AMD Radeon RX 9070
16 GB · 640 GB/s · 220 W
In short
The AMD Radeon RX 9070 has 16 GB of GDDR6 memory with 640 GB/s of bandwidth, which sets how fast it generates text. At 32k context it runs 35 of the 125 open models we track fully in memory; the most capable is gpt-oss 20B at MXFP4, at about 65 tok/s (estimated). We have no current price for the AMD Radeon RX 9070 in the US.
Updated . Speeds are estimates; how they are worked out.
Prices
Every retailer we track in the US, cheapest first. Buy links may earn us a commission.
No retailer in this market lists it right now.
What it runs
Every model in the catalogue, with the best quant for this hardware. Estimates, not measurements.
| Model | Quant | Fits | Memory | Max context | Decode @32k | Prefill | Good for |
|---|---|---|---|---|---|---|---|
| gpt-oss 20B OpenAI | MXFP4 | fits | 13.0 GB | 106k128k with q8 cache | 65tok/sest. | 1086 tok/s | chat, coding, agents, long documents |
| Gemma 4 12B | Q8_0 | fits | 13.6 GB | 106k171k with q8 cache | 26tok/sest. | 473 tok/s | chat, coding, long documents |
| Qwen 3.5 9B Alibaba Qwen | Q8_0 | fits | 10.9 GB | 147k243k with q8 cache | 33tok/sest. | 589 tok/s | chat, coding, long documents |
| Qwen3 14B Alibaba Qwen | Q4_K_M | fits | 14.4 GB | 37k40k with q8 cache | 25tok/sest. | 294 tok/s | chat, coding |
| Ministral 3 14B Mistral AI | Q5_K_M | fits | 14.9 GB | 33k61k with q8 cache | 24tok/sest. | 327 tok/s | chat, coding |
| DeepSeek R1 0528 Qwen3 8B DeepSeek | Q8_0 | fits | 13.6 GB | 42k78k with q8 cache | 26tok/sest. | 485 tok/s | chat, coding |
| Qwen3 8B Alibaba Qwen | Q8_0 | fits | 13.6 GB | 40k | 26tok/sest. | 485 tok/s | chat, coding |
| Qwen 3.5 4B Alibaba Qwen | Q8_0 | fits | 6.2 GB | 256k | 63tok/sest. | 1102 tok/s | chat, coding, long documents |
| Ministral 3 8B Mistral AI | Q8_0 | fits | 13.6 GB | 43k78k with q8 cache | 26tok/sest. | 469 tok/s | chat, coding |
| Granite 4.2 8B IBM | Q8_0 | fits | 14.7 GB | 35k64k with q8 cache | 24tok/sest. | 446 tok/s | chat, coding |
| Gemma 4 E4B | Q8_0 | fits | 9.5 GB | 128k | 38tok/sest. | 740 tok/s | chat, coding, long documents |
| LFM2.5 8B-A1B Liquid AI | Q8_0 | fits | 10.9 GB | 113k125k with q8 cache | 79tok/sest. | 2420 tok/s | chat, coding, long documents |
| Llama 3.1 8B Meta | Q6_K | fits | 11.1 GB | 63k114k with q8 cache | 32tok/sest. | 513 tok/s | chat, coding |
| Qwen2.5 0.5B Alibaba Qwen | Q8_0 | fits | 1.9 GB | 32k | 377tok/sest. | 5270 tok/s | chat, coding, long documents |
| Qwen3.5 0.8B Alibaba Qwen | Q8_0 | fits | 2.1 GB | 256k | 285tok/sest. | 4951 tok/s | chat, coding, agents, long documents |
| Qwen3.5 2B Alibaba Qwen | Q5_K_M | fits | 2.7 GB | 256k | 190tok/sest. | 2360 tok/s | chat, coding, agents, long documents |
| DeepSeek R1 Distill Qwen 1.5B DeepSeek | Q8_0 | fits | 3.6 GB | 128k | 124tok/sest. | 1983 tok/s | chat, coding, long documents |
| MiniCPM5 2B openbmb | Q8_0 | fits | 4.8 GB | 128k | 86tok/sest. | 1184 tok/s | chat, coding, agents, long documents |
| Spark X2.5 4B XHToken | mlx-4bit | fits | 5.0 GB | 280k470k with q8 cache | 82tok/sest. | 1163 tok/s | chat, coding, agents, long documents |
