gpt-oss 20B: VRAM requirements & speed
gpt-oss 20B (20.91B MoE, 3.6B active, apache-2.0). Native context 128k. Smallest GGUF: 11.62 GB.
Memory needed by quantization and context
GGUF file size plus f16 KV cache (and fixed overhead of ~1 GB per GPU not shown). Source: unsloth/gpt-oss-20b-GGUF.
| Quant | File | +8k ctx | +32k ctx | +128k ctx |
|---|---|---|---|---|
| Q4_K_M | 11.62 GB | 11.8 GB | 12.4 GB | 14.8 GB |
| Q5_K_M | 11.72 GB | 11.9 GB | 12.5 GB | 14.9 GB |
| Q6_K | 12.04 GB | 12.2 GB | 12.8 GB | 15.3 GB |
| Q8_0 | 12.11 GB | 12.3 GB | 12.9 GB | 15.3 GB |
KV cache size by context length
Computed from the model's attention layout (sliding-window layers capped at their window).
| Context | f16 | q8_0 | q4_0 |
|---|---|---|---|
| 8k | 0.20 GB | 0.11 GB | 0.06 GB |
| 32k | 0.81 GB | 0.43 GB | 0.23 GB |
| 128k | 3.22 GB | 1.71 GB | 0.91 GB |
Which hardware runs gpt-oss 20B? · 16k context
| Hardware | Quant | Fit | Memory | Tok/s | Speed source |
|---|---|---|---|---|---|
| RTX 5090 32GB | Q8_0 | fits fully | 12.5 GB | 298 | scaled from measured |
| RTX PRO 6000 96GB (bonus page) | Q8_0 | fits fully | 12.5 GB | 287 | scaled from measured |
| Mac Studio M5 Ultra 256GB | Q8_0 | fits fully | 12.5 GB | 285 | estimated |
| Mac Studio M5 Ultra 512GB (ships late Oct 2026) | Q8_0 | fits fully | 12.5 GB | 285 | estimated |
| RTX 4090 24GB | Q8_0 | fits fully | 12.5 GB | 258 | measured |
| 2x RTX 3090 (48GB total) | Q8_0 | fits fully | 12.5 GB | 241 | estimated |
| AMD RX 7900 XTX 24GB | Q8_0 | fits fully | 12.5 GB | 219 | estimated |
| RTX 3090 24GB | Q8_0 | fits fully | 12.5 GB | 204 | measured |
| RTX 5070 Ti 16GB | Q8_0 | fits fully | 12.5 GB | 189 | scaled from measured |
| RTX 5080 16GB | Q8_0 | fits fully | 12.5 GB | 187 | scaled from measured |
| Mac Studio M5 Max 128GB | Q8_0 | fits fully | 12.5 GB | 180 | estimated |
| MacBook Pro 14/16 M5 Max 128GB | Q8_0 | fits fully | 12.5 GB | 180 | estimated |
| RTX 5070 12GB | Q4_K_M | MoE offload to CPU | 12.0 GB | 132 | estimated |
| RTX 4070 12GB | Q4_K_M | MoE offload to CPU | 12.0 GB | 121 | estimated |
| Mac Studio (2025) M3 Ultra 96GB | Q8_0 | fits fully | 12.5 GB | 115 | scaled from measured |
| MacBook Pro 14/16 (2024) M4 Max 128GB | Q8_0 | fits fully | 12.5 GB | 97 | estimated |
| 16GB VRAM class (RTX 4060 Ti 16GB, 32GB RAM) | Q8_0 | fits fully | 12.5 GB | 94 | measured |
| AMD Ryzen AI Max+ 395 (Strix Halo) 128GB | Q8_0 | fits fully | 12.5 GB | 93 | scaled from measured |
| NVIDIA DGX Spark 128GB | Q8_0 | fits fully | 12.5 GB | 86 | scaled from measured |
| MacBook Pro 14/16 M5 Pro 48GB | Q8_0 | fits fully | 12.5 GB | 80 | estimated |
| RTX 3060 12GB (32GB RAM) | Q4_K_M | MoE offload to CPU | 12.0 GB | 77 | estimated |
| Mac mini (2024) M4 Pro 64GB | Q8_0 | fits fully | 12.5 GB | 59 | estimated |
| 8GB VRAM class (RTX 4060 8GB, 32GB RAM) | Q4_K_M | MoE offload to CPU | 12.0 GB | 44 | estimated |
| Mac mini M6 32GB | Q8_0 | fits fully | 12.5 GB | 40 | estimated |
| MacBook Air M5 32GB | Q8_0 | fits fully | 12.5 GB | 37 | estimated |
| Mac mini / iMac M4 16GB | - | does not fit | - | - |
Discrete GPUs are modelled with 32 GB of system RAM available for offload. Does not fit even with offload: Mac mini / iMac M4 16GB.
Links
Model card · GGUF files · Open in the advisor · last verified 2026-10-14
FAQ
How much VRAM does gpt-oss 20B need?
The smallest quant (Q4_K_M) is 11.62 GB, so you need roughly 13 GB for an 8k context including KV cache and overhead. At 128k context the f16 KV cache alone is 3.2 GB.
Can I run gpt-oss 20B on a RTX 5090 32GB?
Yes: on RTX 5090 32GB it fits fully as Q8_0 at about 298 tok/s (scaled from measured). 21 of 26 hardware configurations fit it fully at 16k context.
What is the maximum context for gpt-oss 20B?
128k tokens natively. Memory, not the model, is usually the limit: see the KV cache table above.
All models: model index · by memory: 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · 48 GB · 64 GB · 96 GB · 128 GB · 192 GB · 256 GB · 512 GB. Numbers use the same fit math as the advisor; sources and methods.