2x RTX 3090 (used) vs 2x Radeon RX 7900 XTX
The 2x Radeon RX 7900 XTX has slightly more memory bandwidth (960 against 936 GB/s, per card for multi-GPU builds), which sets the ceiling on generation speed. The 2x RTX 3090 (used) has no complete price: GPU and risers prices not included. Rows where the two differ are tinted.
Multi-GPU
2x RTX 3090 (used)
Two used 3090s give 48 GB for 70B-class models at 4-bit. The 3090 is the newest GeForce card with an NVLink connector, which only matters for tensor-parallel serving.
gpt-oss-20b, 4-bit: ~117–171 tok/sestimate
Multi-GPU
2x Radeon RX 7900 XTX
AMD's cheapest new 48 GB: two 24 GB RDNA 3 cards running llama.cpp on ROCm/HIP or Vulkan. Same VRAM as 2x 3090 without CUDA.
gpt-oss-20b, 4-bit: ~120–175 tok/sestimate
| Specification | 2x RTX 3090 (used) | 2x Radeon RX 7900 XTX |
|---|---|---|
| Kind | Multi-GPU · DIY | Multi-GPU · DIY |
| Chip or GPUs | 2× RTX 3090 | 2× Radeon RX 7900 XTX |
| Memory | 48 GB VRAM + 64 GB RAM | 48 GB VRAM + 64 GB RAM |
| Memory for models | 46.5 GB (VRAM across all cards) | 46.5 GB (VRAM across all cards) |
| Memory bandwidth | 936 GB/s per card | 960 GB/s per card |
| FP16 compute | 71.2 TFLOPS per card | 123 TFLOPS per card |
| Price | ~$3,639 + unpriced parts (GPU and risers prices not included) | ~$5,637 (sum of parts) |
| Largest at 4-bit, 8k context | Llama 3.3 70B Instruct | Llama 3.3 70B Instruct |
| Software | CUDA, Vulkan | ROCm, Vulkan |
| Power | 1500 W power supply | 1500 W power supply |
What each one runs
Every model in the catalog that at least one of the two holds at 4-bit with 8,192 tokens of context. Copper speeds are measured medians; outlined ones are estimates. Rows where only one fits, or where the speed differs by more than a quarter, are tinted.
| Model | 2x RTX 3090 (used) | 2x Radeon RX 7900 XTX |
|---|---|---|
| Llama 3.3 70B Instruct70B · 39.6 GB file | 20 tok/smedian of 1 run | ~11–17 tok/sestimate |
| Nemotron 3.5 Lightning 30B-A3B30B, 3B active · 23.7 GB file | ~91–133 tok/sestimate | ~94–137 tok/sestimate |
| Qwen3.6 35B-A3B35B, 3B active · 20.8 GB file | ~122–178 tok/sestimate | ~125–183 tok/sestimate |
| Ornith 1.5 35B-A3B35B, 3B active · 20.2 GB file | ~125–183 tok/sestimate | ~128–187 tok/sestimate |
| Qwen3 Coder 30B-A3B Instruct30.5B, 3.3B active · 17.3 GB file | ~116–169 tok/sestimate | ~119–174 tok/sestimate |
| Gemma 4 31B IT30.7B · 17.1 GB file | ~25–39 tok/sestimate | ~26–40 tok/sestimate |
| Granite 4.2 30B29.3B · 16.8 GB file | ~25–40 tok/sestimate | ~26–41 tok/sestimate |
| Qwen3.8 27B27B · 16.2 GB file | ~26–41 tok/sestimate | ~27–42 tok/sestimate |
| Gemma 4 26B-A4B IT25.2B, 3.8B active · 15.9 GB file | ~91–132 tok/sestimate | ~93–136 tok/sestimate |
| Qwen3.6 27B27B · 15.7 GB file | ~27–43 tok/sestimate | ~28–44 tok/sestimate |
| Devstral Small 2 24B (2512)24B · 13.3 GB file | ~32–50 tok/sestimate | ~33–52 tok/sestimate |
| gpt-oss-20b21B, 3.6B active · 10.8 GB file | ~117–171 tok/sestimate | ~120–175 tok/sestimate |
| Phi-414.7B · 8.3 GB file | ~52–81 tok/sestimate | ~53–83 tok/sestimate |
| Ministral 3 14B Reasoning (2512)13.9B · 7.7 GB file | ~56–87 tok/sestimate | ~57–90 tok/sestimate |
| Gemma 4 12B IT11.95B · 6.6 GB file | ~64–101 tok/sestimate | ~66–104 tok/sestimate |
| Qwen3.5 9B9B · 5.3 GB file | ~81–127 tok/sestimate | ~83–130 tok/sestimate |
| Granite 4.2 8B8.8B · 5.2 GB file | ~83–130 tok/sestimate | ~85–133 tok/sestimate |
| Gemma 4 E4B IT8B · 5.0 GB file | ~151–237 tok/sestimate | ~155–243 tok/sestimate |
| Qwen3 8B8.2B · 4.7 GB file | ~91–143 tok/sestimate | ~94–147 tok/sestimate |
| Llama 3.1 8B Instruct8B · 4.6 GB file | ~93–146 tok/sestimate | ~96–150 tok/sestimate |
| Qwen3.5 4B4B · 2.6 GB file | ~167–263 tok/sestimate | ~172–270 tok/sestimate |
Full pages: 2x RTX 3090 (used) · 2x Radeon RX 7900 XTX · check a different quant or context length