Quick Answer: RX 9070 XT offers 16GB VRAM and starts around current market pricing. It delivers approximately 130 tokens/sec on apple/OpenELM-1_1B-Instruct. It typically draws 280W under load.
This GPU offers reliable throughput for local AI workloads. Pair it with the right model quantization to hit your desired tokens/sec, and monitor prices below to catch the best deal.
With 16GB VRAM, RX 9070 XT can run models up to approximately 40B parameters using 4-bit quantization. This handles most popular models including Llama 3 70B, Mistral 7B, and larger.
Consider RTX 4090 — Double the VRAM for larger models.
| Model | Quantization | Tokens/sec | VRAM used |
|---|---|---|---|
| apple/OpenELM-1_1B-Instruct | Q4 | 129.63 tok/sEstimated Auto-generated benchmark | 1GB |
| unsloth/gemma-3-1b-it | Q4 | 129.20 tok/sEstimated Auto-generated benchmark | 1GB |
| tencent/HunyuanOCR | Q4 | 127.68 tok/sEstimated Auto-generated benchmark | 1GB |
| ibm-research/PowerMoE-3b | Q4 | 127.64 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-2b | Q4 | 126.68 tok/sEstimated Auto-generated benchmark | 1GB |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Q4 | 126.51 tok/sEstimated Auto-generated benchmark | 1GB |
| bigcode/starcoder2-3b | Q4 | 126.03 tok/sEstimated Auto-generated benchmark | 2GB |
| LiquidAI/LFM2-1.2B | Q4 | 125.86 tok/sEstimated Auto-generated benchmark | 1GB |
| deepseek-ai/DeepSeek-OCR | Q4 | 125.67 tok/sEstimated Auto-generated benchmark | 2GB |
| allenai/OLMo-2-0425-1B | Q4 | 125.59 tok/sEstimated Auto-generated benchmark | 1GB |
| deepseek-ai/DeepSeek-OCR-2 | Q4 | 125.13 tok/sEstimated Auto-generated benchmark | 2GB |
| nineninesix/kani-tts-2-en | Q4 | 124.52 tok/sEstimated Auto-generated benchmark | 1GB |
| WeiboAI/VibeThinker-1.5B | Q4 | 124.49 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen3-ASR-1.7B | Q4 | 122.98 tok/sEstimated Auto-generated benchmark | 2GB |
| google/embeddinggemma-300m | Q4 | 121.67 tok/sEstimated Auto-generated benchmark | 1GB |
| facebook/sam3 | Q4 | 120.46 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen2.5-3B | Q4 | 120.30 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.2-1B | Q4 | 119.08 tok/sEstimated Auto-generated benchmark | 1GB |
| unsloth/Llama-3.2-3B-Instruct | Q4 | 118.28 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-3.2-3B | Q4 | 118.21 tok/sEstimated Auto-generated benchmark | 2GB |
| nari-labs/Dia2-2B | Q4 | 116.62 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-2-2b-it | Q4 | 116.53 tok/sEstimated Auto-generated benchmark | 1GB |
| inference-net/Schematron-3B | Q4 | 116.47 tok/sEstimated Auto-generated benchmark | 2GB |
| meta-llama/Llama-Guard-3-1B | Q4 | 115.07 tok/sEstimated Auto-generated benchmark | 1GB |
| ibm-granite/granite-3.3-2b-instruct | Q4 | 114.49 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-3B-Instruct | Q4 | 113.81 tok/sEstimated Auto-generated benchmark | 2GB |
| google/gemma-3-1b-it | Q4 | 112.76 tok/sEstimated Auto-generated benchmark | 1GB |
| Qwen/Qwen2.5-3B-Instruct | Q4 | 112.75 tok/sEstimated Auto-generated benchmark | 2GB |
| unsloth/Llama-3.2-1B-Instruct | Q4 | 112.24 tok/sEstimated Auto-generated benchmark | 1GB |
| google-bert/bert-base-uncased | Q4 | 111.13 tok/sEstimated Auto-generated benchmark | 1GB |
| meta-llama/Llama-3.2-1B-Instruct | Q4 | 110.00 tok/sEstimated Auto-generated benchmark | 1GB |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q4 | 107.81 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2-1.5B-Instruct | Q4 | 107.67 tok/sEstimated Auto-generated benchmark | 3GB |
| kaitchup/Phi-3-mini-4k-instruct-gptq-4bit | Q4 | 107.56 tok/sEstimated Auto-generated benchmark | 2GB |
