← All GPUs
Intel · Consumer · Xe2 (Battlemage BMG-G21)
Intel Arc B580
12 GB only; PCIe 4.0 x8.
VRAM
12 GB
GDDR6
Memory bandwidth
456 GB/s
FP16 compute
not published
Price
~$249
new, maker or suggested price
What fits at 4-bit
The 4-bit file plus 8,192 tokens of context and 1.5 GB of overhead, against the VRAM of one card. Speeds are measured medians where someone has published a run at that card count; a blank speed means no run, not a slow one. Machine pages carry estimates for the gaps.
| Model | Needs | 1 card (12 GB) |
|---|---|---|
| Qwen3.5 4B4B · Q4_K_M | 4.3 GB | fits |
| Llama 3.1 8B Instruct8B · Q4_K_M | 7.1 GB | fits |
| Qwen3 8B8.2B · Q4_K_M | 7.3 GB | fits |
| Gemma 4 E4B IT8B · Q4_K_M | 6.5 GB | 64 tok/smedian of 4 runs |
| Granite 4.2 8B8.8B · Q4_K_M | 7.9 GB | fits |
| Qwen3.5 9B9B · Q4_K_M | 7.0 GB | fits |
| Gemma 4 12B IT11.95B · Q4_K_M | 8.2 GB | 41 tok/smedian of 1 run |
| Ministral 3 14B Reasoning (2512)13.9B · Q4_K_M | 10.4 GB | fits |
| Phi-414.7B · Q4_K_M | 11.3 GB | fits |
Measured speeds
One card
| Model | Quant | Backend | Generation | Prompt | Context | Source |
|---|---|---|---|---|---|---|
| Gemma 4 12B IT | Q4_KM | llama.cpp (build not stated) SYCL | 41 tok/s | – | 0 | llama.cpp benchmark discussion2026-06 |
| Gemma 4 12B IT | Q4_KM | llama.cpp (build not stated) SYCL | – | 1,320 tok/s512-token prompt | 0 | llama.cpp benchmark discussion2026-06 |
| Gemma 4 E4B IT | Q4_0 | llama.cpp (build not stated) SYCL | 82 tok/s | – | 0 | llama.cpp benchmark discussion2026-06 |
| Gemma 4 E4B IT | Q4_K_M | Ollama 0.33.0 (llama.cpp Vulkan, Mesa 26.1.7) | 67 tok/s | – | 0 | llama.cpp benchmark discussion2026-08 |
| Gemma 4 E4B IT | Q4_K_M | Ollama (version not stated, Vulkan, Mesa 26.2.3) | 61 tok/s | 270 tok/s | 0 | llama.cpp benchmark discussion2026-09 |
| Gemma 4 E4B IT | Q4_K_M | Ollama 0.33.0 (llama.cpp Vulkan, Mesa 26.0.8) | 30 tok/s | – | 0 | llama.cpp benchmark discussion2026-08 |
| Gemma 4 E4B IT | Q4_0 | llama.cpp (build not stated) SYCL | – | 2,763 tok/s512-token prompt | 0 | llama.cpp benchmark discussion2026-06 |
Specifications
- Memory
- 12 GB GDDR6
- FP16 basis
- Intel publishes 233 INT8 TOPS (peak); no FP16 figure.
- Board power
- 190 W
- PCIe
- Gen 4 x8
- NVLink
- No
- Cooling
- Open-air
- Software
- oneAPI/SYCL, Vulkan
- Released
- 2024
- Availability
- sold new
- Price basis
- Intel 'Recommended Customer Price' on its spec page.
- Maker page
- intel.com
Sources
- intel.com: 12 GB GDDR6 192-bit, 456 GB/s, 190 W, PCIe 4.0 x8, 233 TOPS, Q4'24, $249
- raw.githubusercontent.com: llama.cpp build docs: SYCL backend supports Intel Arc; HIP backend for ROCm GPUs; Vulkan backend