Can the RTX 5090 D run Llama 3.3 70B?
No. Llama 3.3 70B needs about 41.8 GB at Q4_K_M (weights plus 1.8 GB runtime overhead), and the RTX 5090 D has 32 GB, so it won't fit without offloading layers to system RAM, which is far slower.
Every quantization
Llama 3.3 70B (70B) on 32 GB
| Quant | Weights | Needs | Fit | Est. speed |
|---|---|---|---|---|
| Q4_K_M | 40 GB | 41.8 GB | Too big | - |
| Q5_K_M | 48 GB | 49.8 GB | Too big | - |
| Q8_0 | 75 GB | 76.8 GB | Too big | - |
| FP16 | 140 GB | 141.8 GB | Too big | - |
Estimates. Needs = weights + 1.8 GB for KV cache and runtime at modest context. Speed is single-stream decode scaled from the RTX 5090 D's 1790 GB/s bandwidth (Estimated from memory bandwidth (llama.cpp decode)). Long contexts need more memory and run slower.
What to do instead
Two ways forward
Cheapest cards that run it
- NVIDIA RTX A6000 - 48 GB, ~$3,299
- Mac Studio M4 Ultra - 192 GB, ~$4,699
- NVIDIA RTX 6000 Ada Generation - 48 GB, ~$5,800
- NVIDIA L40S - 48 GB, ~$8,500
Models that fit the RTX 5090 D
- Qwen3.6 35B A3B - 36B, 21 GB at Q4
- Qwen3.5 35B A3B - 36B, 21 GB at Q4
- dolphin-2.9.1-yi-1.5-34b - 34.4B, 20 GB at Q4
- Gemma 4 31B - 32.7B, 19 GB at Q4
Compare
Same model, other hardware - and other models on this card
Llama 3.3 70B on other hardware
- RTX 6000D~23 tok/s
- MacBook Pro 16" M4 Max~7.8 tok/s
- DGX Spark (GB10)~3.9 tok/s
- RTX Spark N1X (128 GB)~3.9 tok/s
Also runs on the RTX 5090 D
- Qwen3.6 35B A3B~391 tok/s
- Qwen3.5 35B A3B~391 tok/s
- dolphin-2.9.1-yi-1.5-34b~51 tok/s
- Gemma 4 31B~54 tok/s
- Qwen3 32B~54 tok/s
- GLM-4.7-Flash~395 tok/s
- Qwen3-30B-A3B-Instruct-2507~387 tok/s
- Qwen3.6 27B~64 tok/s
Frequently asked
Quick answers
Can the GeForce RTX 5090 D run Llama 3.3 70B?
No. Llama 3.3 70B needs about 41.8 GB at Q4_K_M (weights plus 1.8 GB runtime overhead), and the RTX 5090 D has 32 GB, so it won't fit without offloading layers to system RAM, which is far slower.
How much VRAM does Llama 3.3 70B need?
About 40 GB of weights at Q4_K_M, 75 GB at Q8_0 and 140 GB at FP16. Add roughly 1.8 GB for the KV cache and runtime at modest context lengths.
How fast is Llama 3.3 70B on the RTX 5090 D?
It doesn't fit in the RTX 5090 D's memory, so speed depends on how many layers you offload to system RAM - typically single-digit tokens per second or lower.