AMD Strix Halo (128GB)
The AMD Strix Halo (128GB) has 128 GB VRAM and 256 GB/s memory bandwidth. It can run 58 of our 70 tracked models natively in VRAM at 8k context.
With 128 GB LPDDR5X, the AMD Strix Halo (128GB) is a laptop-tier GPU that can run 58 models natively. It handles 70B-class models at Q4 quantization.
AMD Strix Halo (128GB): 2025 unified memory APU with 128GB LPDDR5X at 256 GB/s. RDNA 3.5 architecture.
70B at Q4 native. 405B at Q2 with offload. ~5-10 t/s for 7B.
Vulkan via llama.cpp works cross-platform. ROCm Linux experimental. Unified memory helps but bandwidth is limiting.
| Vendor | AMD |
| Architecture | RDNA 3.5 |
| VRAM | 128 GB (unified) |
| Memory type | LPDDR5X |
| Memory bandwidth | 256 GB/s |
| Compute backend | VULKAN |
| Tier | Laptop |
| Released | 2025 |
| Models (native) | 58 / 70 |
| Models (offload) | 0 / 70 |
Software: Vulkan backend via llama.cpp works cross-platform. ROCm on Linux is experimental but offers better performance.
Popular models for this GPU
Models this GPU runs natively in VRAM (58)
- DeepSeek V4 Flash 284B284B · MMLU-Pro 86.3Q2_K · ~65.8 t/s
- Qwen3 235B-A22B (MoE)235B · MMLU-Pro 84.4Q3_K_M · ~29.8 t/s
- MiniMax M2.5 229B229B · MMLU-Pro 84.8Q3_K_M · ~65.5 t/s
- MiniMax M2.7 229B229B · MMLU-Pro 86.0Q3_K_M · ~65.5 t/s
- Mixtral 8x22B Instruct v0.1141B · MMLU-Pro 40.0Q5_K_M · ~11.2 t/s
- Qwen 3.5 122B-A10B (MoE)122B · MMLU-Pro 86.7Q6_K · ~34.3 t/s
- Nemotron 3 Super 120B120B · MMLU-Pro 83.7Q6_K · ~28.6 t/s
- GPT-OSS 120B117B · MMLU-Pro 80.7Q6_K · ~68.7 t/s
- Llama 4 Scout 109B109B · MMLU-Pro 74.3Q6_K · ~20.2 t/s
- GLM-4.5 Air 106B106B · MMLU-Pro 81.4Q8_0 · ~23.5 t/s
- GLM-4.6V 106B106B · MMLU-Pro 79.9Q8_0 · ~23.5 t/s
- Qwen 2.5 72B Instruct72B · MMLU-Pro 71.1Q8_0 · ~3.6 t/s
- Llama 3.3 70B Instruct70B · MMLU-Pro 68.9Q8_0 · ~3.7 t/s
- DeepSeek R1 Distill Llama 70B70B · MMLU-Pro 70.0Q8_0 · ~3.7 t/s
- Llama 3.1 70B Instruct70B · MMLU-Pro 66.4Q8_0 · ~3.7 t/s
- Mixtral 8x7B Instruct v0.146.7B · MMLU-Pro 29.7BF16 · ~10.9 t/s
- Command-R 35B35B · MMLU-Pro 33.0BF16 · ~3.7 t/s
- Qwen 3.5 35B-A3B (MoE)35B · MMLU-Pro 84.2BF16 · ~46.9 t/s
- Qwen 3.6 35B35B · MMLU-Pro 85.2BF16 · ~3.7 t/s
- Yi 1.5 34B Chat34.4B · MMLU-Pro 37.0BF16 · ~3.7 t/s
- Qwen3 32B32.8B · MMLU-Pro 65.5BF16 · ~3.9 t/s
- Qwen 2.5 32B Instruct32.5B · MMLU-Pro 69.0BF16 · ~3.9 t/s
- Qwen 2.5 Coder 32B Instruct32.5B · MMLU-Pro 50.4BF16 · ~3.9 t/s
- DeepSeek R1 Distill Qwen 32B32.5B · MMLU-Pro 65.0BF16 · ~3.9 t/s
- Nemotron 3 Nano 30B32B · MMLU-Pro 78.3BF16 · ~46.9 t/s
- Gemma 4 31B31B · MMLU-Pro 85.2BF16 · ~4.1 t/s
- Qwen3 30B-A3B (MoE)30B · MMLU-Pro 61.5BF16 · ~46.9 t/s
- Gemma 2 27B Instruct27.2B · MMLU-Pro 38.0BF16 · ~4.7 t/s
- Gemma 3 27B Instruct27B · MMLU-Pro 67.5FP32 · ~2.4 t/s
- Qwen 3.6 27B27B · MMLU-Pro 86.2FP32 · ~2.4 t/s
- Gemma 4 26B (MoE)26B · MMLU-Pro 82.6FP32 · ~18.5 t/s