| LFM2.5 2.6B Liquid AI | Q8_0 | fits | 5.5 GB | 128k | 72tok/sest. | 1341 tok/s | chat, coding, agents, long documents |
| LFM2.5 VL 3B Liquid AI | Q8_0 | fits | 5.5 GB | 32k | 72tok/sest. | 1230 tok/s | chat, coding, agents, long documents |
| Kimi VL A3B Thinking 2506 Moonshot AI | Q4_K_M | fits | 11.8 GB | 128k | 72tok/sest. | 1313 tok/s | chat, coding, long documents |
| Mellum2.1 12B A2.5B Thinking JetBrains | Q8_0 | fits | 13.5 GB | 118k128k with q8 cache | 67tok/sest. | 1747 tok/s | chat, coding, agents, long documents |
| FrogNano 4B 2609 Microsoft | Q8_0 | fits | 6.3 GB | 256k | 61tok/sest. | 1102 tok/s | chat, coding, agents, long documents |
| Kimi VL A3B Moonshot AI | mlx-6bit | fits | 14.8 GB | 42k70k with q8 cache | 60tok/sest. | 1313 tok/s | chat, coding, agents, long documents |
| granite vision 4.1 4b IBM | Q8_0 | fits | 6.9 GB | 128k | 56tok/sest. | 859 tok/s | chat, coding, long documents |
| granite 4.2 3b IBM | Q8_0 | fits | 7.1 GB | 128k | 54tok/sest. | 901 tok/s | chat, coding, agents, long documents |
| Ministral 3 3B Instruct 2512 Mistral AI | Q8_0 | fits | 7.6 GB | 102k183k with q8 cache | 49tok/sest. | 861 tok/s | chat, coding, agents |
| GLM 4 9B 0414 Z.ai | mlx-5bit | fits | 9.9 GB | 32k | 37tok/sest. | 428 tok/s | chat, coding |
| DeepSeek R1 Distill Qwen 7B DeepSeek | Q8_0 | fits | 10.3 GB | 113k128k with q8 cache | 35tok/sest. | 579 tok/s | chat, coding |
| GLM 4.6V Flash Z.ai | Q8_0 | fits | 11.6 GB | 113k128k with q8 cache | 31tok/sest. | 403 tok/s | chat, coding |
| DeepSeek R1 Distill Llama 8B DeepSeek | Q8_0 | fits | 12.9 GB | 49k89k with q8 cache | 27tok/sest. | 513 tok/s | chat, coding |
| GLM 4.1V 9B Thinking Z.ai | mlx-8bit | fits | 13.2 GB | 64k | 27tok/sest. | 403 tok/s | chat, coding |
| Ministral 3 8B Reasoning 2512 Mistral AI | Q8_0 | fits | 13.6 GB | 43k78k with q8 cache | 26tok/sest. | 469 tok/s | chat, coding |
| Ministral 3 14B Reasoning 2512 Mistral AI | Q5_K_M | fits | 14.9 GB | 33k61k with q8 cache | 24tok/sest. | 327 tok/s | chat, coding |
| Qwen 3.8 27B Alibaba Qwen | UD-Q4_K_M | offload | 18.5 GB | – | 9tok/sest. | 181 tok/s | – |
| Qwen 3.5 122B-A10B Alibaba Qwen | UD-Q2_K_XL | offload | 40.8 GB | – | 11tok/sest. | 293 tok/s | – |
| Qwen 3.6 27B Alibaba Qwen | Q4_K_M | offload | 18.8 GB | – | 9tok/sest. | 179 tok/s | – |
| Gemma 4 31B | Q4_K_M | offload | 21.3 GB | – | 6tok/sest. | 149 tok/s | – |
| Mistral Small 4 119B Mistral AI | UD-Q2_K_XL | offload | 39.1 GB | – | 15tok/sest. | 331 tok/s | chat |
| Qwen 3.6 35B-A3B Alibaba Qwen | MXFP4 | offload | 21.9 GB | – | 30tok/sest. | 1176 tok/s | chat, coding |
| Qwen 3.5 27B Alibaba Qwen | Q4_K_M | offload | 18.7 GB | – | 9tok/sest. | 180 tok/s | – |
| Gemma 4 26B-A4B | MXFP4 | offload | 17.2 GB | – | 36tok/sest. | 1198 tok/s | chat, coding, agents, long documents |
| Qwen 3.5 35B-A3B Alibaba Qwen | Q4_K_M | offload | 22.2 GB | – | 29tok/sest. | 1168 tok/s | chat, coding |
| GLM 4.7 Flash Z.ai | Q4_K_M | offload | 19.7 GB | – | 19tok/sest. | 650 tok/s | chat |
