This repo was created for anyone curious as to what is going on with AMD GPUs,ROCm (Radeon Open Compute) and the machine learning space.
The GPUs referenced in this tutorial:
- MSI Gaming Radeon 7900 XTX 24 GB.
One thing of annoyance is the long term support of GPUs with a ongoing comment section here. To summarize NVIDIA usually supports their GPUs for 8+ years and it looks like AMD only for ~4 years. Keep this in mind when working with these GPUs!
For each ROCm version ensure you check the GPU Support and OS Compatibility (Linux) first prior to obtaining and using an AMD GPU.
Check out the following blog that I found on this news article.
LINK JONSBO D31 MESH SC BLACK Micro ATX Computer Case with Sub HD-LCD Display, M-ATX/DTX/ITX Mainboard/Support RTX 4090(335-400mm) GPU 360/280AIO,Power ATX/SFX: 100mm-220mm Multiple Tool-free Design,Black
LINK be quiet! PURE LOOP 360mm AIO CPU Water Cooler | All In One Water Cooling System | Intel and AMD Support | Low Noise | BW008
LINK Iceberg Thermal IceGALE Silent 120mm (Black) 3-PACK Quiet Optimized Airflow 3-Pin Case Fans
LINK Bgears Vortex 120 ARGB LED Case Fans b-ARGB Vortex 120
LINK Thermal Grizzly Kryonaut Thermal Grease Paste - 1.0 Gram
LINK AMD Ryzen 9 5900X - Ryzen 9 5000 Series Vermeer (Zen 3) 12-Core 3.7 GHz Socket AM4 105W None Integrated Graphics Desktop Processor - 100-100000061WOF
LINK MSI Gaming AMD Radeon RX 6600 XT 128-bit 8GB GDDR6 DP/HDMI Dual Torx Fans FreeSync DirectX 12 VR Ready OC Graphics Card (RX 6600 XT MECH 2X 8G OC) LINK MSI Gaming Radeon RX 7900 XTX 24GB GDDR6 PCI Express 4.0 ATX Video Card RX 7900 XTX GAMING TRIO CLASSIC 24G
[LINK] (https://www.newegg.com/samsung-970-evo-plus-500gb/p/N82E16820147742?Item=N82E16820147742) SAMSUNG 970 EVO PLUS M.2 2280 500GB PCIe Gen 3.0 x4, NVMe 1.3 V-NAND Internal Solid State Drive (SSD) MZ-V7S500B/AM
LINK 990 PRO w/ Heatsink PCIe®4.0 NVMe™ SSD 1TB
LINK G.SKILL TridentZ RGB Series 128GB (4 x 32GB) 288-Pin PC RAM DDR4 3600 (PC4 28800) Desktop Memory Model F4-3600C18Q-128GTZR
LINK MSI MAG B550M MORTAR MAX WIFI DDR4 AM4 AMD B550 SATA 6Gb/s Micro-ATX Wi-Fi 6E 2.5Gbps LAN M.2 (Key-E) Motherboards - AMD
LINK Deepcool PQ850M R-PQ850M-FA0B-US 850 W ATX12V V2.4 80 PLUS GOLD Certified Full Modular Active PFC Power Supply
Version Used 6.0.0 Compatible GPU List
ROCm is an open-source stack, composed primarily of open-source software, designed for graphics processing unit (GPU) computation. ROCm consists of a collection of drivers, development tools, and APIs that enable GPU programming from low-level kernel to end-user applications.
With ROCm, you can customize your GPU software to meet your specific needs. You can develop, collaborate, test, and deploy your applications in a free, open source, integrated, and secure software ecosystem. ROCm is particularly well-suited to GPU-accelerated high-performance computing (HPC), artificial intelligence (AI), scientific computing, and computer aided design (CAD).
ROCm is powered by AMD’s Heterogeneous-computing Interface for Portability (HIP), an open-source software C++ GPU programming environment and its corresponding runtime. HIP allows ROCm developers to create portable applications on different platforms by deploying code on a range of platforms, from dedicated gaming GPUs to exascale HPC clusters.
ROCm supports programming models, such as OpenMP and OpenCL, and includes all necessary open source software compilers, debuggers, and libraries. ROCm is fully integrated into machine learning (ML) frameworks, such as PyTorch and TensorFlow.Key Components of ROCm:
-
ROCm Runtime: ROCm includes a runtime that provides the necessary low-level GPU support, including hardware abstraction, memory management, and kernel execution. It allows developers to write and execute GPU-accelerated code using a variety of programming languages, including C, C++, and Python.
-
ROCm Math Libraries (RocBLAS, RocFFT, etc.): ROCm provides a suite of math libraries optimized for AMD GPUs. These libraries include linear algebra routines (RocBLAS), Fast Fourier Transform (FFT) operations (RocFFT), and more. They are crucial for scientific and engineering applications.