| vikhyatk/moondream2 | Q4 | 107.36 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/DeepSeek-V3-0324 | Q4 | 107.26 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen2.5-Coder-7B-Instruct | Q4 | 107.24 tok/sEstimated Auto-generated benchmark | 4GB |
| dicta-il/dictalm2.0-instruct | Q4 | 107.12 tok/sEstimated Auto-generated benchmark | 4GB |
| microsoft/phi-4 | Q4 | 107.05 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-Embedding-8B | Q4 | 107.04 tok/sEstimated Auto-generated benchmark | 4GB |
| deepseek-ai/deepseek-coder-1.3b-instruct | Q4 | 106.92 tok/sEstimated Auto-generated benchmark | 2GB |
| tencent/HunyuanVideo-1.5 | Q4 | 106.85 tok/sEstimated Auto-generated benchmark | 4GB |
| context-labs/meta-llama-Llama-3.2-3B-Instruct-FP16 | Q4 | 106.80 tok/sEstimated Auto-generated benchmark | 2GB |
| google-t5/t5-3b | Q4 | 106.58 tok/sEstimated Auto-generated benchmark | 2GB |
| Nanbeige/Nanbeige4.1-3B | Q4 | 106.19 tok/sEstimated Auto-generated benchmark | 3GB |
| black-forest-labs/FLUX.2-dev | Q4 | 106.13 tok/sEstimated Auto-generated benchmark | 4GB |
| HuggingFaceTB/SmolLM-135M | Q4 | 106.09 tok/sEstimated Auto-generated benchmark | 4GB |
| Qwen/Qwen3-0.6B | Q4 | 105.39 tok/sEstimated Auto-generated benchmark | 3GB |
| lmstudio-community/Qwen3-4B-Thinking-2507-MLX-6bit | Q4 | 104.98 tok/sEstimated Auto-generated benchmark | 2GB |
| deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct | Q4 | 104.89 tok/sEstimated Auto-generated benchmark | 4GB |
Note: Performance estimates are calculated. Real results may vary. Methodology · Submit real data
| Model | Quantization | Verdict | Estimated speed | VRAM needed |
|---|---|---|---|---|
| openai-community/gpt2 | Q8 | Fits comfortably | 72.64 tok/sEstimated | 7GB (have 16GB) |
| openai-community/gpt2 | FP16 | Fits (tight) | 34.29 tok/sEstimated | 15GB (have 16GB) |
| Qwen/Qwen2.5-7B-Instruct | Q4 | Fits comfortably | 103.83 tok/sEstimated | 4GB (have 16GB) |
| Qwen/Qwen2.5-7B-Instruct | Q8 | Fits comfortably | 62.51 tok/sEstimated | 7GB (have 16GB) |
| Qwen/Qwen2.5-7B-Instruct | FP16 | Fits (tight) | 38.98 tok/sEstimated | 15GB (have 16GB) |
| Qwen/Qwen3-0.6B | Q4 | Fits comfortably | 105.39 tok/sEstimated | 3GB (have 16GB) |
| Qwen/Qwen3-0.6B | Q8 | Fits comfortably | 71.31 tok/sEstimated | 6GB (have 16GB) |
| Qwen/Qwen3-0.6B | FP16 | Fits comfortably | 37.53 tok/sEstimated | 13GB (have 16GB) |
| Gensyn/Qwen2.5-0.5B-Instruct | Q4 | Fits comfortably | 92.44 tok/sEstimated | 3GB (have 16GB) |
| Gensyn/Qwen2.5-0.5B-Instruct | Q8 | Fits comfortably | 66.12 tok/sEstimated | 5GB (have 16GB) |
| Gensyn/Qwen2.5-0.5B-Instruct | FP16 | Fits comfortably | 37.18 tok/sEstimated | 11GB (have 16GB) |
| meta-llama/Llama-3.1-8B-Instruct | Q4 | Fits comfortably | 97.38 tok/sEstimated | 4GB (have 16GB) |
| meta-llama/Llama-3.1-8B-Instruct | Q8 | Fits comfortably | 74.65 tok/sEstimated | 9GB (have 16GB) |
| meta-llama/Llama-3.1-8B-Instruct | FP16 | Not supported | 39.83 tok/sEstimated | 17GB (have 16GB) |
| dphn/dolphin-2.9.1-yi-1.5-34b | Q4 | Not supported | 33.04 tok/sEstimated | 17GB (have 16GB) |
| dphn/dolphin-2.9.1-yi-1.5-34b | Q8 | Not supported | 25.28 tok/sEstimated | 35GB (have 16GB) |
| dphn/dolphin-2.9.1-yi-1.5-34b | FP16 | Not supported | 12.81 tok/sEstimated | 70GB (have 16GB) |
| openai/gpt-oss-20b | Q4 | Fits comfortably | 55.38 tok/sEstimated | 10GB (have 16GB) |
| openai/gpt-oss-20b | Q8 | Not supported | 39.82 tok/sEstimated | 20GB (have 16GB) |
| openai/gpt-oss-20b | FP16 | Not supported | 19.82 tok/sEstimated | 41GB (have 16GB) |
| google/gemma-3-1b-it | Q4 | Fits comfortably | 112.76 tok/sEstimated | 1GB (have 16GB) |
| google/gemma-3-1b-it | Q8 | Fits comfortably | 75.33 tok/sEstimated | 1GB (have 16GB) |
