Sparkle Intel Arc Pro B70 32 GB Single Slot Blower GPU Announced

Sparkle has introduced a single-slot blower-style variant of the Intel Arc Pro B70 graphics card, designed for workstation environments where space and thermal efficiency matter. The card features 32 GB of GDDR6 memory and 32 Xe2 compute cores, with a boost clock rated at 2800 MHz. It draws power through a single 16-pin PCIe connector and maintains a 160W total board power design.

Compact blower design enables dense multi-GPU AI workstation configurations

The blower-style cooler directs airflow straight out the rear I/O plate, which carries four display outputs. This form factor allows tighter packing in multi-GPU setups without relying on open-air cooling that can recirculate heat inside chassis enclosures. The single-slot footprint is a deliberate engineering choice for dense workstation builds rather than consumer gaming rigs.

Specifications

• Memory: 32 GB GDDR6
• Compute Cores: 32 Xe2 cores
• Form Factor: Single-slot
• Cooler Type: Blower-style cooler
• Power Connector: 16-pin PCIe

Sparkle positions the card to support up to eight units per workstation, yielding 256 GB of aggregate VRAM. This configuration targets AI and large language model workloads that require over 200 billion parameters. The company claims this setup enables inference and training tasks for models exceeding that scale, though sustained performance under heavy multi-GPU loads remains unquantified in available reports.

No pricing details have been disclosed alongside the announcement. Availability is listed as soon, with no specific launch window or regional rollout dates provided yet. The absence of cost information leaves buyers unable to evaluate total system investment for eight-card configurations at this stage.

The single-slot design may influence sustained performance under prolonged workloads due to thermal constraints inherent in blower coolers and restricted airflow paths. Intel and Sparkle have not released benchmark data or thermal throttling thresholds for multi-GPU operation, so real-world efficiency claims remain unverified against published standards.