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 64k context it runs 31 of the 125 open models we track fully in memory; the most capable is gpt-oss 20B at MXFP4, at about 52 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 @64k | Prefill | Good for |
|---|---|---|---|---|---|---|---|
| gpt-oss 20B OpenAI | MXFP4 | fits | 14.0 GB | 106k128k with q8 cache | 52tok/sest. | 851 tok/s | chat, coding, agents, long documents |
| Gemma 4 12B | Q8_0 | fits | 14.3 GB | 106k171k with q8 cache | 25tok/sest. | 437 tok/s | chat, coding, long documents |
| Qwen 3.5 9B Alibaba Qwen | Q8_0 | fits | 12.1 GB | 147k243k with q8 cache | 30tok/sest. | 535 tok/s | chat, coding, long documents |
| DeepSeek R1 0528 Qwen3 8B DeepSeek | Q4_K_M | fits | 14.9 GB | 66k120k with q8 cache | 24tok/sest. | 354 tok/s | chat, coding |
| Qwen3 8B Alibaba Qwen | Q4_K_M | fits | 14.9 GB | 40k | 24tok/sest. | 354 tok/s | chat, coding |
| Qwen 3.5 4B Alibaba Qwen | Q8_0 | fits | 7.4 GB | 256k | 53tok/sest. | 928 tok/s | chat, coding, long documents |
| Ministral 3 8B Mistral AI | Q4_K_M | fits | 14.5 GB | 68k125k with q8 cache | 25tok/sest. | 350 tok/s | chat, coding |
| Gemma 4 E4B | Q8_0 | fits | 10.6 GB | 128k | 35tok/sest. | 701 tok/s | chat, coding, long documents |
| LFM2.5 8B-A1B Liquid AI | Q8_0 | fits | 12.6 GB | 113k125k with q8 cache | 49tok/sest. | 1497 tok/s | chat, coding, long documents |
| Llama 3.1 8B Meta | Q5_K_M | fits | 14.5 GB | 69k125k with q8 cache | 25tok/sest. | 380 tok/s | chat, coding |
| Qwen2.5 0.5B Alibaba Qwen | Q8_0 | fits | 2.4 GB | 32k | 263tok/sest. | 3319 tok/s | chat, coding, long documents |
| Qwen3.5 0.8B Alibaba Qwen | Q8_0 | fits | 2.7 GB | 256k | 215tok/sest. | 3763 tok/s | chat, coding, agents, long documents |
| Qwen3.5 2B Alibaba Qwen | Q5_K_M | fits | 3.3 GB | 256k | 156tok/sest. | 2052 tok/s | chat, coding, agents, long documents |
| DeepSeek R1 Distill Qwen 1.5B DeepSeek | Q8_0 | fits | 4.7 GB | 128k | 93tok/sest. | 1375 tok/s | chat, coding, long documents |
| MiniCPM5 2B openbmb | Q8_0 | fits | 6.3 GB | 128k | 64tok/sest. | 775 tok/s | chat, coding, agents, long documents |
| Spark X2.5 4B XHToken | mlx-4bit | fits | 6.3 GB | 280k470k with q8 cache | 64tok/sest. | 951 tok/s | chat, coding, agents, long documents |
| Mellum2.1 12B A2.5B Thinking JetBrains | Q8_0 | fits | 14.1 GB | 118k128k with q8 cache | 58tok/sest. | 1387 tok/s | chat, coding, agents, long documents |
| Kimi VL A3B Thinking 2506 Moonshot AI | Q4_K_M | fits | 12.9 GB | 128k | 53tok/sest. | 954 tok/s | chat, coding, long documents |
| FrogNano 4B 2609 Microsoft | Q8_0 | fits | 7.5 GB | 256k | 52tok/sest. | 928 tok/s | chat, coding, agents, long documents |
| LFM2.5 2.6B Liquid AI | Q8_0 | fits | 7.6 GB | 128k | 51tok/sest. | 940 tok/s | chat, coding, agents, long documents |
