A800 NVLink 英伟达 基于Ampere架构的A100系列计算卡 

据相关媒体报道，英伟达推出A800系列计算卡，中国市场使用。与原有的A100系列计算卡相比，A800系列的规格基本相同，比较大的区别在于NVLink互连总线的连接速率，前者为600 GB/s，后者限制在了400 GB/s。据了解，英伟达新产品是为了绕开相关的出口限制。

A800系列计算卡分为三个版本以及SXM和PCIe两种不同形态，分别有40GB PCIe版、80GB PCIe版和80GB SXM版，不过这些计算卡FP64和FP32计算性能是相同的，分别是9.7 TFLOPS和19.5 TFLOPS。其中40GB PCIe版配备了40GB的HBM2显存，带宽为1555 GB/s，功耗为250W；80GB PCIe版配备了80GB的HBM2e显存，带宽为1935 GB/s，功耗为300W；80GB SXM版配备了80GB的HBM2e显存，带宽为2039 GB/s，功耗为400W。

英伟达官方在一份声明中称，A800系列计算卡已于2022年第三季度投入生产，作为目前A100系列计算卡的替代品出口到中国市场。由于出口管制的原因，英伟达曾表示可能会导致一个季度里损失约4亿美元。

According to relevant media reports, Nvidia launched the A800 series of computing cards for the Chinese market. Compared with the original A100 series compute card, the A800 series specifications are basically the same, the big difference is the NVLink interconnect bus connection rate, the former is 600 GB/s, the latter is limited to 400 GB/s. It is understood that Nvidia's new product is to circumvent the relevant export restrictions.

The A800 series compute cards are divided into three versions and SXM and PCIe two different forms, respectively, 40GB PCIe version, 80GB PCIe version and 80GB SXM version, but these compute cards FP64 and FP32 compute performance is the same, 9.7 TFLOPS and 19.5 TFLOPS, respectively. The 40GB PCIe version is equipped with 40GB of HBM2 memory, a bandwidth of 1555 GB/s, and a power consumption of 250W; The 80GB PCIe version comes with 80GB of HBM2e video memory, a bandwidth of 1935 GB/s, and a power consumption of 300W. The 80GB SXM edition comes with 80GB of HBM2e video memory, a bandwidth of 2039 GB/s, and a power consumption of 400W.

Nvidia said in a statement that the A800 series computing card has been put into production in the third quarter of 2022, as a replacement for the current A100 series computing card exported to the Chinese market. Because of the export controls, Nvidia has said it could lose about $400 million in a quarter.