← All GPUs
Intel · Workstation · Xe2 (Battlemage BMG-G21)
Intel Arc Pro B60
Cooler varies by partner board. Native x8 link, so x8/x8 splits cost it nothing. Intel's LLM Scaler containers target multi-GPU vLLM on Linux.
VRAM
24 GB
GDDR6 ECC
Memory bandwidth
456 GB/s
FP16 compute
not published
Price
~$599
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 1 and 2 cards. 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 (24 GB) | 2 cards (48 GB) |
|---|---|---|---|
| Qwen3.5 4B4B · Q4_K_M | 4.3 GB | 37 tok/smedian of 2 runs | fits |
| Llama 3.1 8B Instruct8B · Q4_K_M | 7.1 GB | fits | fits |
| Qwen3 8B8.2B · Q4_K_M | 7.3 GB | fits | fits |
| Gemma 4 E4B IT8B · Q4_K_M | 6.5 GB | fits | fits |
| Granite 4.2 8B8.8B · Q4_K_M | 7.9 GB | fits | fits |
| Qwen3.5 9B9B · Q4_K_M | 7.0 GB | 23 tok/smedian of 3 runs | fits |
| Gemma 4 12B IT11.95B · Q4_K_M | 8.2 GB | fits | fits |
| Ministral 3 14B Reasoning (2512)13.9B · Q4_K_M | 10.4 GB | fits | fits |
| Phi-414.7B · Q4_K_M | 11.3 GB | fits | fits |
| gpt-oss-20b21B, 3.6B active · MXFP4 (native) | 12.5 GB | fits | fits |
| Devstral Small 2 24B (2512)24B · Q4_K_M | 16.1 GB | fits | fits |
| Qwen3.6 27B27B · Q4_K_M | 17.7 GB | fits | fits |
| Gemma 4 26B-A4B IT25.2B, 3.8B active · Q4_K_M | 17.4 GB | fits | fits |
| Qwen3.8 27B27B · Q4_K_M | 18.2 GB | fits | fits |
| Granite 4.2 30B29.3B · Q4_K_M | 20.3 GB | fits | fits |
| Gemma 4 31B IT30.7B · Q4_K_M | 18.9 GB | fits | fits |
| Qwen3 Coder 30B-A3B Instruct30.5B, 3.3B active · Q4_K_M | 19.5 GB | fits | fits |
| Ornith 1.5 35B-A3B35B, 3B active · Q4_K_M | 21.9 GB | fits | fits |
| Qwen3.6 35B-A3B35B, 3B active · Q4_K_M | 22.4 GB | fits | fits |
| Nemotron 3.5 Lightning 30B-A3B30B, 3B active · Q4_K_M | 25.3 GB | too large | fits |
| Llama 3.3 70B Instruct70B · Q4_K_M | 43.6 GB | too large | fits |
| Qwen3 Coder Next80B, 3B active · Q4_K_M | 46.9 GB | too large | fits |
Machines built on it
- 2x Intel Arc Pro B60~$4,367
Measured speeds
One card
| Model | Quant | Backend | Generation | Prompt | Context | Source |
|---|---|---|---|---|---|---|
| Qwen3.5 4B | Q4_0 | llama.cpp b8175 (Vulkan) | 43 tok/s | 812 tok/s2,048-token prompt | 0 | owner report2026-03 |
| Qwen3.5 4B | Q4_K_M | llama.cpp b8175 (Vulkan) | 31 tok/s | 821 tok/s2,048-token prompt | 0 | owner report2026-03 |
| Qwen3.5 9B | Q4_K_XL (UD) | llama.cpp b8688 (SYCL) | 33 tok/s | 1,634 tok/s2,048-token prompt | 0 | owner report2026-04 |
| Qwen3.5 9B | Q4_K_XL (UD) | llama.cpp b8175 (SYCL) | 23 tok/s | 329 tok/s2,048-token prompt | 0 | owner report2026-03 |
| Qwen3.5 9B | Q4_K_XL (UD) | llama.cpp b8175 (Vulkan) | 18 tok/s | 560 tok/s2,048-token prompt | 0 | owner report2026-03 |
| Qwen3.5 9B | Q4_K_XL (UD) | llama.cpp b8175 (SYCL) | – | 317 tok/s16,384-token prompt | 0 | owner report2026-03 |
| Qwen3.5 9B | Q4_K_XL (UD) | llama.cpp b8175 (Vulkan) | – | 518 tok/s16,384-token prompt | 0 | owner report2026-03 |
| Qwen3.5 9B | Q4_K_XL (UD) | llama.cpp b8688 (SYCL) | – | 1,464 tok/s16,384-token prompt | 0 | owner report2026-04 |
Speculative decoding, several users, or long context
Kept apart because each measures a different workload from one person chatting near the start of a conversation.
| Model | Quant | Backend | Generation | Prompt | Context | Measured on | Source |
|---|---|---|---|---|---|---|---|
| gpt-oss-20b | MXFP4 | vLLM (intel/vllm docker, >=0.10.2 XPU)75 requests at once | 1,211 tok/s | – | 0 | Arc Pro B60 | vLLM project blog2025-11 |
| gpt-oss-20b | MXFP4 | vLLM (intel/vllm docker, >=0.10.2 XPU)38 requests at once | 819 tok/s | – | 0 | Arc Pro B60 | vLLM project blog2025-11 |
| gpt-oss-20b | MXFP4 | vLLM (intel/vllm docker, >=0.10.2 XPU)15 requests at once | 417 tok/s | – | 0 | Arc Pro B60 | vLLM project blog2025-11 |
Specifications
- Memory
- 24 GB GDDR6 ECC
- FP16 basis
- Intel publishes 197 INT8 TOPS (XMX, dense); no FP16 figure.
- Board power
- 200 W
- PCIe
- Gen 5 x8
- NVLink
- No
- Software
- oneAPI/SYCL, Vulkan
- Released
- 2025
- Availability
- sold new
- Price basis
- Reported MSRP / US listing of $599 (Intel does not publish a price on its spec page).
- Maker page
- intel.com
Sources
- intel.com: 24 GB GDDR6 192-bit, 456 GB/s, TBP 200 W (120-200 W), PCIe 5.0 x8, 197 INT8 TOPS, Q2'25
- intel.com: Intel Arc Pro B-series quick reference (Mar 2026): B70/B65/B60/B50 memory, bandwidth, INT8 TOPS, board power
- guru3d.com: $599 listing
- vllm.ai: vLLM multi-GPU support on Arc Pro B-series
- raw.githubusercontent.com: llama.cpp build docs: SYCL backend supports Intel Arc; HIP backend for ROCm GPUs; Vulkan backend