RunsOnMy ModelsGPUs & MacsBy VRAMBest GPUCan I run…?API pricesVRAM calculatorNew

Can gpt-oss Safeguard 20B run on the Arc B580?

Not fully in VRAM. Even at 2-bit with 8K context, gpt-oss Safeguard 20B needs about 11.7 GB, more than the 11.4 GB usable on the Arc B580. It can still run with offload (slower): Q4_K_M with part of the model in system RAM, about 68–91 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 21 of 24 MoE layers on the Arc B580 (about 9.8 GB), and the experts of the other 3 layers in system RAM (about 1.2 GB). Needs a PC with at least 32 GB of system RAM. Estimated speed: 68–91 tokens/sec (usable for chat), versus a bandwidth ceiling of ~176 tok/s if it fit fully in VRAM.

llama-server -hf unsloth/gpt-oss-safeguard-20b-GGUF:Q4_K_M -c 8192 -ngl 999 --n-cpu-moe 3

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

Quant4K ctx8K ctx32K ctx128K ctxMax tok/s (est., 8K)
2-bit (Q2_K) •11.6 GB No11.7 GB No12.2 GB No14.5 GB No—
3-bit (Q3_K_M) •11.6 GB No11.7 GB No12.3 GB No14.5 GB No—
4-bit (Q4_K_M) •11.7 GB No11.8 GB No12.4 GB No14.6 GB No—
5-bit (Q5_K_M) •11.8 GB No11.9 GB No12.4 GB No14.7 GB No—
6-bit (Q6_K) •12.1 GB No12.2 GB No12.8 GB No15.0 GB No—
8-bit (Q8_0) •12.2 GB No12.3 GB No12.8 GB No15.1 GB No—
16-bit (F16/BF16) •13.7 GB No13.8 GB No14.4 GB No16.6 GB No—

• real GGUF file size from unsloth/gpt-oss-safeguard-20b-GGUF. Speed is a bandwidth ceiling (456 GB/s), not a benchmark. Methodology.

Hardware that does fit it

RTX 4060 Ti 16GBRTX 4070 Ti SuperRTX 4080RTX 4080 SuperRTX 5060 Ti 16GBRTX 5070 Ti

What fits instead

"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 20B on every GPU and Mac · Every model that runs on the Arc B580 · Other models that fit 12 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-20b), Intel specs. Model data last checked 2026-11-17; refreshed hourly.