The RTX 5090 is NVIDIA's statement product — 32GB GDDR7, 575W TGP, and a $1,999 MSRP that routinely sells for $2,500-3,000 at street prices due to acute scarcity. The RTX 4090 was the previous statement product and can still be found used for $900-1,100. The performance gap is real. The value gap is absurd.

Specs Comparison

Architecture: GB202 at Full Specification

The RTX 5090 uses the full GB202 Blackwell die — 21,760 CUDA cores across 170 SMs. This is the largest consumer GPU NVIDIA has ever shipped. The RTX 4090 used the full AD102 die with 16,384 CUDA cores. Raw CUDA core advantage goes to the 5090 by 33%.

Memory configuration: 32GB GDDR7 on a 512-bit bus delivering approximately 1,792 GB/s. The RTX 4090's 24GB GDDR6X on a 384-bit bus achieves approximately 1,008 GB/s. The 5090 has 78% more memory bandwidth — a generational leap rather than an incremental improvement.

NVIDIA rates the 5090 at approximately 3.352 petaFLOPS of FP32 performance versus the 4090's approximately 82.6 TFLOPS. The jump is transformative in any workload that saturates compute and memory bandwidth simultaneously.

Gaming Performance at 4K and 8K

At 4K Ultra in Cyberpunk 2077 with path tracing, the RTX 5090 averages approximately 65-70 fps native versus the 4090's 45-50 fps — a 40% improvement. Enable DLSS 4 on the 5090 and Multi Frame Generation brings the displayed output above 120 fps, making path-traced Cyberpunk genuinely playable at 4K without compromise.

In Alan Wake 2 at 4K with full ray tracing, the 5090 leads by 35-45% in native rendering. The 4090 with DLSS 3 achieves similar playable frame rates in this title; the 5090 achieves them with higher image quality via DLSS 4's transformer model upscaling.

The 5090 is the first consumer GPU that makes 8K gaming with DLSS upscaling from 4K a plausible setup for single-GPU configurations. In Hellblade 2 and other current titles at 8K DLSS Quality mode, it delivers playable performance above 60 fps for the first time on a single card.

The Scarcity Premium Problem

The RTX 5090 launched at $1,999 MSRP in January 2026. Within the first month, AIB partner cards were selling for $2,500-3,200 at major retailers — Newegg, Micro Center, and Amazon all showed substantial premiums. By May 2026, supply has improved but premium models (MSI Suprim Liquid X, ASUS ROG Strix) still trade $400-600 above MSRP.

This scarcity premium is not manufacturer-imposed — NVIDIA's MSRP is $1,999. The premium is market-driven, driven by low GB202 die yields and high global demand. It is not clear when RTX 5090 cards will be consistently available at $1,999.

Any buyer spending $2,400+ on a 5090 today should recognize they're paying a scarcity tax. The same $2,400 buys a used RTX 4090 plus an RTX 5080 for dual-GPU workstation use, or an RTX 5080 with $1,400 remaining for other system components.

Who Should Actually Buy This

The legitimate buyer profiles for an RTX 5090 are narrow: a professional who runs both gaming and large AI/ML inference workloads and needs 32GB VRAM in a single card; a content creator who exports 8K video and plays games on the same machine; a simulation researcher who needs maximum single-precision FP32 throughput.

The buyer profile that should not buy an RTX 5090 is: anyone who plays games at 1440p or 4K and owns any card from the RTX 3080 generation or newer. The RTX 5080 at $999 captures over 75% of the 5090's gaming performance for half the price.

The RTX 4090 at $900-1,100 used is still a fundamentally excellent GPU for gaming. It handles 4K Ultra in every current game at playable frame rates with DLSS 3. The 5090 is faster — sometimes by 40% — but that speed costs $1,500-2,000 more. That math doesn't work for gaming alone.

Best For

• NVIDIA GeForce RTX 5090 Professionals running large AI inference workloads who also game, or buyers with truly unlimited budgets
• NVIDIA GeForce RTX 4090 Any gamer who wants 4K Ultra performance without paying a scarcity premium

FAQ