Ascend Product Models

Ascend AI includes a series of training and inference products, as listed in Table 1. The training products support training and inference services. You can refer to the list of supported products of each component to check the details of supported product scope.

**Table 1** Ascend product series
Product Series	Product Model
Atlas A3 training products	Atlas 9000 A3 SuperPoD cluster computing system, Atlas 900 A3 SuperPoD cluster basic unit
Atlas A2 training products	Atlas 800T A2 training server, Atlas 900 A2 PoD cluster basic unit, Atlas 200T A2 Box16 heterogeneous subrack
Atlas training products	Atlas 800 training server (model 9000), Atlas 800 training server (model 9010), Atlas 900 PoD (model 9000), Atlas 900T PoD Lite, Atlas 300T training card (model 9000), and Atlas 300T Pro training card (model 9000)
Atlas 800I A2 inference products	Atlas 800I A2 inference server
Atlas inference products	Accelerator cards: Atlas 300I Pro inference card, Atlas 300V video analysis card, Atlas 300V Pro video analysis card, Atlas 300I Duo inference card
	Accelerator modules: Atlas 200I SoC A1 core board
	Edge servers: The Atlas 500 Pro AI edge server supports the Atlas 300I Pro inference card, Atlas 300V video analysis card, and Atlas 300V Pro video analysis card.
	Center inference servers: The Atlas 800 inference server (model 3000) and Atlas 800 inference server (model 3010) supports the Atlas 300I Pro inference card, Atlas 300V video analysis card, Atlas 300V Pro video analysis card, and Atlas 300I Duo inference card.
Atlas 200I/500 A2 inference products	Atlas 500 A2 edge station, Atlas 200I DK A2, Atlas 200I A2 accelerator module
Atlas 200/300/500 inference product	Atlas 200 AI accelerator module, Atlas 300I inference card (model 3000), Atlas 300I inference card (model 3010), Atlas 500 AI edge station, Atlas 200 DK

Ascend products can be classified based on the PCIe operating modes of the Ascend AI Processor, namely the root complex (RC) mode and endpoint (EP) mode. If the PCle works in the primary mode and can be extended with peripherals, it is called the RC mode. If the PCle works in the secondary mode, it is called the EP mode.

**Table 2** PCle operating modes
PCle Operating Mode	Supported Ascend Products	Description
RC mode	Atlas 200I A2 accelerator module (RC mode) Atlas 200 AI accelerator module (RC mode) Atlas 500 A2 edge station Atlas 200I SoC A1 core board Atlas 200I DK A2 Atlas 200 DK	The CPUs of such products run the AI service software specified by the running user directly and connect to peripherals such as network cameras, I²C sensors, and SPI monitors as secondary devices.
EP mode	Inference products: Atlas inference products Atlas 800I A2 inference products Atlas 300I inference card (model 3000) Atlas 300I inference card (model 3010) Atlas 200 AI accelerator module (EP mode) Atlas 200I A2 accelerator module (EP mode) Atlas 500 AI edge station	In EP mode, the host works as the primary side, and the device works as the secondary side. Customers' AI service programs run on the host. The Ascend product functions as a device system and connects to the host system through the PCIe channel. The host interacts with the device system through the PCIe channel and loads AI tasks to the Ascend AI Processor on the device.
EP mode	Training products: Atlas A3 training products Atlas A2 training products Atlas training products

Figure 1 RC and EP scenarios

In the heterogeneous computing architecture, the Ascend AI Processor interworks with the CPU of the server through the PCIe bus, involving the host and device sides.

Host: an x86 server or Arm server where the CPU is located. The host is connected to the hardware device (for example, Atlas 300I inference card) powered by the Ascend AI Processor. The host leverages the neural-network (NN) computing capabilities provided by the Ascend AI Processor to implement services.
Device: a hardware backend equipped with the Ascend AI Processor. It is connected to the server over the PCIe interface and provides NN computing capabilities required by the server.