GPT OSS 20B NPU2 — Hardware Requirements & GPU Compatibility
ChatFunctions
GPT OSS 20B NPU2 is a 20B-parameter open language model from FastFlowLM in the GPT-OSS family. It supports a context window of up to 131,072 tokens. At Q4_K_M it needs about 12.37 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 12 GB (Q3_K_M · 64K)Best on 24 GB (Q4_K_M · ~227 tok/s · 128K)Macs from 16 GB (Q4_K_M · ~52 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run GPT OSS 20B NPU2
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
vllm serve FastFlowLM/GPT-OSS-20B-NPU2 --max-model-len 16384vLLM repo with its own quantisation
Download: Weights on Hugging Face
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX A4000 (16 GB) (3 GB headroom) · Q4_K_M
- Generation speed
- ~121 tok/s
- generation speed
- Cost per 1M output tokens
- $0.20
- per 1M output tokens
How Much VRAM Does GPT OSS 20B NPU2 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 8.9 GB | 9.4 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 128K context · ~283 tok/s | |
| 10.1 GB | 10.6 GB | 3-bit medium — 3-bit medium quantization | Fits: 128K context · ~260 tok/s | |
| 12.4 GB | 12.9 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 128K context · ~227 tok/s | |
| 14.6 GB | 15.1 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 128K context · ~201 tok/s | |
| 16.9 GB | 17.4 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 128K context · ~181 tok/s | |
| 20.4 GB | 20.9 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 128K context · ~156 tok/s | |
| 40.4 GB | 40.9 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Does not fit: needs 40.6 GB |
Selected Q4_K_M
4-bit medium quantization12.6GBat 16K context
- Weights
- 12.0 GB
- KV cache
- 0.3 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 11.4 GB free · ~227 tok/s
≥50 tok/s needs ≥167 GB/s · RTX 5090 ~327, RTX 3090 Ti ~227
Which Hardware Can Run GPT OSS 20B NPU2?
Q4_K_M · 12.4 GBGPT OSS 20B NPU2 (Q4_K_M) requires 12.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 17+ GB is recommended. Using the full 131K context window can add up to 2.2 GB, bringing total usage to 14.6 GB. 26 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 5080. 150 devices with unified memory can run GPT OSS 20B NPU2, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M2 Pro (16 GB).
16–18 GB
16–18 GB tier →fits · 3.4 GB free · up to 128K ctx
18 GPUs · 2 Mac classes · 2 device families
RTX 5080~219Simulate output on RTX 5080RTX 5070 Ti~208Simulate output on RTX 5070 TiRTX 4080 SUPER~180Simulate output on RTX 4080 SUPERM516 GB~42Simulate output on MacBook Air 13" M5 (16 GB)M3 Pro*18 GB~42Simulate output on MacBook Pro 14" M3 Pro (18 GB)iPad~35Simulate output on iPad Pro M4 13" (16 GB)Orin NX~32Simulate output on NVIDIA Jetson Orin NX 16GB20 GB
fits · 7.4 GB free · up to 128K ctx
1 GPU
RX 7900 XT~111Simulate output on RX 7900 XT32–40 GB
fits · 19.4 GB free · up to 128K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~327Simulate output on RTX 5090M5 Max36 GB~90Simulate output on Mac Studio M5 Max (36 GB)M532 GB~42Simulate output on MacBook Air 13" M5 (32 GB)Orin~60Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~39Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to GPT OSS 20B NPU2
GPT-OSS family · 14 models →Similar models
4| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Internlm 20BInternLM | — | 45.3 GB | —needs > 24 GB | same size |
| Tri 21B Thinktrillionlabs | — | 45.1 GB | —needs > 24 GB | same size |
| ERNIE 4.5 21B A3B PTBaidu | — | 13.8 GB | ~254 | same size |
| GPT Neox 20BEleutherAI | — | 13.7 GB | ~48 | same size |
Specifications
- Publisher
- FastFlowLM
- Family
- GPT-OSS
- Parameters
- 20B
- Active parameters
- 3.3B per token
- Architecture
- GptOssForCausalLM
- Context Length
- 131,072 tokens
- Vocabulary Size
- 201,088
- Release Date
- 2025-09-26
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does GPT OSS 20B NPU2 need?
GPT OSS 20B NPU2 requires 12.4 GB of VRAM at Q4_K_M, or 40.4 GB at BF16. Full 131K context adds up to 2.2 GB (14.6 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 20B × 4.8 bits ÷ 8 = 12 GB
KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 2.6 GB (at full 131K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run GPT OSS 20B NPU2?
Yes, at Q8_0 (20.4 GB) or lower. Higher quantizations like BF16 (40.4 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for GPT OSS 20B NPU2?
For GPT OSS 20B NPU2, Q4_K_M (12.4 GB) offers the best balance of quality and VRAM usage. Q5_K_M (14.6 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 8.9 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run GPT OSS 20B NPU2 on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run GPT OSS 20B NPU2. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 12.4 GB of usable unified memory (RAM minus macOS overhead).
Can I run GPT OSS 20B NPU2 locally?
Yes — GPT OSS 20B NPU2 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is GPT OSS 20B NPU2?
At Q4_K_M, GPT OSS 20B NPU2 can reach ~192 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~227 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 2.0 GB active ÷ (8000 × 0.65) = 0.39 ms, plus 24 layers × 0.055 ms = 1.32 ms, so 1000 ÷ 1.71 ms = ~585 tok/s
Estimated speed at Q4_K_M (12.4 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of GPT OSS 20B NPU2?
At Q4_K_M, the download is about 12.00 GB. The full-precision BF16 version is 40.00 GB. The smallest option (Q2_K) is 8.50 GB.
Which GPUs can run GPT OSS 20B NPU2?
26 consumer GPUs can run GPT OSS 20B NPU2 at Q4_K_M. Top options include AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090, AMD Radeon RX 6800. 8 GPUs have plenty of headroom for comfortable inference.
Which devices can run GPT OSS 20B NPU2?
150 devices with unified memory can run GPT OSS 20B NPU2 at Q4_K_M (12.4 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.