Annotated Bibliography: NVIDIA vs AMD

Introduction

Nvidia and Advanced Micro Devices(AMD) are two top vendors of the graphics processing unit(GPU) in the world today. The preference for a GPU from either of the firms is determined by a number of factors and consumer needs, including performance and efficiency rates. The following annotated bibliography lays a foundation for a comparative analysis of Nvidia and AMD as GPU producers.

Is your time best spent reading someone else’s essay? Get a 100% original essay FROM A CERTIFIED WRITER!

Save 25%

Annotated Bibliography

Chauhan, Harsh. "AMD vs. NVIDIA: Here's How the GPU War Is Playing Out". 19 Sept 2017. The Motley Fool, https://www.fool.com/investing/2017/09/19/amd-vs-nvidia-heres-how-the-gpu-war-is-playing-out.aspx (Accessed 19 Feb 2019).

In this news article, Chauhan examines the performance of both AMD and NVidia at the market. The author places emphasis on the market share and the ability of the two firms to meet consumer demands. On average, Chauhan observes that AMD is providing value-for-money since it delivers high performance at affordable costs. For the reasons mentioned, AMD is taking the larger market share in comparison to NVidia. Of the discrete GPU market, AMD takes 29.4%, which is higher than of NVidia. Apparently, it also seems that AMD has incorporated the latest trends into its GPUs, as in the case of its new Vega 56 GPU. According to Chauhan, Vega 56, though not superior to NVidia's GTX 1070, delivers 90% of the performance at 80% of the cost of GTX 1070.

From the article, it is clear that the comparative analysis between AMD and NVidia ends at the market. Through the analysis of how various GPU products are performing within the product market, the author will seek to create an understanding on the effectiveness of the architectural designs of different GPUs offered by both AMD and NVidia. This will also enable the author to establish the correlation between trends and the marketability of GPUs.

Elteir, Marwa, et al. "StreamMR: an optimized MapReduce framework for AMD GPUs." 2011 IEEE 17th International Conference on Parallel and Distributed Systems. IEEE, 2011. Retrieved from https://www.academia.edu/7738722/StreamMR_An_Optimized_MapReduce_Framework_for_AMD_GPUs

Elteir, Marwa, et al. aimed at revisiting the design of MapReduce framework on AMD GPUs. The article found out that the atomic operations that are used in high end GPUs MapReduce framework can result to servere performance reductionon AMD GPUs because of their significant architectural dissimilarities with NVIDIA GPUs. The article stated that the existing MapReduce implementations on GPUs primarily apply on NVIDIA GPUs. Therefore, the design improvement methods for this particular implementation might not be applicable to the AMD GPUs because they have distinctively varying architecture than NVIDIA GPUs. The article states the difference between NVIDIA and AMD based on the MapReduce dependence on atomic operation to execute coordination which is different for both models. The article proposes StreamMR, which is an OpenCL MapReduce framework that is designed for AMD GPUs to provide an efficient atomic-free algorithm for coordinating output from various threads.

Gao, Yuxiang and Peng Zhang. "A Survey of Homogeneous and Heterogeneous System Architectures in High-Performance Computing". IEEE International Conference on Smart Cloud (SmartCloud). 2016, https://ieeexplore.ieee.org/abstract/document/7796169/ (Accessed 19 Feb 2019).

Acknowledging the role played by high-performance computing systems in the architectural design of supercomputing systems, this article presents features of homogenous and heterogeneous interconnected systems in high-performance computing systems. The authors argue that while general-purpose GPU(GPGPU) are based on parallel computing, there is a trend toward the heterogeneous computing model. They cite Linpack and power efficiency as some of the factors that are driving these trends, making heterogeneous computing a more attractive solution for producers of GPU. The results of the comparative analysis between 25% of Top 100 and Top 500 supercomputers indicate that progress I the computing power is also another key factor shaping the trend in the industry.

This article is based on a comprise analysis that is well-informed and which has been undertaken by experts in the field. The highlighted trends in the industry among GPU manufacturers as well as influential factors will provide a good foundation for gaining insights into which of the two firms to be compared - NVidia and AMD - is keen on the industry trends. More specifically, the article will help in arguing for a heterogeneous computing model all while taking into consideration the efforts made by each firm to incorporate the idea or trend into their respective GPU.