| google/gemma-3-1b-it | FP16 | Fits comfortably | 44.56 tok/sEstimated | 2GB (have 16GB) |
| Qwen/Qwen3-Embedding-0.6B | Q4 | Fits comfortably | 94.35 tok/sEstimated | 3GB (have 16GB) |
| Qwen/Qwen3-Embedding-0.6B | Q8 | Fits comfortably | 67.65 tok/sEstimated | 6GB (have 16GB) |
| Qwen/Qwen3-Embedding-0.6B | FP16 | Fits comfortably | 34.31 tok/sEstimated | 13GB (have 16GB) |
| Qwen/Qwen2.5-1.5B-Instruct | Q4 | Fits comfortably | 100.80 tok/sEstimated | 3GB (have 16GB) |
| Qwen/Qwen2.5-1.5B-Instruct | Q8 | Fits comfortably | 64.39 tok/sEstimated | 5GB (have 16GB) |
| Qwen/Qwen2.5-1.5B-Instruct | FP16 | Fits comfortably | 33.69 tok/sEstimated | 11GB (have 16GB) |
| facebook/opt-125m | Q4 | Fits comfortably | 93.80 tok/sEstimated | 4GB (have 16GB) |
| facebook/opt-125m | Q8 | Fits comfortably | 64.81 tok/sEstimated | 7GB (have 16GB) |
| facebook/opt-125m | FP16 | Fits (tight) | 40.18 tok/sEstimated | 15GB (have 16GB) |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Q4 | Fits comfortably | 126.51 tok/sEstimated | 1GB (have 16GB) |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | Q8 | Fits comfortably | 82.98 tok/sEstimated | 1GB (have 16GB) |
| TinyLlama/TinyLlama-1.1B-Chat-v1.0 | FP16 | Fits comfortably | 46.65 tok/sEstimated | 2GB (have 16GB) |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q4 | Fits comfortably | 107.81 tok/sEstimated | 4GB (have 16GB) |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | Q8 | Fits comfortably | 69.20 tok/sEstimated | 7GB (have 16GB) |
| trl-internal-testing/tiny-Qwen2ForCausalLM-2.5 | FP16 | Fits (tight) | 34.32 tok/sEstimated | 15GB (have 16GB) |
| Qwen/Qwen3-4B-Instruct-2507 | Q4 | Fits comfortably | 94.82 tok/sEstimated | 2GB (have 16GB) |
| Qwen/Qwen3-4B-Instruct-2507 | Q8 | Fits comfortably | 68.96 tok/sEstimated | 4GB (have 16GB) |
| Qwen/Qwen3-4B-Instruct-2507 | FP16 | Fits comfortably | 37.22 tok/sEstimated | 9GB (have 16GB) |
| meta-llama/Llama-3.2-1B-Instruct | Q4 | Fits comfortably | 110.00 tok/sEstimated | 1GB (have 16GB) |
| meta-llama/Llama-3.2-1B-Instruct | Q8 | Fits comfortably | 88.00 tok/sEstimated | 1GB (have 16GB) |
| meta-llama/Llama-3.2-1B-Instruct | FP16 | Fits comfortably | 47.10 tok/sEstimated | 2GB (have 16GB) |
| openai/gpt-oss-120b | Q4 | Not supported | 19.26 tok/sEstimated | 59GB (have 16GB) |
| openai/gpt-oss-120b | Q8 | Not supported | 13.07 tok/sEstimated | 117GB (have 16GB) |
| openai/gpt-oss-120b | FP16 | Not supported | 6.73 tok/sEstimated | 235GB (have 16GB) |
| Qwen/Qwen2.5-3B-Instruct | Q4 | Fits comfortably | 112.75 tok/sEstimated | 2GB (have 16GB) |
| Qwen/Qwen2.5-3B-Instruct | Q8 | Fits comfortably | 86.24 tok/sEstimated | 3GB (have 16GB) |
| openai-community/gpt2 | Q4 | Fits comfortably | 100.24 tok/sEstimated | 4GB (have 16GB) |
Note: Performance estimates are calculated. Real results may vary. Methodology · Submit real data
Buy directly on Amazon with fast shipping and reliable customer service.
Essential accessories to pair with RX 9070 XT
Total Bundle Price
All items from Amazon
💡 Not ready to buy? Try cloud GPUs first
Test RX 9070 XT performance in the cloud before investing in hardware. Pay by the hour with no commitment.
Explore how RTX 5070 stacks up for local inference workloads.
Explore how RTX 4060 Ti 16GB stacks up for local inference workloads.
Explore how RX 6800 XT stacks up for local inference workloads.
Explore how RTX 4070 Super stacks up for local inference workloads.
Explore how RTX 3080 stacks up for local inference workloads.
RPG • 2020
RPG • 2023
Action RPG • 2023
RPG • 2023
Survival Horror • 2023
Action RPG • 2022
Action RPG • 2024
Action Adventure • 2025
Survival Horror • 2023
Action • 2022
Action Adventure • 2023
Action Adventure • 2019