HomeCan I run it?Llama 3.3 70BLlama 3.3 70B on an RTX 5090
No. Llama 3.3 70B at Q4_K_M needs about 50.2 GB, and the GeForce RTX 5090 has 32.0 GB usable at a ~4k context.
Can I run Llama 3.3 70B on an RTX 5090?
Llama 3.3 70B at Q4_K_M needs about 50.2 GB, and the GeForce RTX 5090 has 32.0 GB usable. That figure is 32 GB of VRAM on the spec sheet, of which 32.0 GB is usable for weights and cache, at 1792 GB/s memory bandwidth. Meta’s flagship 70B instruct model, the usual target for a high-end local chat box. These are estimates from our formula, not measured benchmarks.
- Quant that fits
- None of Q4, Q8 or FP16 needs 50.2 GB
- Memory needed
- 50.2 GB at Q4_K_M, ~4k context
- Usable VRAM
- 32.0 GB of 32 GB VRAM
- Est. tokens/sec
- n/a does not fit
Memory at each quant
| Quant | Need vs 32.0 GB | Verdict |
|---|---|---|
| Q4_K_M | 50.2 GB | Q4Won’t fit |
| Q8_0 | 87.2 GB | Q8Won’t fit |
| FP16 | 163.3 GB | FP16Won’t fit |
Estimated tokens per second
No speed estimate: Llama 3.3 70B does not fit in the 32.0 GB this machine can use, so we do not publish a tokens/sec figure for CPU offload.
Context headroom
Even at a short ~4k context, the weights do not fit. Extending context would only increase the gap.
| Context | KV cache (FP16) | Total at Q4_K_M | Verdict |
|---|---|---|---|
| 4k | 1.3 GB | 50.2 GB | Won’t fit |
| 8k | 2.7 GB | 51.5 GB | Won’t fit |
| 16k | 5.4 GB | 54.2 GB | Won’t fit |
| 32k | 10.7 GB | 59.6 GB | Won’t fit |
| 64k | 21.5 GB | 70.3 GB | Won’t fit |
| 128k | 42.9 GB | 91.8 GB | Won’t fit |
What to do instead
- Run R1 Distill 32B instead (23.8 GB at Q4). It is a different model, not a drop-in, but it does fit this machine.
- To hold Llama 3.3 70B at Q4, you need about 50.2 GB. The MacBook Pro M5 Max 128GB has 96.0 GB usable ($6699 Apple list price, 14-inch M5 Max 18C CPU/40C GPU, 128GB/2TB).
- Two used RTX 3090s give 48 GB combined. Our Q4 estimate for this 70B is just over that, so we don’t mark the pair as a fit; see the dual 3090 build and multi-GPU notes.
- CPU offload can load the weights anyway, but it is slow. We do not publish an offload tokens/sec figure.
- See the best GPU for this model family and the Llama-3.3-70B-Instruct model page.
Other hardware for Llama 3.3 70B
- Llama 3.3 70B on an RTX 3060 12GBWon’t fit
- Llama 3.3 70B on an RTX 3070 TiWon’t fit
- Llama 3.3 70B on an RTX 3080Won’t fit
- Llama 3.3 70B on an RTX 3090Won’t fit
- Llama 3.3 70B on an RTX 4060 Ti 16GBWon’t fit
- Llama 3.3 70B on an RTX 4070 Ti SuperWon’t fit
Other models on an RTX 5090
- Qwen3-32B on an RTX 5090Fits
- Qwen3-30B-A3B on an RTX 5090Fits
- Qwen3.8-27B on an RTX 5090Fits
- Gemma 3 27B on an RTX 5090Fits
- Gemma 4 31B on an RTX 5090Fits
- gpt-oss-20b on an RTX 5090Fits
Quick answers
- Can I run Llama 3.3 70B on an RTX 5090?
- No. Llama 3.3 70B at Q4_K_M needs about 50.2 GB, and the GeForce RTX 5090 has 32.0 GB usable.
- How much VRAM does Llama 3.3 70B need?
- About 50.2 GB at Q4_K_M, 87.2 GB at Q8_0 and 163.3 GB at FP16, estimated for a ~4k context.
- How fast is Llama 3.3 70B on an RTX 5090?
- We do not estimate tokens/sec when the model does not fit. CPU offload is possible but slow.
- What context length can I use?
- The weights already do not fit, so longer context is not the limiting factor.