How much VRAM does the NVIDIA RTX 5090 have?

The NVIDIA RTX 5090 has 32 GB of GDDR7 with 1792 GB/s memory bandwidth.

What is the NVIDIA RTX 5090 best for?

With 32 GB of VRAM, the NVIDIA RTX 5090 is well-suited for running 7B–32B models at Q4 with room for context, making it a great all-rounder for local LLM inference.

What LLMs can the NVIDIA RTX 5090 run locally?

The NVIDIA RTX 5090 can run 47 of the 70 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.3 70B Instruct at Q2_K, Llama 3.1 8B Instruct at BF16, Llama 3.2 3B Instruct at FP32.

Can the NVIDIA RTX 5090 run Llama 3.3 70B Instruct?

Yes. The NVIDIA RTX 5090 runs Llama 3.3 70B Instruct natively in VRAM at Q2_K quantization, achieving approximately 77.8 tokens per second.

Can the NVIDIA RTX 5090 run Qwen 3.6 27B?

Yes. The NVIDIA RTX 5090 runs Qwen 3.6 27B natively in VRAM at NVFP4 quantization, achieving approximately 132.7 tokens per second.

Can the NVIDIA RTX 5090 run Llama 3.1 8B Instruct?

Yes. The NVIDIA RTX 5090 runs Llama 3.1 8B Instruct natively in VRAM at BF16 quantization, achieving approximately 112 tokens per second.

NVIDIA RTX 5090

The NVIDIA RTX 5090 has 32 GB VRAM and 1792 GB/s memory bandwidth. It can run 47 of our 70 tracked models natively in VRAM at 8k context.

With 32 GB GDDR7, the NVIDIA RTX 5090 is a consumer-tier GPU that can run 47 models natively. It handles 70B-class models at Q4 quantization.

The NVIDIA RTX 5090 is the flagship Blackwell consumer GPU, featuring 32GB of GDDR7 memory on a 512-bit bus delivering 1,792 GB/s bandwidth — a 78% increase over the RTX 4090. With 21,760 CUDA cores and 680 5th-gen Tensor Cores, it handles 30B-class models at high quantization entirely in VRAM and accelerates FP4 inference for next-generation LLMs.

NVIDIA RTX 5090: February 2025 Blackwell architecture with 32GB GDDR7 at 1792 GB/s bandwidth — highest bandwidth consumer GPU ever.

Runs all models up to 32B at Q4 natively. 70B models need CPU offload. Tokens/sec ~15-25 for 7B, ~8-12 for 32B at Q4.

Full CUDA support out of box. CUDA 12.x and driver 525+ recommended. TensorRT-LLM and vLLM fully optimized.

Vendor	NVIDIA
Architecture	Blackwell
VRAM	32 GB
Memory type	GDDR7
Memory bandwidth	1792 GB/s
Compute backend	CUDA
Tier	Consumer
Released	2025
Models (native)	47 / 70
Models (offload)	7 / 70

Software: Full llama.cpp and Ollama support out of the box. CUDA 12.x recommended; driver ≥ 525 required.

Popular models for this GPU

Qwen 2.5 72B Instruct Llama 3.3 70B Instruct DeepSeek R1 Distill Llama 70B Llama 3.1 70B Instruct Mixtral 8x7B Instruct v0.1

Models this GPU runs natively in VRAM (47)

Models that fit with CPU offload (7)

These use system RAM for layers that don't fit in VRAM — expect much slower inference.

Too large for this GPU (16)

Compare NVIDIA RTX 5090 with other GPUs

Frequently asked questions

How much VRAM does the NVIDIA RTX 5090 have?: The NVIDIA RTX 5090 has 32 GB of GDDR7 with 1792 GB/s memory bandwidth.
What is the NVIDIA RTX 5090 best for?: With 32 GB of VRAM, the NVIDIA RTX 5090 is well-suited for running 7B–32B models at Q4 with room for context, making it a great all-rounder for local LLM inference.
What LLMs can the NVIDIA RTX 5090 run locally?: The NVIDIA RTX 5090 can run 47 of the 70 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.3 70B Instruct at Q2_K, Llama 3.1 8B Instruct at BF16, Llama 3.2 3B Instruct at FP32.
Can the NVIDIA RTX 5090 run Llama 3.3 70B Instruct?: Yes. The NVIDIA RTX 5090 runs Llama 3.3 70B Instruct natively in VRAM at Q2_K quantization, achieving approximately 77.8 tokens per second.
Can the NVIDIA RTX 5090 run Qwen 3.6 27B?: Yes. The NVIDIA RTX 5090 runs Qwen 3.6 27B natively in VRAM at NVFP4 quantization, achieving approximately 132.7 tokens per second.
Can the NVIDIA RTX 5090 run Llama 3.1 8B Instruct?: Yes. The NVIDIA RTX 5090 runs Llama 3.1 8B Instruct natively in VRAM at BF16 quantization, achieving approximately 112 tokens per second.