skip to main content

Title: AMD GPUs as an Alternative to NVIDIA for Supporting Real-Time Workloads
Graphics processing units (GPUs) manufactured by NVIDIA continue to dominate many fields of research, including real-time GPU-management. NVIDIA’s status as a key enabling technology for deep learning and image processing makes this unsurprising, especially when combined with the company’s push into embedded, safety-critical domains like autonomous driving. NVIDIA’s primary competitor, AMD, has received comparatively little attention, due in part to few embedded offerings and a lack of support from popular deep-learning toolkits. Recently, however, AMD’s ROCm (Radeon Open Compute) software platform was made available to address at least the second of these two issues, but is ROCm worth the attention of safety-critical software developers? In order to answer this question, this paper explores the features and pitfalls of AMD GPUs, focusing on contrasting details with NVIDIA’s GPU hardware and software. We argue that an open software stack such as ROCm may be able to provide much-needed flexibility and reproducibility in the context of real-time GPU research, where new algorithmic or analysis techniques should typically remain agnostic to the underlying GPU architecture. In support of this claim, we summarize how closed-source platforms have obstructed prior research using NVIDIA GPUs, and then demonstrate that AMD may be a viable alternative by modifying components of the ROCm software stack to implement spatial partitioning. Finally, we present a case study using the PyTorch deep-learning framework that demonstrates the impact such modifications can have on complex real-world software.  more » « less
Award ID(s):
1717589 1837337
Author(s) / Creator(s):
Date Published:
Journal Name:
Proceedings of the 32nd Euromicro Conference on Real-Time Systems
Page Range / eLocation ID:
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Obeid, I. ; Selesnik, I. ; Picone, J. (Ed.)
    The Neuronix high-performance computing cluster allows us to conduct extensive machine learning experiments on big data [1]. This heterogeneous cluster uses innovative scheduling technology, Slurm [2], that manages a network of CPUs and graphics processing units (GPUs). The GPU farm consists of a variety of processors ranging from low-end consumer grade devices such as the Nvidia GTX 970 to higher-end devices such as the GeForce RTX 2080. These GPUs are essential to our research since they allow extremely compute-intensive deep learning tasks to be executed on massive data resources such as the TUH EEG Corpus [2]. We use TensorFlow [3] as the core machine learning library for our deep learning systems, and routinely employ multiple GPUs to accelerate the training process. Reproducible results are essential to machine learning research. Reproducibility in this context means the ability to replicate an existing experiment – performance metrics such as error rates should be identical and floating-point calculations should match closely. Three examples of ways we typically expect an experiment to be replicable are: (1) The same job run on the same processor should produce the same results each time it is run. (2) A job run on a CPU and GPU should produce identical results. (3) A job should produce comparable results if the data is presented in a different order. System optimization requires an ability to directly compare error rates for algorithms evaluated under comparable operating conditions. However, it is a difficult task to exactly reproduce the results for large, complex deep learning systems that often require more than a trillion calculations per experiment [5]. This is a fairly well-known issue and one we will explore in this poster. Researchers must be able to replicate results on a specific data set to establish the integrity of an implementation. They can then use that implementation as a baseline for comparison purposes. A lack of reproducibility makes it very difficult to debug algorithms and validate changes to the system. Equally important, since many results in deep learning research are dependent on the order in which the system is exposed to the data, the specific processors used, and even the order in which those processors are accessed, it becomes a challenging problem to compare two algorithms since each system must be individually optimized for a specific data set or processor. This is extremely time-consuming for algorithm research in which a single run often taxes a computing environment to its limits. Well-known techniques such as cross-validation [5,6] can be used to mitigate these effects, but this is also computationally expensive. These issues are further compounded by the fact that most deep learning algorithms are susceptible to the way computational noise propagates through the system. GPUs are particularly notorious for this because, in a clustered environment, it becomes more difficult to control which processors are used at various points in time. Another equally frustrating issue is that upgrades to the deep learning package, such as the transition from TensorFlow v1.9 to v1.13, can