Atlas 300T Training Card
Unlocking superb computing power
ผลิตภัณฑ์ โซลูชั่น และบริการสำหรับองค์กรธุรกิจ
Unlocking superb computing power
Atlas 300T Training Card (Model: 9000)
The Huawei Atlas 300T training card (model: 9000) is based on the Ascend processor and works with servers to provide powerful computing for data centers. A single card provides ultra-high computing power, accelerating deep learning and training. Atlas 300T features the highest computing power, integration, and bandwidth, meeting the AI training and high-performance computing requirements of the Internet, carriers, and finance.
Ultimate computing power
• Industry-leading computing power
Highest integration
• AI computing, general computing, and I/O 3-in-1
• Integrates 16 TaiShan Cores, and 1 100GE RoCE v2 NICs
Highest bandwidth
• Supports PCIe 4.0 and 1 100 Gbit/s RoCE high-speed ports, with a total egress bandwidth of 56.5 Gbit/s
• Boosts the efficiency of data training and gradient synchronization by 10–70% without the need for external NICs
Specifications
| Parameters | Atlas 300T AI training card Model: 9000 |
| Form Factor | Full height 3/4 length, dual-slot |
| AI Processor | Ascend Processors |
| Encoding/ Decoding | • 16-channel 4K (or 64-channel 1080p) 60 FPS H.264/H.265 • JPEG decoding: 1080p 2048 FPS, or equivalent decoding capability; maximum resolution: 8192 x 4320 • PNG decoding: 1080p 240 FPS, or equivalent decoding capability; maximum resolution: 4096 x 2160 • JPEG encoding: 1080p 256 FPS, or equivalent encoding capability; maximum resolution: 8192 x 4320 |
| Network | 1 100GE QSFP-DD ports |
| PCIe | PCIe x16 Gen4.0 |
| Power Consumption | Maximum: 300 W |
| Cooling Mode | Passive air cooling |
| Environment Conditions | • Operating temperature: 5°C to 35°C • Storage temperature: –40°C to +75°C • Operating humidity: 5% to 90% RH (non-condensing) • Storage humidity: 5% to 95% RH (non-condensing) • Altitude: Between 900 m and 3000 m, the maximum operating temperature decreases by 1°C as the altitude increases every 300 m |
Copyright © 2024 Huawei Technologies Co., Ltd. All rights reserved.