Can gpt-oss Safeguard 120B run on the RTX 4090?
Not fully in VRAM. Even at 2-bit with 8K context, gpt-oss Safeguard 120B needs about 59.3 GB, more than the 22.8 GB usable on the RTX 4090. It can still run with offload (slower): Q4_K_M with part of the model in system RAM, about 17–34 tokens/sec (estimate). Details and command.
Runs with offload (slower)
Runs with offload 4-bit (Q4_K_M) at 8K context: all attention layers plus the experts of 12 of 36 MoE layers on the RTX 4090 (about 20.6 GB), and the experts of the other 24 layers in system RAM (about 38.2 GB). Needs a PC with at least 64 GB of system RAM. Estimated speed: 17–34 tokens/sec (usable for chat), versus a bandwidth ceiling of ~326 tok/s if it fit fully in VRAM.
llama-server -hf unsloth/gpt-oss-safeguard-120b-GGUF:Q4_K_M -c 8192 -ngl 999 --n-cpu-moe 24
Estimate, not a benchmark. The command downloads the GGUF from Hugging Face. The speed range spans dual-channel DDR4-3200 to DDR5-6000 system RAM; your CPU, RAM speed and llama.cpp version move it. If llama.cpp runs out of GPU memory, raise --n-cpu-moe by 1-2; if VRAM is left over, lower it for more speed. Recent llama.cpp builds can also pick the split automatically. How we estimate offload.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) • | 59.2 GB No | 59.3 GB No | 60.2 GB No | 63.6 GB No | — |
| 3-bit (Q3_K_M) • | 59.3 GB No | 59.4 GB No | 60.2 GB No | 63.6 GB No | — |
| 4-bit (Q4_K_M) • | 59.4 GB No | 59.5 GB No | 60.4 GB No | 63.7 GB No | — |
| 5-bit (Q5_K_M) • | 59.5 GB No | 59.6 GB No | 60.5 GB No | 63.9 GB No | — |
| 6-bit (Q6_K) • | 59.9 GB No | 60.0 GB No | 60.9 GB No | 64.2 GB No | — |
| 8-bit (Q8_0) • | 60.0 GB No | 60.1 GB No | 60.9 GB No | 64.3 GB No | — |
| 16-bit (F16/BF16) | 219 GB No | 219 GB No | 220 GB No | 223 GB No | — |
• real GGUF file size from unsloth/gpt-oss-safeguard-120b-GGUF. Speed is a bandwidth ceiling (1008 GB/s), not a benchmark. Methodology.
Hardware that does fit it
A100 80GBH100 PCIeH100 SXMRTX PRO 6000 BlackwellMac M2 Max 96GBDGX Spark
What fits instead
- Cheapest listed GPU that runs gpt-oss Safeguard 120B: 2x RTX 5090 (2× 32 GB, launch MSRP $3,998) at Q8_0, fully in VRAM at 8K context.
- Largest popular model that fits the RTX 4090: Gemma 4 31B IT (31B) at Q4_K_M, about 20.0 GB at 8K context.
"Runs" and "fits" mean the whole model in GPU memory at 8K context (4-bit or better for the model pick). Prices are US launch MSRPs, not today's street prices. "Popular" means the 60 most-downloaded models we track.
More
gpt-oss Safeguard 120B on every GPU and Mac · Every model that runs on the RTX 4090 · Other models that fit 24 GB VRAM · Try other settings in the calculator
Guides: Best GPU for local LLMs by budget · Can I run gpt-oss, Llama 70B or Qwen3? · Mac vs PC for local LLMs
Data: Hugging Face Hub (openai/gpt-oss-safeguard-120b), NVIDIA specs. Model data last checked 2026-11-17; refreshed hourly.