CloudCard

F12010

Process

Frequency

16nm

200MHz

Support for Computing Precision

INT8

INT16

FP16

FP32

Peak INT8 TOPS

1.638TOPS

Memory Capacity

2x DDR 4 16GB

bit width: 512bit

bandwidth: 154GB/s

Interface

PCIe Gen3 x 16

Compatible PCIe Gen 4 x 8

bandwidth: 16GB/s

Thermal Cooling Solution

passive

Form Factor

Double slot, full height 3/4 length

F11610E

16nm

1x DDR 4 16GB

bit width: 256bit

bandwidth: 102.4GB/s

PCIe Gen3 x 16

Compatible PCIe Gen 4 x 8

bandwidth: 16GB/s

162MHz

145TOPS

passive

INT8

INT16

FP16

FP32

G41210E

12nm

4x 32GB

bit width: 512bit

bandwidth: GDDR5:512GB/s
GDDR5X:704GB/s
GDDR6:1024GB/s

PCIe Gen4 x 16

bandwidth: 32GB/s

1.2GHz

>1000TOPS

passive

INT8

INT16

FP16

FP32

G40710E

7nm

PCIe Gen4 x 16

bandwidth: 32GB/s

1.6GHz

>4000TOPS

passive

INT8

INT16

FP16

FP32

Double slot, full height and full length

4x 32GB

bit width: 512bit

bandwidth: GDDR5:512GB/s
GDDR5X:704GB/s
GDDR6:1024GB/s

Product Parameters

Support to Various
Application Scenarios

▪ Reconfigurable Computing-in-Memory technology for multi-operator support

▪ Adaptatibility to multiple models

▪ Compatibility with multiple frameworks to support various application scenarios

High Density Surge Computing Power

▪ Advanced Computing-in-Memory architecture with advanced process

▪ Multi-core & multi-chip for efficient and high-density AI computing power

Flexible and Highly Compatible Development Platform

▪ A development platform built on a universal AI platform

▪ Supports multiple mainstream deep learning frameworks

▪ Simplifies the development flow and easily ports existing algorithms

Ultra-Low Energy Cost on Operation

▪ Breaks the traditional power comsumption wall for data transformation, with ultra-high energy efficiency ratio to bring excellent deployment operation cost performance

Application Scenarios