-
HIP (Heterogeneous-Compute Interface for Portability): HIP is a key component of ROCm that allows developers to write GPU-accelerated code in a way that can be easily ported between AMD and NVIDIA GPUs. It offers a high-level API and tools for GPU application development.
-
ROCr (ROCm Runtime API): ROCr is an API that provides low-level control and management of AMD GPUs. It allows developers to interact directly with the GPU hardware and execute custom kernels.
-
ROCProfiler: ROCProfiler is a profiling tool that helps developers analyze the performance of GPU-accelerated code. It provides insights into kernel execution times, memory usage, and other performance metrics.
-
ROCTracer: This tool enables fine-grained GPU kernel tracing and profiling, making it easier to identify performance bottlenecks and optimize GPU code.
-
ROCm SMI (System Management Interface): ROCm SMI allows users to monitor and manage the health and performance of AMD GPUs. It provides information about GPU temperature, power consumption, and other metrics.
Supported Frameworks:
ROCm is compatible with popular machine learning and deep learning frameworks, including TensorFlow and PyTorch. This compatibility allows researchers and data scientists to leverage the power of AMD GPUs for training and inference in machine learning models.
ROCm Use Cases:
-
Scientific Computing: ROCm is widely used in scientific research for simulations, computational fluid dynamics, molecular modeling, and other scientific computations that benefit from GPU acceleration.
-
Machine Learning: ROCm enables training and inference for deep learning models, making it valuable for AI and machine learning applications.
-
Data Analytics: ROCm can accelerate data analytics tasks, helping organizations process large datasets more efficiently.
-
High-Performance Computing (HPC): ROCm is suitable for HPC clusters and supercomputers, where GPU acceleration is essential for scientific simulations and numerical calculations.
Community and Open Source: ROCm is developed as an open-source project, fostering a collaborative community of developers, researchers, and enthusiasts. The open nature of ROCm encourages contributions, improvements, and the development of custom solutions for specific use cases.
Compatibility: ROCm is primarily designed for AMD GPUs, including Radeon Instinct and Radeon Pro series cards. It is compatible with Linux-based operating systems and is often used with popular Linux distributions such as Ubuntu and CentOS.
ROCm Packages:
-
rocm-dkms (ROCm Dynamic Kernel Module Support):
-
rocm-dkms is a package that provides Dynamic Kernel Module Support (DKMS) for the ROCm (Radeon Open Compute) software stack.
-
Dynamic Kernel Module Support (DKMS):
-
DKMS is a framework that enables automatic building and installation of kernel modules when the Linux kernel is updated.
-
It ensures that ROCm-specific kernel modules stay compatible with different kernel versions without manual intervention.
-
Ensures ROCm compatibility with various Linux kernels, making it easier to maintain ROCm functionality.
-
-
-
rocm-libs:
-
rocm-libs contains GPU-accelerated libraries and tools for high-performance computing, machine learning, and more, as part of the ROCm software stack.
-
ROCm is an open-source framework for GPU computing on AMD GPUs.
-
rocm-libs is a fundamental part of this stack, providing essential libraries.
-
ROCm Libraries include optimized GPU libraries such as ROCm Math Libraries (RocBLAS, rocSOLVER) and ROCm Communication Libraries (RCCL).
-
These libraries enhance performance for various computing workloads.
-
Provides GPU-accelerated capabilities for scientific computing, machine learning, and deep learning tasks.
-
Optimized libraries improve the efficiency of GPU-based applications.
-
-
-
rccl (ROCm Communication Collectives Library):
-
rccl is a specific library within the ROCm stack, designed to optimize GPU communication and synchronization for parallel computing workloads.
-
rccl focuses on optimizing communication primitives and collectives for multi-GPU parallel computing.
-
It enhances synchronization and data exchange between GPUs.
-
Valuable for distributed GPU computing, especially in deep learning training across multiple GPUs.
-
Improves the performance and scalability of GPU-accelerated parallel computing applications.
-
Enables efficient communication and coordination between GPUs in multi-GPU systems.
-
-
Why They Are Important:
rocm-dkms: Ensures that ROCm remains compatible with different Linux kernel versions, simplifying the maintenance of GPU compute capabilities on AMD GPUs.
rocm-libs: Provides essential GPU-accelerated libraries for various high-performance computing tasks, making it easier to develop efficient applications.
rccl: Optimizes GPU communication for parallel computing, enhancing the performance and scalability of multi-GPU applications, particularly valuable in deep learning and scientific computing.
[rccl](https://rocmdocs.amd.com/en/latest/reference/library-index.html#rccl) facilitates efficient communication across GPUs in multi-GPU configurations, essential for distributed deep learning and parallel computing tasks.
[rocm-libs](https://rocmdocs.amd.com/en/latest/reference/library-index.html) is a comprehensive suite of AMD-specific, device-side language runtime libraries designed to support the development and execution of high-performance computing applications on AMD GPUs.
