Skip to main content

MetaX Xiyun C290

Product Overview​

The MetaX Xiyun C290 is an early model of MetaX's Xiyun C series general-purpose compute GPUs, built on MetaX's proprietary XCORE 1.0 architecture and instruction set and delivered in the OAM 1.5 accelerator module form factor, positioned for AI training and inference. The Xiyun C series is MetaX's flagship product line (97%+ of revenue); based on the self-developed GPGPU architecture and paired with the MXMACA® software stack, which is fully compatible with the CUDA ecosystem, it is widely used in intelligent computing, general-purpose computing, and data analytics.

Compared with the later mass-produced Xiyun C500 (2023) / C550 / C600 (2025), the C290 is an early product of the same-generation XCORE 1.0 family with very little public specification material. Its compute range can be referenced against the XCORE 1.0 family (e.g., the C500: 15–40 TFLOPS FP32, 480 TOPS INT8, 64GB HBM2e), but the C290's official model-specific parameters are not public; the table below is marked "not disclosed/estimated" to avoid fabrication.

Core Specifications​

ParameterValue
ArchitectureMetaX proprietary XCORE 1.0 architecture and instruction set
Process Node7nm-class (estimated; same-generation XCORE 1.0 products are 7nm, officially not broken out for the C290)
FP32 ComputeNot disclosed (estimated; belongs to the XCORE 1.0 family, range likely near C500 level)
FP16 / BF16 ComputeNot disclosed (estimated; same as above)
INT8 ComputeNot disclosed (estimated; same as above)
Memory CapacityNot disclosed (same-generation XCORE 1.0 products mostly use 64GB HBM2e; refer to official MetaX data for the C290)
Memory TypeNot disclosed (estimated HBM2e)
Memory BandwidthNot disclosed
TDPNot disclosed
InterconnectMetaXLink inter-card interconnect (estimated, consistent with the Xiyun C series)
InterfaceOAM 1.5 (open accelerator module form factor)
Launch2022 (early model of the Xiyun C series)
Mass Production/AvailabilityIn mass production (specific scale not disclosed)

⚠️ Specification Notes: The C290 is only listed as a supported model in MetaX's developer documentation (Xiyun series user guide / out-of-band management manual); MetaX has not separately published a full parameter table for it. The "estimated" items in the table are directional references based on the parameter ranges of same-generation XCORE 1.0 architecture products (e.g., the C500) and are not measured/official C290 data; refer to official MetaX materials.

Key Features​

  • Proprietary XCORE 1.0 architecture: scalar + vector + tensor compute units, mixed multi-precision compute
  • OAM 1.5 form factor: open accelerator module for easy high-density server integration
  • MXMACA software stack: API-level compatibility with the CUDA ecosystem, supporting PyTorch/TensorFlow/PaddlePaddle
  • MetaXLink interconnect: high-speed inter-card interconnect supporting multi-card training clusters (specific port count not disclosed)

Vendor Information​

ParameterValue
CompanyMetaX Integrated Circuits (Shanghai) Co., Ltd.
HeadquartersShanghai
Founded2020-09
ListingSTAR Market 688802

Use Cases​

  • ✅ AI training and inference (general Xiyun C series scenarios)
  • ✅ Intelligent computing centers / general-purpose computing
  • ✅ Data analytics
  • ❌ Scarce public specifications; selection requires confirmation from MetaX
  • ❌ Graphics rendering (the Xiyun C series is a compute GPU, not a full-function GPU)

References​