Skip to main content

NVIDIA RTX 5090 (Blackwell Consumer Flagship)

Product Overview

The NVIDIA RTX 5090, unveiled at CES 2025-01, is the consumer flagship bringing the Blackwell architecture to consumer GPUs for the first time. With 32GB GDDR7 memory, 21,760 CUDA cores, and a 575W TDP, it delivers 3,352 TOPS of AI compute (FP4)2.5× that of the RTX 4090.

Positioned for local LLM inference (70B+ models), Stable Diffusion XL training, and consumer AI developers.

Core Specifications

ParameterValue
ArchitectureBlackwell (GB202)
Process NodeTSMC 4N (custom 5nm)
CUDA Cores21,760
Tensor Cores680 (5th Gen)
RT Cores170 (4th Gen)
Base Clock2.01 GHz
Boost Clock2.41 GHz
Memory32 GB GDDR7
Memory Bandwidth1,792 GB/s (28 Gbps × 512-bit)
FP32 Compute104.8 TFLOPS
FP16 Tensor419 TFLOPS (sparse)
FP8 Tensor838 TFLOPS (sparse)
FP4 Tensor3,352 TOPS (sparse)
INT8 Tensor1,676 TOPS
TDP575 W
Power Connector1× 16-pin (12V-2x6)
MSRP$1,999
Launch Date2025-01-30

RTX 5090 vs RTX 4090 Comparison

MetricRTX 5090RTX 4090Improvement
ArchitectureBlackwellAda LovelaceNew gen
CUDA Cores21,76016,3841.33×
Memory32GB GDDR724GB GDDR6X1.33×
Memory Bandwidth1,792 GB/s1,008 GB/s1.78×
FP16 Tensor419 TFLOPS165 TFLOPS2.5×
FP4 Tensor3,352 TOPSN/ANew
TDP575W450W1.28×
Price$1,999$1,5991.25×

Blackwell New Features

FP4 Precision Support

  • Native FP4 Tensor Cores (first time on consumer GPUs).
  • Reduces inference memory footprint by 50% (vs FP8).
  • 70B LLM can run FP4 quantized within 32GB memory (~40GB model compressed).

DLSS 4 Multi Frame Generation

  • Multi Frame Generation: Generates 3 frames from 1 (vs DLSS 3's 1 frame from 1).
  • Gaming-only, but showcases Blackwell's compute power.

GDDR7 Memory

  • 28 Gbps speed (vs GDDR6X 21 Gbps).
  • 1,792 GB/s bandwidth = 2× RTX 4090.
  • Alleviates the memory-bound bottleneck in LLM inference.

LLM Inference Performance

ModelQuantizationRTX 5090 (32GB)RTX 4090 (24GB)Improvement
Llama 3 8BFP16~95 tok/s~70 tok/s1.36×
Llama 3 70BFP4~28 tok/sOOMBreakthrough
Llama 3 70BINT4~22 tok/s~15 tok/s1.47×
Mixtral 8x7BINT4~45 tok/s~32 tok/s1.41×
Qwen 2.5 72BFP4~26 tok/sOOMBreakthrough

70B model FP4 quantized (~40GB) fully fits in VRAM — 32GB memory is the key enabler.

Vendor Information

ParameterValue
VendorNVIDIA Corporation
Product Pagehttps://www.nvidia.com/en-us/geforce/graphics-cards/50-series/rtx-5090/
MSRP$1,999 (FE Founders Edition)
Target MarketConsumer AI, creators, researchers, local LLM

Use Cases

  • Local 70B LLM inference (FP4 quantized, 32GB VRAM)
  • ✅ Stable Diffusion XL / Flux training and inference
  • ✅ Video production (DaVinci Resolve AI acceleration)
  • ✅ 8K gaming + frame generation
  • ❌ Data center (use H100/B200 instead)
  • ❌ Multi-node training (lacks NVLink)