These libraries are pivotal for developers aiming to leverage AMD's GPU computing capabilities across a wide range of scientific, data analysis, and machine learning applications, providing the foundational components for building and executing compute-intensive applications on AMD's GPU architecture
- Install rccm and rocm-libs.
sudo apt-get install -y rccl rocm-libsA tool shipped with this script to enumerate GPU agents available on a working ROCm stack.
rocminfoMonitor the GPU with rocm-smi.
Usage (use watch -n 1 # replace 1 with second duration for continuous monitoring):
rocm-smiPyTorch is an open-source deep learning framework known for its dynamic computation graph, making it flexible for experimentation and debugging. It's Pythonic, easy to learn, and widely used in research for tasks like computer vision, natural language processing, and reinforcement learning.
Version Used:
- Pytorch has not been as big of a pain setting up as Tensorflow. Go back in versions and try it, unless asked I'll just try later versions).
TensorFlow, developed by Google Brain, is an open-source machine learning framework with a static computation graph, optimized for production deployments. It's scalable, works on CPUs and GPUs, and offers TensorFlow Serving for model deployment. TensorFlow provides high-level APIs like Keras and is used in various machine learning applications, including computer vision, natural language processing, and recommendation systems.
Version Used 2.16.0.600-dev20240219
- 2.15.0.600 specifically this one works.
- 2.14.0.600: The supported AMDGPU versions are gfx1030gfx1100, gfx900, gfx906, gfx908, gfx90a, gfx940, gfx941, gfx942.
- Yes a comma prevents this version from running.
- Recompilation at tensorflow/compiler/xla/stream_executor/device_description.h line 184, which takes an hour on the above build will get tried and this instruction updated if it works.
- 2.13.1.600: The supported AMDGPU versions are gfx1030, gfx900, gfx906, gfx908, gfx90a, gfx940, gfx941, gfx942.
The MSI Gaming Radeon RX 7900 XTX is a high-end graphics card designed for gaming and demanding graphics workloads. With 24GB of GDDR6 memory and support for PCI Express 4.0, it delivers excellent performance and is suitable for gamers and professionals looking for top-tier graphics capabilities.
| Technical Specifications | Details |
|---|---|
| Graphics Processing Unit | AMD Radeon RX 7900 XTX (Navi 31) |
| Architecture | RDNA 3 (gfx1100) |
| Core Clock | Up to 2340 MHz (Boost Clock) |
| Compute Units | 96 |
| Boost Frequency | Up to 2500 MHz |
| Stream Processors | 6144 |
| AI Accelerators | 192 |
| Peak Pixel Fill-Rate | Up to 480 GP/s |
| Peak Texture Fill-Rate | Up to 960 GT/s |
| Peak Half Precision Compute Performance | 123 TFLOPs |
| Peak Single Precision Compute Performance | 61 TFLOPs |
| ROPs | 192 |
| Texture Units | 384 |
| Transistor Count | 58 B |
| Memory | 24GB GDDR6 |
| Memory Speed | Up to 20 Gbps (Effective) |
| Memory Type | GDDR6 |
| Memory Bus | 384-bit |
| Memory Bandwidth | Up to 960 GB/s |
| Effective Memory Bandwidth | Up to 3500 GB/s |
| AMD Infinity Cache Technology | 96 MB |
| PCI Express Version | PCIe 4.0 |
| TDP (Thermal Design Power) | 330W |
| Typical Board Power (Desktop) | 355 W |
| Minimum PSU Recommendation | 800 W |
| Power Connectors | 2 x 8-pin PCIe |
| Display Outputs | - 3 x DisplayPort 1.4a |
| - 1 x HDMI 2.1 | |
| DirectX Support | DirectX 12 Ultimate |
| Ray Tracing Support | AMD Ray Accelerator, Ray Tracing Support |
| VR Ready | Yes |
| CrossFire Support | Yes (With compatible AMD GPUs) |
| RGB Lighting | Mystic Light RGB lighting |
| Form Factor | ATX |
| Dimensions (L X W X H) | 325 mm (12.8 inches) X 141 mm (5.6 inches) X 56 mm (2.2 inches) |
| Slot Size | 2.5 slots |
| Cooling | Triple-fan cooling system with Zero Frozr |
| Recommended PSU | Minimum 750W |
| Operating Systems | - Windows 10/11 (64-bit) |
| - Linux (64-bit) | |
| Warranty | 3 years |
The MSI Gaming Radeon RX 7900 XTX is a powerful graphics card that boasts 24GB of GDDR6 memory and supports PCI Express 4.0, making it ideal for high-resolution gaming, content creation, and professional applications. Its triple-fan cooling system ensures efficient heat dissipation, and it features RGB lighting for customization. With DirectX 12 Ultimate support and ray tracing capabilities, it provides an immersive gaming experience. Additionally, it supports CrossFire for multi-GPU setups and is VR-ready. This graphics card is a solid choice for enthusiasts and gamers seeking top-notch performance.
