Cambrian releases new AI training card MLU370-X8

According to Cambrian official news, Cambrian today officially released a new training accelerator card MLU370-X8, equipped with dual-chip four-core Siyuan 370, integrated Cambrian MLU-Link multi-core interconnection technology, mainly for training tasks.

Cambrian said that in training tasks such as YOLOv3 and Transformer widely used in the industry, the parallel performance of the 8-card computing system reached an average of 155% of the 350W RTX GPU. The MLU370-X8 uses a 7nm process, is equipped with 48GB LPDDR5 memory, and has FP32 single-precision floating-point performance of 24 TFLOPS.

According to reports, the MLU370-X8 smart accelerator card provides a maximum training power consumption of 250W, which can give full play to the FP32, FP16 or BF16 computing performance commonly used in AI training acceleration. For the first time, Cambrian integrated the dual-chip four-core Siyuan 370 into the MLU370-X8 smart accelerator card, providing twice the memory and codec resources of the standard Siyuan 370 accelerator card, and equipped with MLU-Link™ multi-core interconnection Technology.

The MLU370-X8 smart accelerator card supports MLU-Link™ multi-core interconnection technology, providing interconnection functions within and between cards. Cambrian specially designed MLU-Link bridge card for the multi-card system, which can realize the full interconnection of 8 Siyuan 370 chips with 4 accelerator cards as a group, each accelerator card can obtain 200GB/s communication throughput performance, bandwidth It is 3.1 times that of PCIe 4.0, and can efficiently perform multi-core and multi-card training and distributed inference tasks.