launch in January 2027 with an MSRP of $1,999 — a $200 increase over the 4090’s launch price. Availability is expected to be tight, with scalper prices potentially reaching $3,000+ at launch.
A more affordable RTX 5080 Ti with 16GB GDDR7 is rumored for Q2 2027 at $1,199.
Should You Wait?
If you’re gaming at 4K with an RTX 4090, the 5090 isn’t a necessity — the 4090 remains the second-fastest consumer GPU on the planet. But if you’re on a 30-series card or building a new high-end system, the 5090 will be the GPU to beat for years to come.
For AI researchers and content creators, the 32GB VRAM and FP4 support make the 5090 a no-brainer. It’s effectively a data center card in a consumer form factor.
NVIDIA’s Blackwell keynote is scheduled for CES 2027. Full specifications and benchmarks will be revealed then.