- Minimal Installation
- Download updates
ROCm Platform: Offers a comprehensive foundation for HPC, machine learning, and scientific research on AMD GPUs.
Installation instructions take from here
- Download and convert the package signing key.
# Make the directory if it doesn't exist yet.
# This location is recommended by the distribution maintainers.
sudo mkdir --parents --mode=0755 /etc/apt/keyrings
# Download the key, convert the signing-key to a full
# keyring required by apt and store in the keyring directory
wget https://repo.radeon.com/rocm/rocm.gpg.key -O - | gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/nullNote: The GPG key may change; ensure it is updated when installing a new release. If the key signature verification fails while updating, re-add the key from the ROCm to the apt repository as mentioned above. The current rocm.gpg.key is not available in a standard key ring distribution but has the following SHA1 sum hash: 73f5d8100de6048aa38a8b84cd9a87f05177d208 rocm.gpg.key
- Add the AMDGPU repository for the driver.
echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/amdgpu/6.0/ubuntu jammy main" \
| sudo tee /etc/apt/sources.list.d/amdgpu.list
sudo apt update- Add the ROCm repository:
echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/6.0 jammy main" \
| sudo tee --append /etc/apt/sources.list.d/rocm.list
echo -e 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600' \
| sudo tee /etc/apt/preferences.d/rocm-pin-600- Install kernel driver:
sudo apt-get install -y amdgpu-dkms- Reboot:
sudo reboot- Update package list:
sudo apt-get update- Install ROCm packages:
sudo apt-get install -y rocm-hip-sdk- Add yourself to the rend and video group:
sudo usermod -a -G render,video $LOGNAME- Reboot, for changes to take effect:
sudo rebootInstructions taken from here
- Configure the system linker. Instruct the system linker where to find shared objects (.so-files) for ROCm applications.
sudo tee --append /etc/ld.so.conf.d/rocm.conf <<EOF
/opt/rocm/lib
/opt/rocm/lib64
EOF
sudo ldconfig- Configure PATH. Add binary paths to the PATH environment variable.
echo 'export PATH=$PATH:/opt/rocm-6.0.0/bin' >> ~/.bashrc- Close and open your terminal or use the following command:
source ~/.bashrc- Verify kernel-mode driver installation.
dkms status- Verify ROCm installation.
rocminfoResult:
ROCk module is loaded
=====================
HSA System Attributes
=====================
Runtime Version: 1.1
System Timestamp Freq.: 1000.000000MHz
Sig. Max Wait Duration: 18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count)
Machine Model: LARGE
System Endianness: LITTLE
Mwaitx: DISABLED
DMAbuf Support: YES
==========
HSA Agents
==========
*******
Agent 1
*******
Name: AMD Ryzen 9 5900X 12-Core Processor
Uuid: CPU-XX
Marketing Name: AMD Ryzen 9 5900X 12-Core Processor
Vendor Name: CPU
Feature: None specified
Profile: FULL_PROFILE
Float Round Mode: NEAR
Max Queue Number: 0(0x0)
Queue Min Size: 0(0x0)
Queue Max Size: 0(0x0)
Queue Type: MULTI
Node: 0
Device Type: CPU
Cache Info:
L1: 32768(0x8000) KB
Chip ID: 0(0x0)
ASIC Revision: 0(0x0)
Cacheline Size: 64(0x40)
Max Clock Freq. (MHz): 3700
BDFID: 0
Internal Node ID: 0
Compute Unit: 24
SIMDs per CU: 0
Shader Engines: 0
Shader Arrs. per Eng.: 0
WatchPts on Addr. Ranges:1
Features: None
Pool Info:
Pool 1
Segment: GLOBAL; FLAGS: FINE GRAINED
Size: 131816192(0x7db5b00) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
Pool 2
Segment: GLOBAL; FLAGS: KERNARG, FINE GRAINED
Size: 131816192(0x7db5b00) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
Pool 3
Segment: GLOBAL; FLAGS: COARSE GRAINED
Size: 131816192(0x7db5b00) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
ISA Info:
*******
Agent 2
*******
Name: gfx1100
Uuid: GPU-69e905a91bab9202
Marketing Name: Radeon RX 7900 XTX
Vendor Name: AMD
Feature: KERNEL_DISPATCH
Profile: BASE_PROFILE
Float Round Mode: NEAR
Max Queue Number: 128(0x80)
Queue Min Size: 64(0x40)
Queue Max Size: 131072(0x20000)
Queue Type: MULTI
Node: 1
Device Type: GPU
Cache Info:
L1: 32(0x20) KB
L2: 6144(0x1800) KB
L3: 98304(0x18000) KB
Chip ID: 29772(0x744c)
ASIC Revision: 0(0x0)
Cacheline Size: 64(0x40)
Max Clock Freq. (MHz): 2304
BDFID: 11520
Internal Node ID: 1
Compute Unit: 96
SIMDs per CU: 2
Shader Engines: 6
Shader Arrs. per Eng.: 2
WatchPts on Addr. Ranges:4
Coherent Host Access: FALSE
Features: KERNEL_DISPATCH
Fast F16 Operation: TRUE
Wavefront Size: 32(0x20)
Workgroup Max Size: 1024(0x400)
Workgroup Max Size per Dimension:
x 1024(0x400)
y 1024(0x400)
z 1024(0x400)
Max Waves Per CU: 32(0x20)
Max Work-item Per CU: 1024(0x400)