| LFM2.5 VL 3B Liquid AI | Q8_0 | fits | 7.6 GB | 32k | 51tok/sest. | 884 tok/s | chat, coding, agents, long documents |
| Kimi VL A3B Moonshot AI | mlx-5bit | fits | 14.1 GB | 95k128k with q8 cache | 50tok/sest. | 954 tok/s | chat, coding, agents, long documents |
| granite vision 4.1 4b IBM | Q8_0 | fits | 9.5 GB | 128k | 39tok/sest. | 590 tok/s | chat, coding, long documents |
| granite 4.2 3b IBM | Q8_0 | fits | 9.8 GB | 128k | 38tok/sest. | 609 tok/s | chat, coding, agents, long documents |
| Ministral 3 3B Instruct 2512 Mistral AI | Q8_0 | fits | 11.1 GB | 102k183k with q8 cache | 33tok/sest. | 583 tok/s | chat, coding, agents |
| GLM 4 9B 0414 Z.ai | mlx-5bit | fits | 11.4 GB | 32k | 32tok/sest. | 314 tok/s | chat, coding |
| DeepSeek R1 Distill Qwen 7B DeepSeek | Q8_0 | fits | 12.2 GB | 113k128k with q8 cache | 30tok/sest. | 445 tok/s | chat, coding |
| GLM 4.6V Flash Z.ai | Q8_0 | fits | 13.0 GB | 113k128k with q8 cache | 28tok/sest. | 300 tok/s | chat, coding |
| DeepSeek R1 Distill Llama 8B DeepSeek | Q5_K_M | fits | 14.5 GB | 69k125k with q8 cache | 25tok/sest. | 380 tok/s | chat, coding |
| Ministral 3 8B Reasoning 2512 Mistral AI | Q4_K_M | fits | 14.5 GB | 68k125k with q8 cache | 25tok/sest. | 350 tok/s | chat, coding |
| GLM 4.1V 9B Thinking Z.ai | mlx-8bit | fits | 14.6 GB | 64k | 24tok/sest. | 300 tok/s | chat, coding |
| Qwen 3.8 27B Alibaba Qwen | UD-Q4_K_M | offload | 20.6 GB | – | 7tok/sest. | 154 tok/s | – |
| Qwen 3.6 27B Alibaba Qwen | Q4_K_M | offload | 21.0 GB | – | 6tok/sest. | 152 tok/s | – |
| Gemma 4 31B | Q4_K_M | offload | 24.0 GB | – | 5tok/sest. | 130 tok/s | – |
| Mistral Small 4 119B Mistral AI | UD-Q2_K_XL | offload | 40.0 GB | – | 12tok/sest. | 233 tok/s | chat |
| Qwen 3.6 35B-A3B Alibaba Qwen | MXFP4 | offload | 22.7 GB | – | 22tok/sest. | 879 tok/s | chat, coding |
| Qwen 3.5 27B Alibaba Qwen | Q4_K_M | offload | 20.9 GB | – | 6tok/sest. | 153 tok/s | – |
| Gemma 4 26B-A4B | MXFP4 | offload | 18.0 GB | – | 27tok/sest. | 1023 tok/s | chat, coding, agents, long documents |
| Qwen 3.5 35B-A3B Alibaba Qwen | Q4_K_M | offload | 23.0 GB | – | 22tok/sest. | 873 tok/s | chat, coding |
| GLM 4.7 Flash Z.ai | Q4_K_M | offload | 21.5 GB | – | 12tok/sest. | 448 tok/s | chat |
| Devstral Small 2 24B Mistral AI | Q4_K_M | offload | 24.5 GB | 6k11k with q8 cache | 4tok/sest. | 138 tok/s | chat |
| Nemotron 3.5 Lightning 30B-A3B NVIDIA | MXFP4 | offload | 23.2 GB | – | 26tok/sest. | 1071 tok/s | chat, coding, long documents |
| Kimi Linear 48B-A3B Moonshot AI | Q4_K_M | offload | 29.7 GB | – | 23tok/sest. | 620 tok/s | chat, coding, long documents |
| Qwen3 Coder 30B-A3B Alibaba Qwen | Q4_K_M | offload | 24.5 GB | – | 8tok/sest. | 303 tok/s | chat |
