Tip
Huawei: Data Transport Capacity Becomes the Biggest Bottleneck for AI Computing Power, Should Not Obsess Over Single-Chip Process
Huang Jian, General Manager of Huawei's China Government and Enterprise Internet Industry, pointed out that data transport capacity has become the biggest bottleneck for AI computing power. Instead of obsessing over single-chip process, efficiency should be improved through the integration of network, computing, and storage. In the past six months, token consumption has increased by 6 times, and the number of model calls has increased by about 50 to 100 times. Huawei has increased NPO bandwidth by more than eight times and token generation efficiency by five times.
Read the original (opens in a new tab)
News stream data aggregated by AI HOT