| Devstral Small 2 24B Mistral AI | Q4_K_M | offload | 19.3 GB | 6k11k with q8 cache | 8tok/sest. | 186 tok/s | chat |
| Nemotron 3.5 Lightning 30B-A3B NVIDIA | MXFP4 | offload | 22.8 GB | – | 29tok/sest. | 1312 tok/s | chat, coding, long documents |
| Kimi Linear 48B-A3B Moonshot AI | Q4_K_M | offload | 29.3 GB | – | 26tok/sest. | 880 tok/s | chat, coding, long documents |
| Qwen3 Coder 30B-A3B Alibaba Qwen | Q4_K_M | offload | 21.3 GB | – | 16tok/sest. | 549 tok/s | chat |
| Qwen3 32B Alibaba Qwen | Q4_K_M | offload | 27.4 GB | – | 4tok/sest. | 92 tok/s | – |
| Gemma 3 27B | Q4_K_M | offload | 19.3 GB | – | 8tok/sest. | 187 tok/s | – |
| Mistral Small 3.2 24B Mistral AI | Q4_K_M | offload | 19.3 GB | 6k11k with q8 cache | 8tok/sest. | 186 tok/s | chat |
| DeepSeek R1 Distill Qwen 32B DeepSeek | Q4_K_M | offload | 27.5 GB | – | 4tok/sest. | 105 tok/s | – |
| Phi-4 14B Microsoft | Q4_K_M | offload | 15.5 GB | 16k | 20tok/sest. | 292 tok/s | chat, coding |
| LFM2.5 24B A2B Liquid AI | Q4_K_M | offload | 17.8 GB | 1k2k with q8 cache | 29tok/sest. | 1224 tok/s | chat, coding |
| DeepSeek R1 Distill Qwen 14B DeepSeek | Q4_K_M | offload | 15.4 GB | 31k57k with q8 cache | 22tok/sest. | 275 tok/s | chat, coding |
| Muse Glimmer 30B meta-models | Q4_K_M | offload | 17.6 GB | – | 11tok/sest. | 179 tok/s | – |
| Devstral Small 2507 Mistral AI | Q4_K_M | offload | 19.3 GB | 6k11k with q8 cache | 8tok/sest. | 189 tok/s | chat |
| Magistral Small 2506 Mistral AI | Q4_K_M | offload | 19.3 GB | 6k11k with q8 cache | 8tok/sest. | 189 tok/s | chat |
| Magistral Small 2507 Mistral AI | Q4_K_M | offload | 19.3 GB | 6k11k with q8 cache | 8tok/sest. | 189 tok/s | chat |
| Magistral Small 2509 Mistral AI | Q4_K_M | offload | 19.3 GB | 6k11k with q8 cache | 8tok/sest. | 186 tok/s | chat |
| Qwen3.8 27B OBLITERATED OBLITERATUS | mlx-4bit | offload | 22.4 GB | – | 5tok/sest. | 163 tok/s | – |
| granite 4.2 30b IBM | Q4_K_M | offload | 25.8 GB | – | 4tok/sest. | 124 tok/s | – |
| DeepSeek V4 Flash DeepSeek | UD-Q2_K_XL | won't run | 93.9 GB | – | – | – | – |
| GLM 5.3 Flash Z.ai | UD-Q2_K_XL | won't run | 102.7 GB | – | – | – | – |
| Qwen 3.8 Flash Next Alibaba Qwen | UD-Q2_K_XL | won't run | 75.3 GB | – | – | – | – |
| MiniMax M2.7 MiniMax | UD-Q2_K_XL | won't run | 78.9 GB | – | – | – | – |
| Nemotron 3 Super 120B-A12B NVIDIA | UD-Q2_K_XL | won't run | 52.3 GB | – | – | – | – |
| Qwen3 Coder Next Alibaba Qwen | Q4_K_M | won't run | 47.0 GB | – | – | – | – |
| gpt-oss 120B OpenAI | Q4_K_M | won't run | 60.6 GB | – | – | – | – |
| GLM 4.5 Air Z.ai | Q2_K | won't run | 48.9 GB | – | – | – | – |
| Qwen3 Next 80B-A3B Alibaba Qwen | Q4_K_M | won't run | 47.0 GB | – | – | – | – |
| Llama 3.3 70B Meta | Q4_K_M | won't run | 50.6 GB | – | – | – | – |
| DeepSeek R1 DeepSeek | Q4_K_M | won't run | 379.8 GB | – | – | – | – |
| DeepSeek R1 0528 DeepSeek | Q4_K_M | won't run | 380.3 GB | – | – | – | – |
| DeepSeek R1 Distill Llama 70B DeepSeek | Q4_K_M | won't run | 50.6 GB | – | – | – | – |
| DeepSeek R1 Zero DeepSeek | Q4_K_M | won't run | 379.8 GB | – | – | – | – |
| DeepSeek V3 DeepSeek | Q4_K_M | won't run | 379.8 GB | – | – | – | – |
