MetaLlama 3.3 70B Instruct on NVIDIA GeForce RTX 5090
RTX 5090 cannot run Llama 3.3 70B Instruct. 32 GB VRAM is insufficient at any quantization level.
Model Size
70.6B
Device VRAM
32 GB
Bandwidth
1792 GB/s
Quantization
Q4_K_M
Performance by Quantization
OwnRig currently has one published compatibility entry for Llama 3.3 70B Instruct on NVIDIA GeForce RTX 5090 at Q4_K_M. This pairing has limitations — check the rating and notes below.
| Quantization | Speed | TTFT | Fits in VRAM | Rating | Confidence |
|---|---|---|---|---|---|
| Q4_K_M | 8 tok/s | 1900ms | ✗ Offload | Marginal | estimated |
Notes
Q4_K_M
Model data lists Q4 at ~41GB VRAM required; 32GB is below that, so expect offload for Q4. Q3_K_M (~33GB) is still over 32GB VRAM; use offload or a 48GB+ unified Mac (M4 Pro) / larger GPU for cleaner Q4 fits.
About Llama 3.3 70B Instruct
Llama 3.3 70B Instruct (70.6B) is a chat, coding, reasoning, multi-purpose model. Flagship Llama 3.3 model with best-in-class general and coding performance.
View all Llama 3.3 70B Instruct hardware options →About NVIDIA GeForce RTX 5090
NVIDIA GeForce RTX 5090 has 32 GB at 1792 GB/s. Street price: $2,199.
See all models NVIDIA GeForce RTX 5090 can run →Builds with NVIDIA GeForce RTX 5090
Estimate method: Community benchmarks and estimated performance. Reference hardware source: github.com (2026-03-01)
Performance varies by driver version, inference engine, quantization method, context length, and system configuration. Figures shown are estimates based on community benchmarks and may not reflect your exact setup. Product names are trademarks of their respective owners. OwnRig is independent and not affiliated with any hardware or AI model provider.