View on GitHub ROCm_Logo

ROCm, a New Era in Open GPU Computing

Platform for GPU-Enabled HPC and Ultrascale Computing

  Catalyst & AMD GPU Pro ROCm 1.x
Core Features    
Shared Virtual Memory Coarse Grained Coarse Grained HSA Memory Allocation
Memory Allocation    
Page Migration Support NO NO
HMM Support NO NO
Raw Pointers NO NO
Platform Atomics NO YES
User Mode Queues NO YES
Hardware Scheduler for Queues NO YES
Max Number of Queues Restrictions Theoretical Unlimited Kernel Level 1024
Max Number of Processes 512 >512
Max Number of Concurrent Independent Processes 1 15
Process Preemption NO YES
User Mode SDMA NO YES
Defined Memory Model NO YES
Process Concurrency NO YES
Peer to Peer GPU Support with Large BAR NO YES
Peer to Peer GPU VIA RDMA NO YES
Peer to Peer DirectGMA YES NO
DirectGMA  for   SD/IO YES NO
Headless  OpenGL on EGL  Interop NO YES
OpenGL/X11 Interop YES YES
Largest Single Memory Allocation for  GPU Memory 3/4 total GPU Memory Total GPU Memory - 250 Megbytes
Developent Features    
Full HSA API Support NO YES
Open Source Driver and Runtime NO YES
Command Line Profiler YES YES
Open Source Native LLVM AMDGPU Compiler NO YES
Native ISA Programing Binary Interface  ( ABI) NO YES
Finalizer (HSAIL to SC Shader  Compiler ) YES YES
Programing Languges    
HCC Single Source Compiler C & C++ YES  Limited Features YES
HIP C++ Kernel Language with C Runtime API with HIPIFY for Porting CUDA applications NO YES
OpenCL 1.2 YES YES
OpenCL 2.0 YES Kernel Languge and Limited Host API
OpenMP 4.5 Support NO YES - In development
Continum IO Python Support No YES
Command line Offline  Compiler ( Compiles GPU Kernels  independent of the driver. ) NO YES
GCN Assembler  Support NO YES
OpenUCX NO YES - In development
OpenMPI NO YES - In development
MPICH NO NO
System Management  Features    
System Management API NO YES
Clock Management API and Public Tools NO YES
CPU  Support    
Intel Core i3,i5,i7 Optimized for Intel Core  CPU’s Optimized for version V3 or newer
Intel Xeon  E5 & E7 Supported Optimized for version V3 or newer
AMD Ryzen Supported Optimized
AMD Naples Supported Optimized
OpenPower NO IBM Power 8
ARM NO ARM AArch64 /Cavium Tunder X
GPU ASIC Support    
Hawaii ( S9150, S9170, S9100, W9100, W8100) YES YES
Fiji ( Fiji Nano, S9300x2) YES YES
Tonga ( W7100, S7100, S7150, S7150x2) YES YES
Polaris ( WX4100,WX5100,WX7100, RX460,RX470, RX480) YES YES