- Mistral Small 3.1 24B Instruct24B · MMLU-Pro 66.8FP32 · ~2.7 t/s
- Mistral Small 22B22.2B · MMLU-Pro 49.2FP32 · ~2.9 t/s
- GPT-OSS 20B21B · MMLU-Pro 67.9FP32 · ~17.6 t/s
- Qwen3 14B14.8B · MMLU-Pro 61.0FP32 · ~4.3 t/s
- Qwen 2.5 14B Instruct14.7B · MMLU-Pro 63.7FP32 · ~4.4 t/s
- Phi-4 14B Instruct14B · MMLU-Pro 70.4FP32 · ~4.6 t/s
- Mistral Nemo 12B Instruct12.2B · MMLU-Pro 35.6FP32 · ~5.2 t/s
- Gemma 3 12B Instruct12.2B · MMLU-Pro 60.6FP32 · ~5.2 t/s
- Gemma 2 9B Instruct9.2B · MMLU-Pro 32.0FP32 · ~7 t/s
- Llama 3.1 8B Instruct8B · MMLU-Pro 48.3FP32 · ~8 t/s
- DeepSeek R1 Distill Llama 8B8B · MMLU-Pro 41.0FP32 · ~8 t/s
- Qwen3 8B8B · MMLU-Pro 56.7FP32 · ~8 t/s
- Qwen 2.5 7B Instruct7.6B · MMLU-Pro 56.3FP32 · ~8.4 t/s
- Mistral 7B Instruct v0.37.25B · MMLU-Pro 30.0FP32 · ~8.8 t/s
- Gemma 3 4B Instruct4B · MMLU-Pro 43.6FP32 · ~16 t/s
- Gemma 4 E4B4B · MMLU-Pro 69.4FP32 · ~16 t/s
- Phi-3.5 Mini Instruct3.8B · MMLU-Pro 47.4FP32 · ~16.8 t/s
- Llama 3.2 3B Instruct3.2B · MMLU-Pro 24.0FP32 · ~20 t/s
- Qwen 2.5 3B Instruct3.1B · MMLU-Pro 32.4FP32 · ~20.6 t/s
- Gemma 2 2B Instruct2.6B · MMLU-Pro 17.8FP32 · ~24.6 t/s
- Gemma 4 E2B2B · MMLU-Pro 60.0FP32 · ~32 t/s
- SmolLM2 1.7B Instruct1.7B · MMLU-Pro 19.0FP32 · ~37.6 t/s
- Qwen 2.5 1.5B Instruct1.5B · MMLU-Pro 16.8FP32 · ~42.7 t/s
- Llama 3.2 1B Instruct1.24B · MMLU-Pro 12.5FP32 · ~51.6 t/s
- Gemma 3 1B Instruct1B · MMLU-Pro 14.7FP32 · ~64 t/s
- Qwen 2.5 0.5B Instruct0.5B · MMLU-Pro 10.0FP32 · ~128 t/s
- SmolLM2 360M Instruct0.36B · MMLU-Pro 8.0FP32 · ~177.8 t/s
Too large for this GPU (12)
Compare AMD Strix Halo (128GB) with other GPUs
Frequently asked questions
- How much VRAM does the AMD Strix Halo (128GB) have?
- The AMD Strix Halo (128GB) has 128 GB of LPDDR5X with 256 GB/s memory bandwidth (unified system memory, shared between CPU and GPU).
- What is the AMD Strix Halo (128GB) best for?
- With 128 GB of VRAM, the AMD Strix Halo (128GB) is a server-class GPU designed for running the largest open-weight models (70B–405B) at high quantization with ample context.
- What LLMs can the AMD Strix Halo (128GB) run locally?
- The AMD Strix Halo (128GB) can run 58 of the 70 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Llama 3.3 70B Instruct at Q8_0, Llama 3.1 8B Instruct at FP32, Llama 3.2 3B Instruct at FP32.
- Can the AMD Strix Halo (128GB) run Llama 3.3 70B Instruct?
- Yes. The AMD Strix Halo (128GB) runs Llama 3.3 70B Instruct natively in VRAM at Q8_0 quantization, achieving approximately 3.7 tokens per second.
- Can the AMD Strix Halo (128GB) run Qwen 3.6 27B?
- Yes. The AMD Strix Halo (128GB) runs Qwen 3.6 27B natively in VRAM at FP32 quantization, achieving approximately 2.4 tokens per second.
- Can the AMD Strix Halo (128GB) run Llama 3.1 8B Instruct?
- Yes. The AMD Strix Halo (128GB) runs Llama 3.1 8B Instruct natively in VRAM at FP32 quantization, achieving approximately 8 tokens per second.