Aiserver Manufacturer
1 / 5
| Introduction | The Haiguang K100 AI version is the second-generation DCU (Data Center Acceleration Card) launched, specifically designed for AI training and inference. It features a peak FP16 computing performance of 196 TFLOPS and 64GB of high-bandwidth memory, offering about 60% of the performance of the A100. |
| Name | Haiguang K100 AI 64GB |
| FP16/BF16 | 196 TFLOPS |
| INT8 | 392 TOPS |
| FT32 | 98 TFLOPS |
| FP32 | 49 TFLOPS |
| VRAM Type | HBM2E (High Bandwidth Memory, with approximately 30% higher bandwidth compared to HBM2) |
| Process | TSMC 7nm FinFET process, paired with 2.5D CoWoS advanced packaging |
| DirectX Support | Compatible with DirectX 12 (including 12_1 features) |
| Memory Frequency | Effective frequency 3600MHz (800GB/s bandwidth) |
| Bus Width | 2048-bit |
| Memory Speed | 0.55 ns |
| Output Interface | Focus on data center inference/computing (No traditional DP/HDMI/VGA) |
| Interface Type | PCIe Gen4.0 x16 (bidirectional bandwidth 64GB/s) |
| Form Factor | FHFL dual-slot |
| Weight | 1064g |
Security is integrated into every phase of the lifecycle, including protected supply chain and factory-to-site integrity assurance. Silicon-based root of trust anchors end-to-end boot resilience.
Increase efficiency and accelerate operations with advanced management portfolios. Simplify, automate, and centralize management for secure and efficient comprehensive solutions.
Designed for energy efficiency to help reduce the carbon footprint and lower operation costs. Includes innovative options for recycling and responsible system retirement.
✓ Designed for 24x7 enterprise data center operations.
✓ Features energy-efficient hardware optimized for deployment at scale.
✓ Supports Secure Boot with Hardware Root of Trust technology.
✓ Passively cooled, dual-socket design for easy integration.