also result in large fluctuations in error rates when re-running the same experiment. Since TensorFlow is constantly updating functions to support GPU use, maintaining an historical archive of experimental results that can be used to calibrate algorithm research is quite a challenge. This makes it very difficult to optimize the system or select the best configurations. The overall impact of all of these issues described above is significant as error rates can fluctuate by as much as 25% due to these types of computational issues. Cross-validation is one technique used to mitigate this, but that is expensive since you need to do multiple runs over the data, which further taxes a computing infrastructure already running at max capacity. GPUs are preferred when training a large network since these systems train at least two orders of magnitude faster than CPUs [7]. Large-scale experiments are simply not feasible without using GPUs. However, there is a tradeoff to gain this performance. Since all our GPUs use the NVIDIA CUDA® Deep Neural Network library (cuDNN) [8], a GPU-accelerated library of primitives for deep neural networks, it adds an element of randomness into the experiment. When a GPU is used to train a network in TensorFlow, it automatically searches for a cuDNN implementation. NVIDIA’s cuDNN implementation provides algorithms that increase the performance and help the model train quicker, but they are non-deterministic algorithms [9,10]. Since our networks have many complex layers, there is no easy way to avoid this randomness. Instead of comparing each epoch, we compare the average performance of the experiment because it gives us a hint of how our model is performing per experiment, and if the changes we make are efficient. In this poster, we will discuss a variety of issues related to reproducibility and introduce ways we mitigate these effects. For example, TensorFlow uses a random number generator (RNG) which is not seeded by default. TensorFlow determines the initialization point and how certain functions execute using the RNG. The solution for this is seeding all the necessary components before training the model. This forces TensorFlow to use the same initialization point and sets how certain layers work (e.g., dropout layers). However, seeding all the RNGs will not guarantee a controlled experiment. Other variables can affect the outcome of the experiment such as training using GPUs, allowing multi-threading on CPUs, using certain layers, etc. To mitigate our problems with reproducibility, we first make sure that the data is processed in the same order during training. Therefore, we save the data from the last experiment and to make sure the newer experiment follows the same order. If we allow the data to be shuffled, it can affect the performance due to how the model was exposed to the data. We also specify the float data type to be 32-bit since Python defaults to 64-bit. We try to avoid using 64-bit precision because the numbers produced by a GPU can vary significantly depending on the GPU architecture [11-13]. Controlling precision somewhat reduces differences due to computational noise even though technically it increases the amount of computational noise. We are currently developing more advanced techniques for preserving the efficiency of our training process while also maintaining the ability to reproduce models. In our poster presentation we will demonstrate these issues using some novel visualization tools, present several examples of the extent to which these issues influence research results on electroencephalography (EEG) and digital pathology experiments and introduce new ways to manage such computational issues. 
    more » « less
  2. In recent years, we have been enhancing and updating gem5’s GPU support, including enhanced gem5’s GPU support to enable running ML workloads. Moreover, we created, validated, and released a Docker image with the proper software and libraries needed to run AMD’s GCN3 and Vega GPU models in gem5. With this container, users can run the gem5 GPU model, as well as build the ROCm applications that they want to run in the GPU model, out of the box without needing to properly install the appropriate ROCm software and libraries. Additionally, we updated gem5 to make it easier to reproduce results, including releasing support for a number of GPU workloads in gem5-resources and enabling continuous integration testing for a variety of GPU workloads. Current gem5 support focuses on Carrizo- and Vega-class GPUs. Unfortunately, these models do not always provide high accuracy relative to the equivalent ”real” GPUs. This leads to a mismatch in expectations: when prototyping new optimizations in gem5 users may draw the wrong conclusions about the efficacy of proposed optimizations if gem5’s GPU models do not provide high fidelity. Accordingly, to help bridge this divide, we design a series of micro-benchmarks designed expose the latencies, bandwidths, and sizes of a variety of GPU components on real GPUs. By iteratively applying fixes and improvements to gem’s GPU model, we significantly improve its fidelity relative to real AMD GPUs. 