Grid Max Size: 4294967295(0xffffffff)
Grid Max Size per Dimension:
x 4294967295(0xffffffff)
y 4294967295(0xffffffff)
z 4294967295(0xffffffff)
Max fbarriers/Workgrp: 32
Packet Processor uCode:: 550
SDMA engine uCode:: 19
IOMMU Support:: None
Pool Info:
Pool 1
Segment: GLOBAL; FLAGS: COARSE GRAINED
Size: 25149440(0x17fc000) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: FALSE
Pool 2
Segment: GLOBAL; FLAGS: EXTENDED FINE GRAINED
Size: 25149440(0x17fc000) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: FALSE
Pool 3
Segment: GROUP
Size: 64(0x40) KB
Allocatable: FALSE
Alloc Granule: 0KB
Alloc Alignment: 0KB
Accessible by all: FALSE
ISA Info:
ISA 1
Name: amdgcn-amd-amdhsa--gfx1100
Machine Models: HSA_MACHINE_MODEL_LARGE
Profiles: HSA_PROFILE_BASE
Default Rounding Mode: NEAR
Default Rounding Mode: NEAR
Fast f16: TRUE
Workgroup Max Size: 1024(0x400)
Workgroup Max Size per Dimension:
x 1024(0x400)
y 1024(0x400)
z 1024(0x400)
Grid Max Size: 4294967295(0xffffffff)
Grid Max Size per Dimension:
x 4294967295(0xffffffff)
y 4294967295(0xffffffff)
z 4294967295(0xffffffff)
FBarrier Max Size: 32
*** Done *** - Verify system management interface (smi)
rocm-smiResult:
====================================== ROCm System Management Interface ======================================
================================================ Concise Info ================================================
Device [Model : Revision] Temp Power Partitions SCLK MCLK Fan Perf PwrCap VRAM% GPU%
Name (20 chars) (Edge) (Avg) (Mem, Compute)
==============================================================================================================
0 [0x5200 : 0xc8] 45.0°C 74.0W N/A, N/A 42Mhz 1249Mhz 0% auto 291.0W 2% 2%
0x744c
==============================================================================================================
============================================ End of ROCm SMI Log =============================================- Install Python Packages
CppHeaderParser A Python library that parses C++ header files to facilitate easy automation of code generation or documentation preparation tasks.
argparse A standard Python library for parsing command-line arguments; useful for building user-friendly command-line tools.
pip3 install CppHeaderParser argparse- Clone ROCm-Developer-Tools Repository, roctracer is a tool for tracing ROCm APIs, facilitating debugging and performance analysis.
git clone -b amd-master https://github.com/ROCm-Developer-Tools/roctracer- Navigate to the cloned directory, changes the current directory to roctracer, where you can build the project and explore its contents.
cd roctracer- Install dependencies:
cmakeAn open-source, cross-platform family of tools designed to build, test, and package software.doxygenA documentation generator for various programming languages.gcc & g++The GNU Compiler Collection includes front ends for C and C++, among others.graphvizProvides a way of representing structural information as diagrams of abstract graphs and networks.libatomic1Provides support for atomic operations in GCC.libstdc++-12-devDevelopment package for the GNU Standard C++ Library version 12 (solves cmath not found error)makeA tool which controls the generation of executables and other non-source files of a program from the program's source files.rpmBuilder for rpms (Fixes RPM package requires rpmbuild executable).texlive-fullA comprehensive TeX document production system.
sudo apt-get install -y cmake doxygen gcc graphviz g++ libatomic1 libstdc++-12-dev make rpm texlive-full- Build the Project
bash build.sh- Change directory to build
cd build- Compile Custom Test (Optional)
make mytestResult:
[ 3%] Built target dlopen
[ 8%] Built target util
[ 22%] Built target roctracer
[ 26%] Built target roctracer_tool
[ 29%] Built target hip_stats
[ 33%] Built target roctx
[ 36%] Built target MatrixTranspose
[ 40%] Built target MatrixTranspose_test
[ 43%] Built target MatrixTranspose_hipaact_test
[ 47%] Built target MatrixTranspose_mgpu
[ 52%] Built target MatrixTranspose_ctest
[ 56%] Built target codeobj_test
[ 71%] Built target hsaco_targets
[ 75%] Built target copy
[ 78%] Built target roctx_test
[ 82%] Built target backward_compat_test
[ 85%] Built target load_unload_reload_test
[ 89%] Built target trace_buffer
[ 92%] Built target memory_pool
[ 96%] Built target activity_and_callback
[100%] Built target multi_pool_activities
[100%] Built target mytest- Use make to install build.
sudo make install- We’ll use the Mamba package manager to create the Python environment. You can learn more about it in my getting started tutorial.
