The Atlas 300T Pro training card cooperates with the server to provide powerful computing power for the data center. A single card can provide up to 280 TFLOPS FP16 computing power to speed up the deep learning training process. Atlas 300T Pro has the characteristics of super computing power, high degree of integration, and high-speed bandwidth, meeting the computing power needs of education, scientific research, operators, finance, etc. that require artificial intelligence training.
Provides industry-leading 280 TFLOPS FP16 computing power
Support 16 core * 2.2 GHz CPU computing power
AI computing power, general computing power, and I/O capabilities three-in-one
Integrated 30 AI cores, 16 TaiShan cores, 1*100GE RoCE
Support PCIe 4.0 and 1*100GE RoCE high-speed interface
No external network card is required, and the efficiency of training data and gradient synchronization is increased by 10%-70%.
| model | Atlas 300T Pro |
|---|---|
| form | Dual-slot full-height, full-length PCIe card |
| AI computing power | 280 TFLOPS FP16 |
| Memory specifications | 32 GB HBM, total bandwidth 1228 GB/s
Support ECC |
| CPU computing power | 16core*2.2GHz |
| Codec capabilities | Support 16 channel 4K (or 64 channel 1080P) 60 FPS H.264/H.265
JPEG decoding capability 1080P 2048FPS, or equivalent decoding capability, The highest resolution is 8192*4320 PNG decoding capability 1080P 240FPS, or equivalent decoding capability, The highest resolution is 4096*2160 JPEG encoding capability 1080P 256FPS, or equivalent encoding capability, The highest resolution is 8192*4320 |
| network | 1*100GE QSFP-DD interface |
| PCIe interface | PCIe x16 Gen4.0 |
| Maximum power consumption | 300W |
| Heat dissipation method | Passive air cooling |
| Working temperature | 5℃~45℃ (41℉~113℉) |
| structure size | 266.7mm (length) × 39.04mm (width) × 111.15mm (height) |
Reviews
There are no reviews yet.