NVIDIA RTX 5090
Specs, Pinout & AI Analysis
The NVIDIA GeForce RTX 5090 is built on the Blackwell GB202 architecture (TSMC 4NP), with 21,760 CUDA cores, 32 GB GDDR7 on a 512-bit bus, and 1,792 GB/s of memory bandwidth. For AI workloads, 5th-generation Tensor Cores deliver 3,352 TOPS at FP4 — the highest of any consumer GPU. Power delivery uses a single 16-pin 12V-2x6 connector capable of up to 600W.
Blackwell GB202
32 GB GDDR7
3,352 TOPS FP4
575W TDP
16-pin 12V-2x6
PCIe 5.0 ×16
Key Specifications
Architecture Blackwell (GB202)
Process Node TSMC 4NP
CUDA Cores 21,760
Tensor Cores 680 × 5th-gen
RT Cores 170 × 5th-gen
AI Performance 3,352 TOPS (FP4)
VRAM 32 GB GDDR7
Memory Bus 512-bit
Memory Bandwidth 1,792 GB/s
Base / Boost Clock 2,017 / 2,407 MHz
TDP 575 W
Power Connector 16-pin 12V-2x6
PCIe Interface PCIe 5.0 ×16
Display Outputs 3× DP 2.1 + 1× HDMI 2.1
DLSS DLSS 4 + Multi Frame Gen
Launch Price $1,999 USD
16-Pin 12V-2x6 Connector Pinout — Interactive
Click any pin to see its function and specification.
12V Power
Ground
Sense / Sideband
Reserved
Analyse This GPU With Our Tools
Tool 1
Bottleneck & Synergy Calculator
See which CPUs pair best with the RTX 5090 at 1080p, 1440p, and 4K.
RTX 5090 pre-selected →
Tool 2
AI Readiness & NPU Checker
Check which local LLMs and AI models the RTX 5090 can run and at what speed.
RTX 5090 pre-selected →
Tool 3
Chip Anatomy Explorer
Visualise the Blackwell GB202 die topology and compare it side-by-side with other chips.
Blackwell pre-selected →
Tool 4
GPU PSU & Thermal Sanity Check
Calculate minimum PSU wattage, connector safety, and estimated GPU temps for the RTX 5090.
RTX 5090 pre-selected →