The following bash commands will download the latest release, install it, and relaunch the current bash shell to apply the relevant changes:
wget "https://github.com/conda-forge/miniforge/releases/latest/download/Mambaforge-$(uname)-$(uname -m).sh"
bash Mambaforge-$(uname)-$(uname -m).sh -b
~/mambaforge/bin/mamba init
bash- Create a Python Environment in mamba
mamba create --name pytorch-rocm python=3.10 -y
mamba activate pytorch-rocm- Install pip dependencies for rocm.
Important! AMD recommends proceeding with ROCm WHLs available, noting the proper ROCm version at repo.radeon.com.
Here is a sample direction set that was referenced.
- Here is an example as of the writing of these directions:
pip3 install torch==2.1.1 torchvision==0.16.1 -f https://repo.radeon.com/rocm/manylinux/rocm-rel-6.0/ The ROCm WHLs available at PyTorch.org are not tested extensively by AMD as the WHLs change regularly when the nightly builds are updated.
- Using these steps from here. Note that I updated the rocm version in the url.
pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm6.0/Note: This may take several minutes.
- Verify if Pytorch is installed and detecting the GPU compute device.
python3 -c 'import torch' 2> /dev/null && echo 'Success' || echo 'Failure'Expected result:
Success- Enter command to test if the GPU is available.
python3 -c 'import torch; print(torch.cuda.is_available())'Expected result:
True- Enter command to display installed GPU device name.
python3 -c "import torch; print(f'device name [0]:', torch.cuda.get_device_name(0))"Expected result:
device name [0]: Radeon RX 7900 XTX- Enter command to display component information within the current PyTorch environment.
python3 -m torch.utils.collect_envExpected result (repo.radeon.com):
Collecting environment information...
PyTorch version: 2.1.1+rocm6.0
Is debug build: False
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: 6.0.32830-d62f6a171
OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.35
Python version: 3.10.13 | packaged by conda-forge | (main, Dec 23 2023, 15:36:39) [GCC 12.3.0] (64-bit runtime)
Python platform: Linux-6.5.0-18-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: Radeon RX 7900 XTX
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: 6.0.32830
MIOpen runtime version: 3.0.0
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 48 bits physical, 48 bits virtual
Byte Order: Little Endian
CPU(s): 24
On-line CPU(s) list: 0-23
Vendor ID: AuthenticAMD
Model name: AMD Ryzen 9 5900X 12-Core Processor
CPU family: 25
Model: 33
Thread(s) per core: 2
Core(s) per socket: 12
Socket(s): 1
Stepping: 2
Frequency boost: enabled
CPU max MHz: 4950.1948
CPU min MHz: 2200.0000
BogoMIPS: 7400.37
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 sse4_1 sse4_2 x2apic movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif v_spec_ctrl umip pku ospke vaes vpclmulqdq rdpid overflow_recov succor smca fsrm
Virtualization: AMD-V
L1d cache: 384 KiB (12 instances)
L1i cache: 384 KiB (12 instances)
L2 cache: 6 MiB (12 instances)
L3 cache: 64 MiB (2 instances)
NUMA node(s): 1
NUMA node0 CPU(s): 0-23
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; safe RET
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines, IBPB conditional, IBRS_FW, STIBP always-on, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] numpy==1.26.4
[pip3] torch==2.1.1+rocm6.0
[pip3] torchvision==0.16.1+rocm6.0
[conda] numpy 1.26.4 pypi_0 pypi
[conda] torch 2.1.1+rocm6.0 pypi_0 pypi
[conda] torchvision 0.16.1+rocm6.0 pypi_0 pypiEnvironment set-up is complete, and the system is ready for use with PyTorch to work with machine learning models, and algorithms.
- Download here.
PyCharm 2023.2.6 (Community Edition)
Build #PC-232.10300.41, built on February 14, 2024
Runtime version: 17.0.10+7-b1000.48 amd64
VM: OpenJDK 64-Bit Server VM by JetBrains s.r.o.