| Qwen3 32B Alibaba Qwen | Q4_K_M | offload | 35.6 GB | – | 2tok/sest. | 61 tok/s | – |
| Gemma 3 27B | Q4_K_M | offload | 22.0 GB | – | 6tok/sest. | 166 tok/s | – |
| Mistral Small 3.2 24B Mistral AI | Q4_K_M | offload | 24.5 GB | 6k11k with q8 cache | 4tok/sest. | 138 tok/s | chat |
| DeepSeek R1 Distill Qwen 32B DeepSeek | Q4_K_M | offload | 35.7 GB | – | 2tok/sest. | 75 tok/s | – |
| Ministral 3 14B Mistral AI | Q4_K_M | offload | 18.8 GB | 41k75k with q8 cache | 9tok/sest. | 224 tok/s | chat, coding |
| Qwen3 14B Alibaba Qwen | Q4_K_M | offload | 19.6 GB | 37k40k with q8 cache | 8tok/sest. | 192 tok/s | chat, coding |
| Phi-4 14B Microsoft | Q4_K_M | offload | 22.0 GB | 16k | 6tok/sest. | 181 tok/s | chat, coding |
| Granite 4.2 8B IBM | Q4_K_M | offload | 16.3 GB | 57k103k with q8 cache | 15tok/sest. | 309 tok/s | chat, coding |
| LFM2.5 24B A2B Liquid AI | Q4_K_M | offload | 20.5 GB | 1k2k with q8 cache | 13tok/sest. | 717 tok/s | chat, coding |
| Muse Glimmer 30B meta-models | Q4_K_M | offload | 18.2 GB | – | 10tok/sest. | 167 tok/s | – |
| Ministral 3 14B Reasoning 2512 Mistral AI | Q4_K_M | offload | 18.8 GB | 41k75k with q8 cache | 9tok/sest. | 224 tok/s | chat, coding |
| DeepSeek R1 Distill Qwen 14B DeepSeek | Q4_K_M | offload | 21.5 GB | 31k57k with q8 cache | 6tok/sest. | 166 tok/s | chat, coding |
| Devstral Small 2507 Mistral AI | Q4_K_M | offload | 24.5 GB | 6k11k with q8 cache | 4tok/sest. | 140 tok/s | chat |
| Magistral Small 2506 Mistral AI | Q4_K_M | offload | 24.5 GB | 6k11k with q8 cache | 4tok/sest. | 140 tok/s | chat |
| Magistral Small 2507 Mistral AI | Q4_K_M | offload | 24.5 GB | 6k11k with q8 cache | 4tok/sest. | 140 tok/s | chat |
| Magistral Small 2509 Mistral AI | Q4_K_M | offload | 24.5 GB | 6k11k with q8 cache | 4tok/sest. | 138 tok/s | chat |
| Qwen3.8 27B OBLITERATED OBLITERATUS | mlx-4bit | offload | 24.6 GB | – | 4tok/sest. | 140 tok/s | – |
| granite 4.2 30b IBM | Q4_K_M | offload | 34.0 GB | – | 2tok/sest. | 89 tok/s | – |
| DeepSeek V4 Flash DeepSeek | UD-Q2_K_XL | won't run | 96.7 GB | – | – | – | – |
| GLM 5.3 Flash Z.ai | UD-Q2_K_XL | won't run | 103.2 GB | – | – | – | – |
| Qwen 3.8 Flash Next Alibaba Qwen | UD-Q2_K_XL | won't run | 76.2 GB | – | – | – | – |
| MiniMax M2.7 MiniMax | UD-Q2_K_XL | won't run | 86.8 GB | – | – | – | – |
| Qwen 3.5 122B-A10B Alibaba Qwen | UD-Q2_K_XL | won't run | 41.8 GB | – | – | – | – |
| Nemotron 3 Super 120B-A12B NVIDIA | UD-Q2_K_XL | won't run | 52.7 GB | – | – | – | – |
| Qwen3 Coder Next Alibaba Qwen | Q4_K_M | won't run | 47.9 GB | – | – | – | – |
| gpt-oss 120B OpenAI | Q4_K_M | won't run | 61.9 GB | – | – | – | – |
| GLM 4.5 Air Z.ai | Q2_K | won't run | 54.9 GB | – | – | – | – |
