Skip to main content

/

Docs

Nightly

v25.1

Log in

Mojo module

info

Contains information about GPU architectures and their capabilities.

This module provides detailed specifications for various GPU models including NVIDIA and AMD GPUs. It includes information about compute capabilities, memory specifications, thread organization, and performance characteristics.

Aliases​

  • ​A10 = Info("A10", Vendor(__init__[__mlir_type.!pop.int_literal](2)), "cuda", "ampere", "nvptx-short-ptr=true", __init__[__mlir_type.!pop.float_literal](8.5999999999999996), "sm_86", 72, 32, 1536, 32, 64, 2048, 32, 102400, 65536, 256, "warp", 255, 65536, 16, 128, 4, 1024, Flops(__init__[__mlir_type.!pop.int_literal](125), __init__[__mlir_type.!pop.int_literal](250), __init__[__mlir_type.!pop.int_literal](500), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.float_literal](62.5), __init__[__mlir_type.!pop.int_literal](0))):
  • ​A100 = Info("A100", Vendor(__init__[__mlir_type.!pop.int_literal](2)), "cuda", "ampere", "nvptx-short-ptr=true", __init__[__mlir_type.!pop.float_literal](8), "sm_80", 108, 32, 2048, 32, 64, 2048, 32, 167936, 65536, 256, "warp", 255, 65536, 32, 128, 4, 1024, Flops(__init__[__mlir_type.!pop.int_literal](312), __init__[__mlir_type.!pop.int_literal](624), __init__[__mlir_type.!pop.int_literal](1248), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.int_literal](156), __init__[__mlir_type.!pop.float_literal](19.5))):
  • ​DEFAULT_GPU = from_name[::StringLiteral]():
  • ​DEFAULT_GPU_ARCH = _accelerator_arch():
  • ​DEFAULT_GPU_TARGET = from_name[::StringLiteral]().target[::Int]():
  • ​H100 = Info("H100", Vendor(__init__[__mlir_type.!pop.int_literal](2)), "cuda", "hopper", "nvptx-short-ptr=true", __init__[__mlir_type.!pop.float_literal](9), "sm_90a", 132, 32, -1, 32, 64, 2048, 32, 233472, 65536, 256, "warp", 255, 65536, 32, 128, 4, 1024, Flops(__init__[__mlir_type.!pop.int_literal](1979), __init__[__mlir_type.!pop.int_literal](3958), __init__[__mlir_type.!pop.int_literal](7916), __init__[__mlir_type.!pop.int_literal](3958), __init__[__mlir_type.!pop.int_literal](989), __init__[__mlir_type.!pop.int_literal](67))):
  • ​L4 = Info("L4", Vendor(__init__[__mlir_type.!pop.int_literal](2)), "cuda", "ada", "nvptx-short-ptr=true", __init__[__mlir_type.!pop.float_literal](8.9000000000000004), "sm_89", 58, 32, 1536, 32, 64, 2048, 32, 102400, 65536, 256, "warp", 255, 65536, 24, 128, 4, 1024, Flops(__init__[__mlir_type.!pop.int_literal](121), __init__[__mlir_type.!pop.int_literal](242), __init__[__mlir_type.!pop.int_literal](485), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.int_literal](0))):
  • ​MI300X = Info("MI300X", Vendor(__init__[__mlir_type.!pop.int_literal](1)), "hip", "gfx942", "", __init__[__mlir_type.!pop.float_literal](9.4000000000000003), "CDNA3", 304, 64, 2048, 64, 32, 2048, 2, 65536, 65536, 256, "warp", 255, 65536, 2, 128, 4, 1024, Flops(__init__[__mlir_type.!pop.float_literal](1307.4000000000001), __init__[__mlir_type.!pop.float_literal](2614.9000000000001), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.float_literal](1307.4000000000001), __init__[__mlir_type.!pop.float_literal](653.70000000000005), __init__[__mlir_type.!pop.int_literal](0))):
  • ​NoGPU = Info("NoGPU", Vendor(__init__[__mlir_type.!pop.int_literal](0)), "none", "no_gpu", "", __init__[__mlir_type.!pop.int_literal](0), "", 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, "none", 0, 0, 0, 0, 0, 0, Flops(__init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.int_literal](0), __init__[__mlir_type.!pop.int_literal](0))):

Structs​

  • ​Flops: Represents floating point operations per second for different precisions.
  • ​Info: Comprehensive information about a GPU architecture.
  • ​Vendor: Represents GPU vendors.

Functions​

  • ​is_cpu: Checks if the target is a CPU (compile-time version).
  • ​is_gpu: Checks if the target is a GPU (compile-time version).
  • ​is_valid_target: Checks if the target is valid (compile-time version).