Linux 6.5.0-18-generic
GC: G1 Young Generation, G1 Old Generation
Memory: 1500M
Cores: 24
Registry:
ide.experimental.ui=true
Current Desktop: ubuntu:GNOME- Use the following command, onnce environment setup (i.e. conda, mamba etc.) to find the python path.
which python3- Use the following instruction set to set your python interpreter to use your python environment with the result of the last step.
The MNIST dataset is an iconic and foundational dataset in the machine learning community, primarily used for training and testing models in image processing, particularly for digit recognition tasks. It has played a critical role in the evolution of machine learning techniques and continues to be a valuable resource for education and benchmarking in the field.
Origin and Composition
- Source: The MNIST (Modified National Institute of Standards and Technology) dataset was created by Yann LeCun, Corinna Cortes, and Christopher J.C. Burges.
- Contents: It contains 70,000 images of handwritten digits (0 through 9).
- Training and Testing Split: Typically, it's split into a training set of 60,000 examples and a test set of 10,000 examples.
- Image Properties: Each image is a 28x28 pixel grayscale image. The digits have been size-normalized and centered in a fixed-size (28x28) image.
Purpose and Usage
- Benchmarking: It's widely used for training and testing in the field of machine learning, particularly for benchmarking algorithms for classification, image recognition, and computer vision.
- Educational Tool: It's often the first dataset used by beginners in machine learning and deep learning due to its simplicity and manageability.
Characteristics
- Simplicity: The simplicity of the dataset makes it ideal for testing new machine learning techniques or for educational purposes.
- Variability: Despite its simplicity, it includes a wide variety of handwriting styles, making it a robust dataset for training models to recognize handwritten digits.
Historical Significance
- Impact on Research: Since its creation, MNIST has been a cornerstone dataset for machine learning research. It has played a significant role in the development and evaluation of new machine learning techniques, especially in deep learning.
- Evolution of Use: Over time, as machine learning algorithms have become more advanced, the MNIST dataset is often considered too easy and is sometimes replaced by more complex datasets like CIFAR-10, CIFAR-100, or ImageNet for cutting-edge research.
Accessibility
- Open Access: The dataset is publicly available and can be easily accessed and used through various machine learning frameworks like TensorFlow, PyTorch, etc.
- Run the following command to test the hello world.
python pytorch_mnist_numbers.pyResult:
This is an experimnental Hello World using ROCm.
PyTorch Version: 2.1.1+rocm6.0
Torchvision Version: 0.16.1+rocm6.0
Using [cpu|cuda]: cuda
Using device: Radeon RX 7900 XTX
Devices Available 1
Epoch: 1
Test set: Average loss: 0.1299, Accuracy: 9608/10000 (96%)
Epoch: 2
Test set: Average loss: 0.0811, Accuracy: 9754/10000 (98%)
Epoch: 3
Test set: Average loss: 0.0629, Accuracy: 9801/10000 (98%)
Epoch: 4
Test set: Average loss: 0.0665, Accuracy: 9780/10000 (98%)
Epoch: 5
Test set: Average loss: 0.0480, Accuracy: 9842/10000 (98%)
Epoch: 6
Test set: Average loss: 0.0486, Accuracy: 9839/10000 (98%)
Epoch: 7
Test set: Average loss: 0.0409, Accuracy: 9870/10000 (99%)
Epoch: 8
Test set: Average loss: 0.0373, Accuracy: 9873/10000 (99%)
Epoch: 9
Test set: Average loss: 0.0383, Accuracy: 9866/10000 (99%)- Create a Python Environment in mamba
mamba create --name pytorch-rocm python=3.10 -y
mamba activate pytorch-rocm- Install pip dependencies for rocm.
- Create a Python Environment
mamba create --name tensorflow-rocm python=3.10 -y
mamba activate tensorflow-rocm- Install pip dependencies from nightly-rocmfork-develop-upstream:
pip3 install ${NIGHTLY_ARTIFACT}Use the following script to build from source, presently it builds from r2.,15-rocm-enhanced:
cd scripts/test-build-env-tensorflow
sudo bash build_tensorflow.sh $(rocm-smi --version | grep ROCM-SMI-LIB | awk '{print $NF}')- The following worked in the python terminal:
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())Result:
(tensorflow-rocm) flaniganp@amd-gpu-machine:~/Downloads/roctracer/build$ python
Python 3.10.13 | packaged by conda-forge | (main, Dec 23 2023, 15:36:39) [GCC 12.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from tensorflow.python.client import device_lib
2024-02-19 10:12:34.866913: E external/local_xla/xla/stream_executor/plugin_registry.cc:91] Invalid plugin kind specified: DNN
2024-02-19 10:12:34.902824: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: SSE3 SSE4.1 SSE4.2 AVX AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-02-19 10:12:35.190644: E external/local_xla/xla/stream_executor/plugin_registry.cc:91] Invalid plugin kind specified: DNN
>>> print(device_lib.list_local_devices())
2024-02-19 10:12:36.669985: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:12:36.689453: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:12:36.689504: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:12:36.690504: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:12:36.690550: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:12:36.690596: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:12:36.690622: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1928] Created device /device:GPU:0 with 23466 MB memory: -> device: 0, name: Radeon RX 7900 XTX, pci bus id: 0000:2d:00.0
[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 16497019316532335090
xla_global_id: -1
, name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 24605884416
locality {
bus_id: 1
links {
}
}
incarnation: 17729530124763235462
physical_device_desc: "device: 0, name: Radeon RX 7900 XTX, pci bus id: 0000:2d:00.0"
xla_global_id: 416903419
]- The following script will now work.
