Skip to main content

NVIDIA RTX 4090 (Ada Lovelace)

Overview​

The NVIDIA GeForce RTX 4090, released in 2022, is the flagship consumer GPU and has become the de facto standard for consumer AI workloads thanks to its 24GB GDDR6X memory and 2,642 TOPS INT8 compute. It offers exceptional price-performance for local LLM inference, Stable Diffusion, LoRA fine-tuning, and similar workloads.

Core Specifications​

ItemSpec
ArchitectureAda Lovelace AD102
ProcessTSMC 4N
Memory24 GB GDDR6X
Memory Bandwidth1,008 GB/s
CUDA Cores16,384
Tensor Cores568 (4th Gen)
FP3282.6 TFLOPS
FP16/BF16 Tensor Core1,321 TFLOPS (sparse)
FP8 Tensor Core2,642 TFLOPS (sparse)
INT8 Tensor Core2,642 TOPS
TDP450 W
PCIeGen 4 x16

Vendor Information​

ItemDetail
ManufacturerNVIDIA Corporation
Websitehttps://www.nvidia.com
Product Pagehttps://www.nvidia.com/en-us/geforce/graphics-cards/40-series/rtx-4090/
ReleaseSeptember 2022
Price$1,599 MSRP (market price often higher)

Use Cases​

  • Local LLM inference (up to 70B)
  • Stable Diffusion / ComfyUI
  • LoRA fine-tuning
  • 3D rendering (gaming, professional applications)

Considerations​

  • Lacks data center features (no ECC memory, no MIG, no vGPU)
  • 24GB memory limits large models (>30B is challenging)
  • High power consumption, requires 850W+ PSU