    more » « less
  3. In recent years, we have been enhancing and updating gem5's GPU support. First, we have enhanced gem5’s GPU support for ML workloads such that gem5 can now run. Moreover, as part of this support, we created, validated, and released a Docker image that contains the proper software and libraries needed to run GCN3 and Vega GPU models in gem5. With this container, users can run the gem5 GPU model, as well as build the ROCm applications that they want to run in the GPU model, out of the box without needing to properly install the appropriate ROCm software and libraries. Additionally, we have updated gem5 to make it easier to reproduce results, including releasing support for a number of GPU workloads in gem5-resources and enabling continuous integration testing on future GPU commits. However, we currently do not have a way to model validated gem5 configurations for the most recent AMD GPUs. Current support focuses on Carrizo- and Vega-class GPUs. Unfortunately, these models do not always provide high accuracy relative to real GPU runs. This leads to a mismatch between how each instruction is supposedly being executed according to the ISA and how a given GPU model executes a given instruction. These discrepancies are of interest to those developing the gem5 GPU models as they can lead to less accurate simulations. Accordingly, to help bridge this divide, we have created a new tool, GAP (gem5 GPU Accuracy Profiler), to identify discrepancies between real GPU and simulated gem5 GPU behavior. GAP identifies and verifies how accurate these configurations relative to real GPUs by comparing the simulator’s performance counters to those from real GPUs. 
    more » « less
  4. null (Ed.)
    Self-driving systems execute an ensemble of different self-driving workloads on embedded systems in an end-to-end manner, subject to functional and performance requirements. To enable exploration, optimization, and end-to-end evaluation on different embedded platforms, system designers critically need a benchmark suite that enables flexible and seamless configuration of self-driving scenarios, which realistically reflects real-world self-driving workloads’ unique characteristics. Existing CPU and GPU embedded benchmark suites typically (1) consider isolated applications, (2) are not sensor-driven, and (3) are unable to support emerging self-driving applications that simultaneously utilize CPUs and GPUs with stringent timing requirements. On the other hand, full-system self-driving simulators (e.g., AUTOWARE, APOLLO) focus on functional simulation, but lack the ability to evaluate the self-driving software stack on various embedded platforms. To address design needs, we present Chauffeur, the first open-source end-to-end benchmark suite for self-driving vehicles with configurable representative workloads. Chauffeur is easy to configure and run, enabling researchers to evaluate different platform configurations and explore alternative instantiations of the self-driving software pipeline. Chauffeur runs on diverse emerging platforms and exploits heterogeneous onboard resources. Our initial characterization of Chauffeur on different embedded platforms – NVIDIA Jetson TX2 and Drive PX2 – enables comparative evaluation of these GPU platforms in executing an end-to-end self-driving computational pipeline to assess the end-to-end response times on these emerging embedded platforms while also creating opportunities to create application gangs for better response times. Chauffeur enables researchers to benchmark representative self-driving workloads and flexibly compose them for different self-driving scenarios to explore end-to-end tradeoffs between design constraints, power budget, real-time performance requirements, and accuracy of applications. 
    more » « less
  5. Random Forests (RFs) are a commonly used machine learning method for classification and regression tasks spanning a variety of application domains, including bioinformatics, business analytics, and software optimization. While prior work has focused primarily on improving performance of the training of RFs, many applications, such as malware identification, cancer prediction, and banking fraud detection, require fast RF classification. In this work, we accelerate RF classification on GPU and FPGA. In order to provide efficient support for large datasets, we propose a hierarchical memory layout suitable to the GPU/FPGA memory hierarchy. We design three RF classification code variants based on that layout, and we investigate GPU- and FPGA-specific considerations for these kernels. Our experimental evaluation, performed on an Nvidia Xp GPU and on a Xilinx Alveo U250 FPGA accelerator card using publicly available datasets on the scale of millions of samples and tens of features, covers various aspects. First, we evaluate the performance benefits of our hierarchical data structure over the standard compressed sparse row (CSR) format. Second, we compare our GPU implementation with cuML, a machine learning library targeting Nvidia GPUs. Third, we explore the performance/accuracy tradeoff resulting from the use of different tree depths in the RF. Finally, we perform a comparative performance analysis of our GPU and FPGA implementations. Our evaluation shows that, while reporting the best performance on GPU, our code variants outperform the CSR baseline both on GPU and FPGA. For high accuracy targets, our GPU implementation yields a 5-9 × speedup over CSR, and up to a 2 × speedup over Nvidia’s cuML library. 
    more » « less