python tensorflow-rocm.pyResult:
(tensorflow-rocm) flaniganp@amd-gpu-machine:~/Documents/repos/amd-gpu-hello$ python tensorflow-rocm.py
2024-02-19 10:13:23.619015: E external/local_xla/xla/stream_executor/plugin_registry.cc:91] Invalid plugin kind specified: DNN
2024-02-19 10:13:23.658474: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: SSE3 SSE4.1 SSE4.2 AVX AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-02-19 10:13:23.971159: E external/local_xla/xla/stream_executor/plugin_registry.cc:91] Invalid plugin kind specified: DNN
TensorFlow version: 2.16.0.600-dev20240219
CUDA version: None
cuDNN version: None
2024-02-19 10:13:24.618800: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:24.640243: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:24.640300: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:24.640414: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
GPU Details: {'device_name': 'Radeon RX 7900 XTX'}
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-labels-idx1-ubyte.gz
29515/29515 ━━━━━━━━━━━━━━━━━━━━ 0s 1us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/train-images-idx3-ubyte.gz
26421880/26421880 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-labels-idx1-ubyte.gz
5148/5148 ━━━━━━━━━━━━━━━━━━━━ 0s 1us/step
Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/t10k-images-idx3-ubyte.gz
4422102/4422102 ━━━━━━━━━━━━━━━━━━━━ 0s 0us/step
Original Label: 9
One-Hot Encoded Label: [0. 0. 0. 0. 0. 0. 0. 0. 0. 1.]
Label: 4
Class Name: Coat
Image Shape: (28, 28)
Print the train image shape (60000, 28, 28).
Print the train labels shape (60000,).
2024-02-19 10:13:25.921355: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:25.921447: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:25.921482: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:25.921588: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:25.921630: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:25.921670: I external/local_xla/xla/stream_executor/rocm/rocm_executor.cc:874] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2024-02-19 10:13:25.921692: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1928] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 23450 MB memory: -> device: 0, name: Radeon RX 7900 XTX, pci bus id: 0000:2d:00.0
Training model...
Epoch 1/10
WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1708355607.215073 31946 service.cc:145] XLA service 0x7f1f58003d30 initialized for platform ROCM (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1708355607.215104 31946 service.cc:153] StreamExecutor device (0): Radeon RX 7900 XTX, AMDGPU ISA version: gfx1100
2024-02-19 10:13:27.235619: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:268] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.
I0000 00:00:1708355608.478721 31946 device_compiler.h:187] Compiled cluster using XLA! This line is logged at most once for the lifetime of the process.
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 5s 2ms/step - accuracy: 0.7301 - loss: 0.7512 - val_accuracy: 0.8580 - val_loss: 0.4015
Epoch 2/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.8678 - loss: 0.3665 - val_accuracy: 0.8736 - val_loss: 0.3542
Epoch 3/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.8809 - loss: 0.3262 - val_accuracy: 0.8854 - val_loss: 0.3259
Epoch 4/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.8957 - loss: 0.2883 - val_accuracy: 0.8909 - val_loss: 0.3089
Epoch 5/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.9013 - loss: 0.2688 - val_accuracy: 0.8925 - val_loss: 0.3001
Epoch 6/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.9080 - loss: 0.2539 - val_accuracy: 0.8993 - val_loss: 0.2838
Epoch 7/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.9118 - loss: 0.2377 - val_accuracy: 0.8985 - val_loss: 0.2798
Epoch 8/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.9186 - loss: 0.2243 - val_accuracy: 0.8993 - val_loss: 0.2809
Epoch 9/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.9229 - loss: 0.2105 - val_accuracy: 0.8986 - val_loss: 0.2837
Epoch 10/10
1875/1875 ━━━━━━━━━━━━━━━━━━━━ 3s 1ms/step - accuracy: 0.9246 - loss: 0.2056 - val_accuracy: 0.9066 - val_loss: 0.2697
Evaluating model...
313/313 ━━━━━━━━━━━━━━━━━━━━ 0s 700us/step - accuracy: 0.9082 - loss: 0.2783
WARNING:absl:You are saving your model as an HDF5 file via `model.save()` or `keras.saving.save_model(model)`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')` or `keras.saving.save_model(model, 'my_model.keras')`.