← Back to GPUs
NVIDIA · RTX 30
NVIDIA GeForce RTX 3090
$900$1499 MSRP
The NVIDIA GeForce RTX 3090 was the previous-generation flagship with 24GB of GDDR6X memory. In 2026, it remains one of the best used-market options for AI builders — 24GB VRAM with full CUDA support at used prices well below a new RTX 4090. It runs 32B models at Q4 and handles Stable Diffusion easily. The older Ampere architecture means no DLSS 3/4, but for AI inference, raw VRAM matters more than architecture.
Best ForBest used-market value for 24GB VRAM AI builds
VerdictThe budget AI builder's hero on the used market — 24GB CUDA for under .
AI
7/10
Gaming
7/10
Specifications
VRAM24GB GDDR6X
Memory Bandwidth936 GB/s
CUDA Cores10,496
Boost Clock1695 MHz
TDP350W
Power Connector2x 8-pin
Length313mm
Form FactorTriple Slot
Release Year2020
AI Capabilities
Sweet Spot24GB VRAM
The professional standard. Handles most models with smart quantization.
Can run (Q4 quantized)
Llama 3.1 8BQwen 2.5 32BQwen 2.5 14BMistral 7BFLUX.1 DevStable Diffusion XLStable Diffusion 3.5 LargeHunyuanVideoCogVideoX-5BMochi 1LTX VideoStable Video DiffusionWan Video 14BCodestral 22BQwen 2.5 Coder 32BLLaVA 1.6 34BAlphaFold 2ESMFold (ESM-2 15B)ESM-2 3BscGPTRFdiffusionFine-tune Llama 8BTrain SDXL LoRATrain FLUX LoRA
Recommended system RAM for AI: 48GB+ (2x GPU VRAM for model overflow)
Performance Estimates
Estimated tokens/sec for LLM inference based on 936 GB/s memory bandwidth — not hardware benchmarks. Methodology · What is Q4/Q8?
Llama 3.1 8B8B
FP16~29-35 tok/sUsableQwen 2.5 32B32B
Q4~25-31 tok/sUsableQwen 2.5 14B14B
Q8~35-43 tok/sFastMistral 7B7B
FP16~33-40 tok/sFastCodestral 22B22B
Q8~22-27 tok/sUsableQwen 2.5 Coder 32B32B
Q4~25-31 tok/sUsablePros
- +24GB VRAM
- +Great used value for AI
- +Widely available used
Cons
- -Old architecture
- -High power draw
- -No DLSS 3/4
aigaming
Where to Buy
Will It Run?
Llama 3.1 8B8B
FP16Qwen 2.5 32B32B
Q4Qwen 2.5 14B14B
Q8Mistral 7B7B
FP16FLUX.1 Dev12B
Q8Stable Diffusion XL6.6B
FP16Stable Diffusion 3.5 Large8B
FP16HunyuanVideo13B
Q8