2x Radeon RX 7900 XTX vs 2x Intel Arc Pro B60
The 2x Radeon RX 7900 XTX has 2.1× the memory bandwidth (960 against 456 GB/s, per card for multi-GPU builds), which sets the ceiling on generation speed. The cheapest 2x Radeon RX 7900 XTX lists for about $1,270 more than the cheapest 2x Intel Arc Pro B60. Rows where the two differ are tinted.
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
Multi-GPU
2x Intel Arc Pro B60
The cheapest new 48 GB build with a warranty: two 24 GB Battlemage workstation cards at 200 W each. Bandwidth (456 GB/s per card) is half a 3090's, so expect slower token generation.
gpt-oss-20b, 4-bit: ~57–83 tok/sestimate
| Specification | 2x Radeon RX 7900 XTX | 2x Intel Arc Pro B60 |
|---|---|---|
| Kind | Multi-GPU · DIY | Multi-GPU · DIY |
| Chip or GPUs | 2× Radeon RX 7900 XTX | 2× Arc Pro B60 |
| 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 | 960 GB/s per card | 456 GB/s per card |
| FP16 compute | 123 TFLOPS per card | not published |
| Price | ~$5,637 (sum of parts) | ~$4,367 (sum of parts) |
| Largest at 4-bit, 8k context | Llama 3.3 70B Instruct | Llama 3.3 70B Instruct |
| Software | ROCm, Vulkan | oneAPI/SYCL, Vulkan |
| Power | 1500 W power supply | 1000 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 Radeon RX 7900 XTX | 2x Intel Arc Pro B60 |
|---|---|---|
| Llama 3.3 70B Instruct70B · 39.6 GB file | ~11–17 tok/sestimate | ~5.3–8.3 tok/sestimate |
| Nemotron 3.5 Lightning 30B-A3B30B, 3B active · 23.7 GB file | ~94–137 tok/sestimate | ~45–65 tok/sestimate |
| Qwen3.6 35B-A3B35B, 3B active · 20.8 GB file | ~125–183 tok/sestimate | ~59–87 tok/sestimate |
| Ornith 1.5 35B-A3B35B, 3B active · 20.2 GB file | ~128–187 tok/sestimate | ~61–89 tok/sestimate |
| Qwen3 Coder 30B-A3B Instruct30.5B, 3.3B active · 17.3 GB file | ~119–174 tok/sestimate | ~57–83 tok/sestimate |
| Gemma 4 31B IT30.7B · 17.1 GB file | ~26–40 tok/sestimate | ~12–19 tok/sestimate |
| Granite 4.2 30B29.3B · 16.8 GB file | ~26–41 tok/sestimate | ~12–19 tok/sestimate |
| Qwen3.8 27B27B · 16.2 GB file | ~27–42 tok/sestimate | ~13–20 tok/sestimate |
| Gemma 4 26B-A4B IT25.2B, 3.8B active · 15.9 GB file | ~93–136 tok/sestimate | ~44–64 tok/sestimate |
| Qwen3.6 27B27B · 15.7 GB file | ~28–44 tok/sestimate | ~13–21 tok/sestimate |
| Devstral Small 2 24B (2512)24B · 13.3 GB file | ~33–52 tok/sestimate | ~16–24 tok/sestimate |
| gpt-oss-20b21B, 3.6B active · 10.8 GB file | ~120–175 tok/sestimate | ~57–83 tok/sestimate |
| Phi-414.7B · 8.3 GB file | ~53–83 tok/sestimate | ~25–39 tok/sestimate |
| Ministral 3 14B Reasoning (2512)13.9B · 7.7 GB file | ~57–90 tok/sestimate | ~27–43 tok/sestimate |
| Gemma 4 12B IT11.95B · 6.6 GB file | ~66–104 tok/sestimate | ~31–49 tok/sestimate |
| Qwen3.5 9B9B · 5.3 GB file | ~83–130 tok/sestimate | ~39–62 tok/sestimate |
| Granite 4.2 8B8.8B · 5.2 GB file | ~85–133 tok/sestimate | ~40–63 tok/sestimate |
| Gemma 4 E4B IT8B · 5.0 GB file | ~155–243 tok/sestimate | ~74–115 tok/sestimate |
| Qwen3 8B8.2B · 4.7 GB file | ~94–147 tok/sestimate | ~44–70 tok/sestimate |
| Llama 3.1 8B Instruct8B · 4.6 GB file | ~96–150 tok/sestimate | ~45–71 tok/sestimate |
| Qwen3.5 4B4B · 2.6 GB file | ~172–270 tok/sestimate | ~82–128 tok/sestimate |
Full pages: 2x Radeon RX 7900 XTX · 2x Intel Arc Pro B60 · check a different quant or context length