The GPU card cooperated by Tencent and AMD is a semi-custom product

At the Tencent Digital Ecology Conference a few days ago, Tencent disclosed three self-developed chips, which are oriented to AI, video, and network processing in three different directions. In addition, Tencent has also cooperated with AMD to jointly release the new Tencent Cloud “Xinghai” hot-fire heterogeneous server AG251, and Xinghai Wisdom Wood GA01 GPU card.

According to AMD’s official introduction, Tencent and AMD’s cooperative Xinghai Wisdom Wood GA01 GPU card is equipped with the latest Radeon Pro V620. Based on the RDNA 2 architecture, it has 72 CU computing units, that is, 4608 stream processors (same as consumer Radeon RX 6800 XT), configured with 32GB of GDDR6 memory, at a rate of 16 Gbps, and provides 512 GB/s of video memory bandwidth.
AMD Radeon RX 7900
According to the TechnoPixel report, this GPU card of Tencent is a semi-custom product, which is different from the passive cooling structure of AMD Radeon Pro V620. It supports water-cooled heat dissipation, which can be combined with business needs, and is more flexible in deployment and more efficient in overall performance. It also has real-time fault monitoring and early warning functions to make it more reliable in operation. The most important point is that this GPU card has a high degree of customization at the driver level, optimized for rendering, encoding and decoding, and image processing. Tencent said that compared to the industry’s mainstream rendering computing cards, the performance of a single card has increased by more than 120% to meet the demand for GPU performance in multiple scenarios.

Under normal circumstances, AMD’s Radeon Pro series products have their own drivers, but this time the cooperation with Tencent is obviously deeper. Not only has the hardware been specifically designed, but it also allows Tencent to perform self-designed operations on the software. For Tencent, the work of its cloud data center requires a variety of different types of workloads. A multi-purpose and multi-functional custom GPU can improve work efficiency. It seems natural to turn to AMD for semi-custom solutions.