| Qwen3 Next 80B-A3B Alibaba Qwen | Q4_K_M | won't run | 47.9 GB | – | – | – | – |
| Llama 3.3 70B Meta | Q4_K_M | won't run | 60.8 GB | – | – | – | – |
| DeepSeek R1 DeepSeek | Q4_K_M | won't run | 382.1 GB | – | – | – | – |
| DeepSeek R1 0528 DeepSeek | Q4_K_M | won't run | 382.6 GB | – | – | – | – |
| DeepSeek R1 Distill Llama 70B DeepSeek | Q4_K_M | won't run | 60.8 GB | – | – | – | – |
| DeepSeek R1 Zero DeepSeek | Q4_K_M | won't run | 382.1 GB | – | – | – | – |
| DeepSeek V3 DeepSeek | Q4_K_M | won't run | 382.1 GB | – | – | – | – |
| DeepSeek V3 0324 DeepSeek | Q4_K_M | won't run | 382.1 GB | – | – | – | – |
| DeepSeek V3.1 DeepSeek | Q4_K_M | won't run | 383.0 GB | – | – | – | – |
| DeepSeek V3.1 Terminus DeepSeek | Q4_K_M | won't run | 383.0 GB | – | – | – | – |
| DeepSeek V3.2 DeepSeek | Q4_K_M | won't run | 383.0 GB | – | – | – | – |
| DeepSeek V4 Flash 0731 DeepSeek | MXFP4 | won't run | 152.2 GB | – | – | – | – |
| DeepSeek V4 Flash Vision Exp DeepSeek | MXFP4 | won't run | 161.0 GB | – | – | – | – |
| Devstral 2 123B Instruct 2512 Mistral AI | Q4_K_M | won't run | 92.9 GB | – | – | – | – |
| GLM 4.5 Z.ai | Q4_K_M | won't run | 225.8 GB | – | – | – | – |
| GLM 4.5V Z.ai | Q4_K_M | won't run | 71.9 GB | – | – | – | – |
| GLM 4.6 Z.ai | Q4_K_M | won't run | 225.0 GB | – | – | – | – |
| GLM 4.6V Z.ai | Q4_K_M | won't run | 78.3 GB | – | – | – | – |
| GLM 4.7 Z.ai | Q4_K_M | won't run | 225.8 GB | – | – | – | – |
| GLM 5 Z.ai | MXFP4 | won't run | 389.0 GB | – | – | – | – |
| GLM 5.1 Z.ai | MXFP4 | won't run | 426.6 GB | – | – | – | – |
| GLM 5.2 Z.ai | UD-Q4_K_M | won't run | 440.5 GB | – | – | – | – |
| GLM 5.3 Z.ai | Q8_0 | won't run | 753.0 GB | – | – | – | – |
| Kimi Dev 72B Moonshot AI | Q6_K | won't run | 81.1 GB | – | – | – | – |
| Kimi K2 Moonshot AI | Q4_K_M | won't run | 583.6 GB | – | – | – | – |
| Kimi K2 Instruct 0905 Moonshot AI | Q4_K_M | won't run | 584.0 GB | – | – | – | – |
| Kimi K2 Thinking Moonshot AI | Q4_K_M | won't run | 584.0 GB | – | – | – | – |
| Kimi K2.5 Moonshot AI | Q4_K_M | won't run | 584.0 GB | – | – | – | – |
| MiniMax M2 MiniMax | Q4_K_M | won't run | 145.5 GB | – | – | – | – |
| MiniMax M2.1 MiniMax | Q4_K_M | won't run | 145.5 GB | – | – | – | – |
| MiniMax M2.5 MiniMax | MXFP4 | won't run | 131.9 GB | – | – | – | – |
| MiniMax M3 MiniMax | MXFP4 | won't run | 248.4 GB | – | – | – | – |
| Mistral Medium 3.5 128B Mistral AI | Q4_K_M | won't run | 92.9 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 64k context and generates at about 52 tok/s, an estimate from its memory bandwidth. 31 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 97 GB at 64k 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 64k context. See the best hardware for local AI at every budget.