GPT OSS 20B Heretic — Hardware Requirements & GPU Compatibility
Chat
4.2K downloads
98 likesSpecifications
- Publisher
- p-e-w
- Family
- GPT-OSS
- Parameters
- 20.9B
- Architecture
- GptOssForCausalLM
- Context Length
- 131,072 tokens
- Vocabulary Size
- 201,088
- Release Date
- 2025-11-16
- License
- Apache 2.0
Get Started
HuggingFace
How Much VRAM Does GPT OSS 20B Heretic Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| IQ4_NL | 4.50 | 12.1 GB | 16.6 GB | 11.76 GB | Importance-weighted 4-bit, non-linear |
| Q5_1 | 5.50 | 14.8 GB | 19.2 GB | 14.38 GB | 5-bit legacy quantization with offset |
| Q8_0 | 8.00 | 21.3 GB | 25.7 GB | 20.91 GB | 8-bit quantization, near-lossless |
Which GPUs Can Run GPT OSS 20B Heretic?
Q8_0 · 21.3 GBGPT OSS 20B Heretic (Q8_0) requires 21.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 28+ GB is recommended. Using the full 131K context window can add up to 4.4 GB, bringing total usage to 25.7 GB. 5 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti.
Runs great
— Plenty of headroom
Decent
— Enough VRAM, may be tightWhich Devices Can Run GPT OSS 20B Heretic?
Q8_0 · 21.3 GB21 devices with unified memory can run GPT OSS 20B Heretic, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Mini M4 Pro (24 GB).
Runs great
— Plenty of headroomRelated Models
Derivatives (1)
Frequently Asked Questions
- How much VRAM does GPT OSS 20B Heretic need?
GPT OSS 20B Heretic requires 12.1 GB of VRAM at IQ4_NL, or 21.3 GB at Q8_0. Full 131K context adds up to 4.4 GB (16.6 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 20.9B × 4.5 bits ÷ 8 = 11.8 GB
KV Cache + Overhead ≈ 0.3 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 4.8 GB (at full 131K context)
VRAM usage by quantization
IQ4_NL12.1 GBIQ4_NL + full context16.6 GB- What's the best quantization for GPT OSS 20B Heretic?
For GPT OSS 20B Heretic, Q5_1 (14.8 GB) offers the best balance of quality and VRAM usage. Q8_0 (21.3 GB) provides better quality if you have the VRAM. The smallest option is IQ4_NL at 12.1 GB.
VRAM requirement by quantization
IQ4_NL12.1 GB~88%Q5_1 ★14.8 GB~92%Q8_021.3 GB~99%★ Recommended — best balance of quality and VRAM usage.
- Can I run GPT OSS 20B Heretic on a Mac?
GPT OSS 20B Heretic requires at least 12.1 GB at IQ4_NL, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
- Can I run GPT OSS 20B Heretic locally?
Yes — GPT OSS 20B Heretic can run locally on consumer hardware. At IQ4_NL quantization it needs 12.1 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is GPT OSS 20B Heretic?
At IQ4_NL, GPT OSS 20B Heretic can reach ~240 tok/s on AMD Instinct MI300X. On NVIDIA GeForce RTX 4090: ~54 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: AMD Instinct MI300X → 5300 ÷ 12.1 × 0.55 = ~240 tok/s
Estimated speed at IQ4_NL (12.1 GB)
AMD Instinct MI300X~240 tok/sNVIDIA GeForce RTX 4090~54 tok/sNVIDIA H100 SXM~180 tok/sAMD Instinct MI250X~149 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of GPT OSS 20B Heretic?
At IQ4_NL, the download is about 11.76 GB. The full-precision Q8_0 version is 20.91 GB.
- Which GPUs can run GPT OSS 20B Heretic?
17 consumer GPUs can run GPT OSS 20B Heretic at IQ4_NL (12.1 GB). Top options include AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090, AMD Radeon RX 6800. 6 GPUs have plenty of headroom for comfortable inference.
- Which devices can run GPT OSS 20B Heretic?
27 devices with unified memory can run GPT OSS 20B Heretic at IQ4_NL (12.1 GB), including Mac Mini M4 (16 GB), Mac Mini M4 (32 GB), Mac Mini M4 Pro (24 GB), Mac Mini M4 Pro (48 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.