Compare · GPUs
RX 9070 XT vs RTX 5070 Ti for local LLMs
Both fit the same 21 models, but the RTX 5070 Ti generates about 40% faster thanks to 896 GB/s of memory bandwidth versus 640 GB/s.
Both fit the same 21 models, but the RTX 5070 Ti generates about 40% faster thanks to 896 GB/s of memory bandwidth versus 640 GB/s.
Every model at 8K context with the best quantization that fits. Offload assumes 32 GB of system RAM. Green marks the better result. Rows where the two differ come first.
| Model | RX 9070 XT | RTX 5070 Ti |
|---|---|---|
| Qwen3 0.6B | Runs~650 tok/sQ8_0 | Runs~920 tok/sQ8_0 |
| Gemma 3 1B | Runs~390 tok/sQ8_0 | Runs~550 tok/sQ8_0 |
| Llama 3.2 1B | Runs~330 tok/sQ8_0 | Runs~460 tok/sQ8_0 |
| Llama 3.2 3B | Runs~120 tok/sQ8_0 | Runs~170 tok/sQ8_0 |
| Phi-4 Mini 3.8B | Runs~100 tok/sQ8_0 | Runs~140 tok/sQ8_0 |
| Qwen3 4B | Runs~98 tok/sQ8_0 | Runs~140 tok/sQ8_0 |
| Gemma 3 4B | Runs~91 tok/sQ8_0 | Runs~130 tok/sQ8_0 |
| Qwen3.5 4B | Runs~84 tok/sQ8_0 | Runs~120 tok/sQ8_0 |
| Mistral 7B v0.3 | Runs~55 tok/sQ8_0 | Runs~76 tok/sQ8_0 |
| Qwen2.5-Coder 7B | Runs~52 tok/sQ8_0 | Runs~72 tok/sQ8_0 |
| Llama 3.1 8B | Runs~49 tok/sQ8_0 | Runs~69 tok/sQ8_0 |
| DeepSeek-R1 Distill 8B | Runs~49 tok/sQ8_0 | Runs~69 tok/sQ8_0 |
| Qwen3 8B | Runs~48 tok/sQ8_0 | Runs~67 tok/sQ8_0 |
| Qwen3.5 9B | Runs~42 tok/sQ8_0 | Runs~59 tok/sQ8_0 |
| Gemma 4 12B | Runs~42 tok/sQ6_K | Runs~59 tok/sQ6_K |
| Gemma 3 12B | Runs~42 tok/sQ6_K | Runs~58 tok/sQ6_K |
| Mistral Nemo 12B | Runs~42 tok/sQ6_K | Runs~58 tok/sQ6_K |
| Phi-4 14B | Runs~40 tok/sQ5_K_M | Runs~56 tok/sQ5_K_M |
| Qwen3 14B | Runs~40 tok/sQ5_K_M | Runs~55 tok/sQ5_K_M |
| DeepSeek-R1 Distill 14B | Runs~40 tok/sQ5_K_M | Runs~55 tok/sQ5_K_M |
| gpt-oss 20B | Runs~86 tok/sQ4_K_M | Runs~120 tok/sQ4_K_M |
| Codestral 22B | Offload~4.5 tok/sQ4_K_M | Offload~6.2 tok/sQ4_K_M |
| Mistral Small 3 24B | Offload~4.0 tok/sQ4_K_M | Offload~5.5 tok/sQ4_K_M |
| Devstral Small 24B | Offload~4.1 tok/sQ4_K_M | Offload~5.6 tok/sQ4_K_M |
| Gemma 4 26B-A4B (MoE) | Offload~11 tok/sQ4_K_M | Offload~15 tok/sQ4_K_M |
| Gemma 3 27B | Offload~3.3 tok/sQ4_K_M | Offload~4.3 tok/sQ4_K_M |
| Qwen3.6 27B | Offload~3.4 tok/sQ4_K_M | Offload~4.6 tok/sQ4_K_M |
| Qwen3 30B-A3B (MoE) | Offload~13 tok/sQ4_K_M | Offload~16 tok/sQ4_K_M |
| Qwen3-Coder 30B-A3B | Offload~13 tok/sQ4_K_M | Offload~16 tok/sQ4_K_M |
| Gemma 4 31B | Offload~2.8 tok/sQ4_K_M | Offload~3.6 tok/sQ4_K_M |
| Qwen3 32B | Offload~2.7 tok/sQ4_K_M | Offload~3.5 tok/sQ4_K_M |
| DeepSeek-R1 Distill 32B | Offload~2.7 tok/sQ4_K_M | Offload~3.5 tok/sQ4_K_M |
| Qwen2.5-Coder 32B | Offload~2.7 tok/sQ4_K_M | Offload~3.5 tok/sQ4_K_M |
| Qwen3.6 35B-A3B (MoE) | Offload~13 tok/sQ4_K_M | Offload~16 tok/sQ4_K_M |
| Llama 3.3 70B | Won’t fit | Won’t fit |
| DeepSeek-R1 Distill 70B | Won’t fit | Won’t fit |
| Llama 4 Scout 109B-A17B | Won’t fit | Won’t fit |
| gpt-oss 120B | Won’t fit | Won’t fit |
| Qwen3 235B-A22B (MoE) | Won’t fit | Won’t fit |
Which should you buy?
The RTX 5070 Ti is the better card for local AI. Both fit the same 21 models, but the RTX 5070 Ti generates about 40% faster thanks to 896 GB/s of memory bandwidth versus 640 GB/s.
Software matters as much as memory. NVIDIA cards work out of the box with Ollama, LM Studio and llama.cpp through CUDA. AMD cards run through ROCm or Vulkan and Intel Arc through Vulkan or IPEX-LLM; both work, but setup takes longer and new features usually land on NVIDIA first.
See every result on its own page: RX 9070 XT and RTX 5070 Ti.
Questions
Is the RX 9070 XT or the RTX 5070 Ti better for running LLMs?
The RTX 5070 Ti is the better card for local AI. Both fit the same 21 models, but the RTX 5070 Ti generates about 40% faster thanks to 896 GB/s of memory bandwidth versus 640 GB/s.
Which models can the RX 9070 XT run that the RTX 5070 Ti can't?
None. Both cards fit the same models; the difference is speed.
Does the RX 9070 XT or RTX 5070 Ti need anything else to run local AI?
Only the software. Install Ollama or LM Studio, pick a model that fits in VRAM, and it runs on the GPU. 32 GB of system RAM helps when a model spills over, but that path is much slower.
More comparisons
Estimates from memory size and bandwidth at 8K context. Real speed varies with software, drivers and settings.