+8615201351407
1. Acceleration of pre-processing for large models: Given the bottleneck of GPU computing power, our company utilizes FPGA for text/image/voice data cleaning, format conversion, denoising, and tensor packing, offloading CPU/GPU workload, reducing bandwidth usage, and shortening latency by over 40%. Cloud providers (Alibaba Cloud, Huawei Cloud, Tianyi Cloud) have made bulk purchases.
2. Large model low-latency inference cluster (CXL interconnection architecture): fragmented inference of trillion-parameter large models, KV cache pooling, Versal AI series + VU high-end FPGA to build distributed inference nodes, capable of completing scenarios where models can be iterated at any time (recommendation systems, customer service large models)
3. Integration of intelligent network card + DPU: Our FPGA accelerator card supports 800G/400G high-speed network cards, memory virtualization, and computing power scheduling. It is a standard accessory for intelligent computing servers, with up to dozens of FPGAs in a single cabinet
4. Video cloud ultra-large-scale transcoding + AI analysis: capable of completing multi-channel live streaming, parallel on-demand decoding + behavior recognition
Previous：No
Next：No
