Product Overview
Intel Arc Pro B60 is a professional workstation GPU based on the Xe2 (Battlemage) architecture, released by Intel in May 2025 at COMPUTEX Taipei. It features 24GB GDDR6 memory, 12.28 TFLOPS FP32 compute, 160 XMX AI engines, and 197 TOPS INT8 compute. It supports multi-GPU expansion (up to 8 cards), providing a combined 192GB of memory, capable of running AI models up to 150 billion parameters. It is a cost-effective professional card for AI development, deep learning, and 3D rendering.
Strategic Position: In Intel's professional GPU product line, the Arc Pro B60 is the mid-range professional card for 2025-2026, positioned between the consumer-grade Arc B580 and high-end data center GPUs. Compared to the NVIDIA RTX 6000 Ada, it offers larger memory capacity (24GB vs 48GB, but at a lower price), making it the preferred choice for budget-sensitive AI developers.
Core Specifications
| Item | Parameter |
|---|
| Architecture | Intel Xe2 (Battlemage) |
| Process | TSMC 4nm (estimated) |
| Transistors | 19.6 billion |
| Die Size | 272 mm² |
| Xe Cores | 20 Xe2 cores |
| XMX AI Engines | 160 |
| FP32 | 12.28 TFLOPS |
| INT8 | 197 TOPS |
| Memory | 24 GB GDDR6 |
| Memory Bandwidth | 456 GB/s (192-bit) |
| TDP | 200W |
| PCIe | Gen 5 x16 (estimated) |
| ECC | Not supported |
| Release Date | May 2025 |
| Price | Approximately $500-$700 (estimated) |
Comparison with Competitors
| Metric | Intel Arc Pro B60 | NVIDIA RTX 6000 Ada | NVIDIA RTX 4090 |
|---|
| Architecture | Xe2 (Battlemage) | Ada Lovelace | Ada Lovelace |
| Memory | 24GB GDDR6 | 48GB GDDR6 ECC | 24GB GDDR6X |
| Bandwidth | 456 GB/s | 960 GB/s | 1,008 GB/s |
| FP32 | 12.28 TFLOPS | 91.1 TFLOPS | 82.6 TFLOPS |
| INT8 | 197 TOPS | 2,916 TOPS | 1,321 TOPS |
| TDP | 200W | 300W | 450W |
| ECC | Not supported | Supported | Not supported |
| Price | ~$600 | ~$6,800 | ~$1,600 |
Arc Pro B60 Advantages: Large memory capacity (24GB), low price (estimated $500-700), excellent multi-GPU expansion (up to 8 cards, 192GB total).
Multi-GPU Expansion: Project Battlematrix
| Dimension | Specification |
|---|
| Max GPU Count | 8 Arc Pro B60 GPUs |
| Total Memory | 192GB (24GB × 8) |
| Target Platform | Code-named "Project Battlematrix" configurable workstation-class Xeon platform |
| Supported Models | Up to 150 billion parameter AI models |
| Partners | ASRock, Gunnir, Leadtek, MAXSUN, Onix, Shandian, Sprak |
Key Features
- 24GB GDDR6 memory: Capacity is 2× that of consumer-grade Arc B580 (12GB)
- 160 XMX AI engines: Optimized for AI compute
- Multi-GPU expansion: Up to 8 cards, 192GB total memory
- ISV certification: Extensive independent software vendor certification, optimized for AEC (Architecture, Engineering, Construction) workstations and inference scenarios
- Linux support: Supports containerized software stack, simplifies AI deployment
Suitable Scenarios
- ✅ AI development workstations (model training, fine-tuning, inference)
- ✅ Large model local inference (70B+ usable, multi-GPU configuration)
- ✅ Professional graphics (CAD, 3D rendering)
- ✅ Budget-sensitive (lower price compared to NVIDIA professional cards)
- ❌ Top-tier cutting-edge AI models (FP32 compute relatively low)
- ❌ ECC memory requirement (ECC not supported)
External Links