| DeepSeek V3 0324 DeepSeek | Q4_K_M | won't run | 379.8 GB | – | – | – | – |
| DeepSeek V3.1 DeepSeek | Q4_K_M | won't run | 380.7 GB | – | – | – | – |
| DeepSeek V3.1 Terminus DeepSeek | Q4_K_M | won't run | 380.7 GB | – | – | – | – |
| DeepSeek V3.2 DeepSeek | Q4_K_M | won't run | 380.7 GB | – | – | – | – |
| DeepSeek V4 Flash 0731 DeepSeek | MXFP4 | won't run | 149.3 GB | – | – | – | – |
| DeepSeek V4 Flash Vision Exp DeepSeek | MXFP4 | won't run | 158.1 GB | – | – | – | – |
| Devstral 2 123B Instruct 2512 Mistral AI | Q4_K_M | won't run | 81.7 GB | – | – | – | – |
| GLM 4.5 Z.ai | Q4_K_M | won't run | 214.1 GB | – | – | – | – |
| GLM 4.5V Z.ai | Q4_K_M | won't run | 66.0 GB | – | – | – | – |
| GLM 4.6 Z.ai | Q4_K_M | won't run | 213.3 GB | – | – | – | – |
| GLM 4.6V Z.ai | Q4_K_M | won't run | 72.3 GB | – | – | – | – |
| GLM 4.7 Z.ai | Q4_K_M | won't run | 214.1 GB | – | – | – | – |
| GLM 5 Z.ai | MXFP4 | won't run | 386.1 GB | – | – | – | – |
| GLM 5.1 Z.ai | MXFP4 | won't run | 423.7 GB | – | – | – | – |
| GLM 5.2 Z.ai | UD-Q4_K_M | won't run | 437.6 GB | – | – | – | – |
| GLM 5.3 Z.ai | Q8_0 | won't run | 750.1 GB | – | – | – | – |
| Kimi Dev 72B Moonshot AI | Q6_K | won't run | 70.9 GB | – | – | – | – |
| Kimi K2 Moonshot AI | Q4_K_M | won't run | 581.3 GB | – | – | – | – |
| Kimi K2 Instruct 0905 Moonshot AI | Q4_K_M | won't run | 581.7 GB | – | – | – | – |
| Kimi K2 Thinking Moonshot AI | Q4_K_M | won't run | 581.7 GB | – | – | – | – |
| Kimi K2.5 Moonshot AI | Q4_K_M | won't run | 581.7 GB | – | – | – | – |
| MiniMax M2 MiniMax | Q4_K_M | won't run | 137.6 GB | – | – | – | – |
| MiniMax M2.1 MiniMax | Q4_K_M | won't run | 137.6 GB | – | – | – | – |
| MiniMax M2.5 MiniMax | MXFP4 | won't run | 124.0 GB | – | – | – | – |
| MiniMax M3 MiniMax | MXFP4 | won't run | 244.4 GB | – | – | – | – |
| Mistral Medium 3.5 128B Mistral AI | Q4_K_M | won't run | 81.7 GB | – | – | – | – |
Specifications
From the manufacturer's published specifications; sources are linked where we have them.
- GPU
- Radeon RX 9070
- Memory
- 16 GB GDDR6
- Memory bus
- 256-bit
- Memory bandwidth
- 640 GB/s
- FP16 compute
- 36.1 TFLOPS
- Software
- ROCm
- Board power
- 220 W
- PCIe generation
- Gen 5
- PCIe lanes
- x16
Compare with
Questions
- How much memory does the AMD Radeon RX 9070 have for AI models?
- The AMD Radeon RX 9070 has 16 GB of GDDR6 memory with 640 GB/s of bandwidth, which sets how fast it generates text.
- What is the best model the AMD Radeon RX 9070 can run locally?
- gpt-oss 20B (MXFP4) fits fully in its memory at 32k context and generates at about 65 tok/s, an estimate from its memory bandwidth. 35 of the 125 models we track fit at that context.
- Can the AMD Radeon RX 9070 run DeepSeek V4 Flash?
- No: even at UD-Q2_K_XL, DeepSeek V4 Flash needs about 94 GB at 32k context, more than the AMD Radeon RX 9070 and system RAM can hold.
- How much does the AMD Radeon RX 9070 cost in the US?
- We have no current price for the AMD Radeon RX 9070 in the US.
Speeds on this page are for 32k context. See the best hardware for local AI at every budget.