GeForce RTX 4060 Ti 16 GB
NVIDIA
ExcellentRuns gpt-oss-20b MXFP4 entirely in memory
Model · OpenAI
gpt-oss-20b: VRAM and hardware requirements
At MXFP4 with an 8K context, gpt-oss-20b needs about 13.6 GB: a GPU with 16 GB or more runs it entirely in VRAM.
- Parameters20.9B
- Active3.6B
- Max context128K
- LicenseApache-2.0
Specifications
- Organization
- OpenAI
- Architecture
- gpt_oss
- Released
- August 2025
- Layers
- 24
- Hidden size
- 2880
- Attention heads
- 64
- KV heads
- 8
- Head dimension
- 64
- Vocabulary
- 201,088
- KV cache per 1K tokens (f16)
- 0.026 GB
- Repository
- openai/gpt-oss-20b
Official specData updated 24 days agoSource: Hugging Face config.json + model card
Quantizations and memory
| Quantization | File size | Memory needed | Minimum GPU memory | Quality |
|---|---|---|---|---|
| MXFP4Most used | 11.3 GB | ~13.6 GB | 16 GB | Reference (unquantized) |
File sizes: not available until imported from the model repository. Memory = weights (estimated from bits per weight) + KV cache + runtime overhead + safety margin. (MXFP4 ≈ 4.9 bpw)
Q4, Q5 or Q8: how to choose a quantization
GPUs that run gpt-oss-20b
| RTX 509032 GB · 64 GB RAM~13 GB · 264–291 tok/s measured | Excellent |
|---|---|
| RTX 409024 GB · 64 GB RAM~13 GB · 212–243 tok/s measured | Excellent |
| RTX 309024 GB · 64 GB RAM~13 GB · 147–163 tok/s measured | Excellent |
| RX 7900 XTX24 GB · 64 GB RAM~13 GB · 146–201 tok/s estimate | Excellent |
| RTX 5060 Ti 16 GB16 GB · 64 GB RAM~13 GB · 96–117 tok/s measured | Excellent |
| RTX 5070 Ti16 GB · 64 GB RAM~13 GB · 163–199 tok/s measured | Excellent |
| RTX 508016 GB · 64 GB RAM~13 GB · 176–215 tok/s measured | Excellent |
| RTX 4060 Ti 16 GB16 GB · 64 GB RAM~13 GB · 82–91 tok/s measured | Excellent |
| RTX 4070 Ti SUPER16 GB · 64 GB RAM~13 GB · 85–234 tok/s estimate | Excellent |
| RX 9070 XT16 GB · 64 GB RAM~13 GB · 97–134 tok/s estimate | Excellent |
| RTX 3060 12 GB12 GB · 64 GB RAM~13 GB · 41–75 tok/s estimate | CPU/GPU offload |
| RTX 407012 GB · 64 GB RAM~13 GB · 46–88 tok/s estimate | CPU/GPU offload |
| RTX 507012 GB · 64 GB RAM~13 GB · 51–98 tok/s estimate | CPU/GPU offload |
| RTX 40608 GB · 64 GB RAM~13 GB · 18–33 tok/s estimate | CPU/GPU offload |
Macs that run gpt-oss-20b
| M3 Ultra (80-core GPU) · 512 GBMac Studio~12.8 GB · 105–116 tok/s measured | Excellent |
|---|---|
| M4 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~12.8 GB · 62–140 tok/s estimate | Excellent |
| M3 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~12.8 GB · 45–103 tok/s estimate | Excellent |
| M4 Pro (20-core GPU) · 64 GBMac mini, MacBook Pro 14~12.8 GB · 31–70 tok/s estimate | Excellent |
| M4 Max (32-core GPU) · 36 GBMacBook Pro 14, MacBook Pro 16~12.8 GB · 85–94 tok/s measured | Excellent |
| M4 · 32 GBMacBook Air, MacBook Pro 14~12.8 GB · 14–31 tok/s estimate | Excellent |
| M5 · 32 GBMacBook Pro 14~12.8 GB · 18–39 tok/s estimate | Excellent |
| M3 · 24 GBMacBook Air, MacBook Pro 14~12.8 GB · 12–26 tok/s estimate | Excellent |
EstimateData updated 24 days agoSource: Run LLM Locally engine
Limits in plain words
gpt-oss-20b is a mixture-of-experts model: all 20.9B parameters must be in memory, but only 3.6B are read per token, so it generates much faster than a dense model of the same size.
The KV cache grows with context: at 32K it adds about 0.8 GB on top of the weights.
With under 8 GB of VRAM, use a smaller quantization or a smaller model of the same family.
Where to buy
Prices and links follow the market you choose.
Radeon RX 7600 XT
AMD
ExcellentRuns gpt-oss-20b MXFP4 entirely in memory
AMD Ryzen AI Max+ 395
Unified memory PC
ExcellentRuns gpt-oss-20b MXFP4 entirely in memory
No current offer imported for this market: the link opens a marketplace search.
Frequently asked questions
How much VRAM does gpt-oss-20b need?
About 13.6 GB at MXFP4 with an 8K context, including the KV cache and runtime overhead. At MXFP4 it needs about 13.6 GB. A longer context adds more: about 0.8 GB of KV cache at 32K.
Can gpt-oss-20b run on a Mac?
Yes on a Mac whose GPU can use at least 12.8 GB, e.g. M3 24 GB. macOS lets the GPU use about two thirds to three quarters of unified memory by default.
How fast is gpt-oss-20b?
On a RTX 4090, we estimate 212–243 tok/s at MXFP4. This is an estimate from memory bandwidth (confidence medium) until benchmarks are imported.