ROCm/ROCm
ROCm
Features
- AMD Hardware Acceleration - Optimizes machine learning and AI computations by leveraging AMD Instinct and Radeon GPUs through an open-source platform.
- GPU Acceleration - Runs machine learning, engineering, and scientific workloads on GFX9 and CDNA GPUs using the ROCm software stack.
- Deep Learning Frameworks - Integrates with popular deep learning frameworks to enable accelerated training and inference on AMD GPUs.
- Computation Abstraction Layers - Provides a unified runtime and compiler infrastructure mapping multiple programming models to AMD GPU hardware.
- Matrix Fused Multiply-Add Engines - Performs matrix fused multiply-add on KxN matrices using mixed-precision inputs to accelerate ML and HPC workloads.
- GPU-Accelerated HPC Development Platforms - Runs high-performance computing workloads on AMD GPUs using HIP, OpenCL, and OpenMP interfaces.
- GPU Debugger Extensions - Ships a GDB-based debugger for inspecting and controlling GPU-accelerated programs on AMD hardware.
- GDB-Based GPU Debuggers - Inspects and controls GPU-accelerated programs using a GDB-based debugger for AMD hardware.
- CMake Integrations - Provides config-file CMake packages that automatically locate and link ROCm components without custom modules.
- Platform Library Distributions - Organizes GPU runtime, compiler, math libraries, and profiling tools into layered, versioned packages.
- Mixed-Precision Matrix Accelerators - Provides optimized libraries for mixed-precision matrix operations and HPC workloads on AMD GPUs.
- GPU Debugger Extensions - Extends GDB with GPU thread inspection, memory access, and breakpoint support for AMD hardware.
- GPU Programming Runtimes - Provides the HIP runtime that translates CUDA-like code to run on AMD GPUs.
- Multi-Language GPU Programming Interfaces - Provides HIP, OpenCL, and OpenMP interfaces for developing high-performance computing applications on AMD GPUs.
- GDB-Based GPU Debuggers - Inspects and controls GPU-accelerated programs with a GDB-based debugger for AMD hardware.
- GPU Health Monitors - Tracks and manages data center GPU health metrics through a command-line interface wrapping device driver calls.
- GPU - Provides a command-line tool for monitoring and managing AMD GPU health, reliability, and performance metrics.
- Kernel Dispatchers - Launches compute kernels directly on GPU hardware via a command processor for minimal latency.
- Mixed-Precision Compute Engines - Accelerates matrix multiply-add operations by dispatching FP32, FP16, BF16, and Int8 tiles to dedicated hardware units.
- Mixed-Precision Computing - Performs matrix operations with FP32, FP16, BF16, and Int8 inputs to speed up ML and HPC workloads.
- GPU Health Monitors - Monitors and manages data center GPU health metrics through a command-line interface.
- Config-File Package Integrations - Exports pre-built CMake config files that automatically locate and link ROCm components without custom Find modules.
- Inter-GPU Bandwidth Probes - Reports the throughput of the xGMI interconnect between GPU nodes via a dedicated library API.
- HPC System Tuners - Offers guidance on system optimization and performance validation for HPC workloads on AMD GPUs.
