MIGraphX precision support

2026-06-25

1 min read time

Applies to Linux

This topic lists the data type support for the MIGraphX library on AMD GPUs.

This page lists the data types supported by the library itself and does not indicate hardware support. A type listed here is only usable if the GPU architecture also supports it; otherwise it is unsupported. For data type support across the other ROCm libraries and by GPU architecture, see the Data types and precision support page.

Supported data types

The following table lists the primitive compute types supported by MIGraphX — types that can be specified as tensor element types and computed on directly.

Icon	Definition
✅	Fully supported as both an input and output type.
⚠️	Partially supported as an input or output type.

Data types not listed in the table below are not supported.

Data type	Support
int8	✅
int16	✅
int32	✅
int64	✅
float8 (E4M3)	✅
float8 (E5M2)	✅
float16	✅
bfloat16	✅
float32	✅
float64	✅

Supported quantized model formats

In addition to the compute types above, MIGraphX can execute models that use the following quantized weight formats. These are not native compute types — MIGraphX unpacks them internally around compute operations rather than operating on them directly.

Format	Description
MXFP4 (E2M1)	OCP microscaling 4-bit floating-point format. Supported via the ONNX MXQuantizeDequantize and DynamicScale operators.
INT4	4-bit integer format used in bitsandbytes and GGML quantized models. Supported via the ONNX MatMulBnb4 and MatMulNBits operators.
NF4	NormalFloat 4-bit format used in bitsandbytes quantized models. Supported via the ONNX MatMulBnb4 operator.