Harvey, Matt J., and Gianni De Fabritiis. "Swan: A tool for porting CUDA programs to OpenCL." Computer Physics Communications 182.4 (2011): 1093-1099. Retrieved from https://doi.org/10.1016/j.cpc.2010.12.052

The authors present a source-to-source translation tool that is appropriate for simplifying and reducing hardware dependence of the tasks that use CUDA programming model that is primarily supported by GPUs designed by NVIDIA. The article highlights that this "swan" facilitates the conversion of CUDA code that is NVIDIA based to OpenCl model used by the AMD. This is presented as a way of assisting programmers who are more experienced in NVIDIA in evaluating AMD or other alternative hardware. The article highlights that the kernel in CUDA and OpenCL are largely identical, however, the launch of a kernel in OpenCL is more hardware independent while in CUDA it is multifaceted because it is executed by its driver and runtime. This article is distinctive in exploring the difference in programming model between NVIDIA and AMD, GPUs from NVIDIA are primarily programed with CUDA code while AMD is based on OpenCL programming that has almost the same capabilities. Therefore, "swan" presented in the article completes the translation of CUDA kernel source code into an OpenCL equivalent. Moreover, the article also established that "swan" simplifies the kernel invocation but generating the C source codes for point entry.

Parrish, K. "The Nvidia vs. AMD Battle Has Only Got More Interesting in 2018." Digital Trends, Digital Trends, 6 Dec. 2018, www.digitaltrends.com/computing/nvidia-vs-amd/.

In this article, the author compares Nvidia GPUS to AMD GPUs from the top end to the entry level of the GPU market. At the top end, the author compares Nvidias, RTX 2080 Ti to AMD's Radeon RX Vega 64. With regards to price, the author approximates the RTX 2080 Ti to cost about $1000 to $1200 while the RAdeon RX Vega 64 is estimated to cost between $400 and $500. In terms of performance, the RTX 2080 Ti was found to be better. At the midrange category, the author finds that AMD's Radeon RX 580 and RX 570 outperform Nvidia's GEforce GTX 1060 and are also quite cheaper making AMD the better midrange GPU manufacturer. In the laptop GPU market, the author reports that Nvidia is again the better performer as in his testing, the RAdeon RX Vega M GL fell below Nvidia's vanilla GTX 1050 chips

Shimobaba, Tomoyoshi, et al. "Fast calculation of computer-generated-hologram on AMD HD5000 series GPU and OpenCL." Optics express 18.10 (2010): 9955-9960.

The authors of the article aimed at reporting fast CGH calculation using RV870 GPU and OpenCL by calculating 1,9201,024 resolution obtained from a 3D object that consisted of 1024 points in 30 ms. Moreover, the article compares the calculation performance speed between AMD RV870 GPU and the one developed by NVIDIA. The authors highlighted the loop unrolling is a universally known technique for optimizing kernel function. This can be realized by significantly reducing the number of repetitions and duplications of the body of the loop. The loop frequency could be done through loop unrolling. The article will play a significant role in helping the author to determine which of the two vendors of GPUs is fast in CGH calculation. This article is distinctive in exploring the performance between the RV870 GPU and the GPU made by NVIDIA.

Stegailov, Vladimir, and Vyacheslav Vecher. "Efficiency analysis of Intel, AMD and Nvidia 64-bit hardware for memory-bound problems: a case study of ab initio calculations with VASP." International Conference on Parallel Processing and Applied Mathematics. Springer, Cham, 2017.

The article discusses the VASP as a benchmark and result based test model that offers the option of comparing between various CPUs and opt for the best regarding the time-to-solution and energy-to-solution criteria. The article presents the efficiency analysis of an Intel based model AMD x86 64 CPUs using the VASP model. This article is relevant to the research as it will assist in selecting the best CPU. The article states that AMD released Zen which were the first processors based on the novel x86_64 architecture and assumed that the model for HPC applications would be comparable to the modern Intel. However, the various types of CPUs complicates the choice for the ultimate HPC system. The primary standard is mainly the time-to-solution of the specific computational task

Vanek, Jan, et al. "Optimized acoustic likelihoods computation for NVIDIA and ATI/AMD graphics processors." IEEE Transactions on Audio, Speech, and Language Processing 20.6 (2012): 1818-1828. Retrieved from DOI: 10.1109/TASL.2012.2190928

The main aim of the article was to create a high speed GPU implementation that primary achieves the computing power the GPUs of both NVIDIA and AMD. The authors aimed at establishing this to ascertain the possibility of utilizing the actual LVCSR systems and other related speech applications in both new devices and laptops. Moreover, the developed GPU could be shared amongst extra recognizers and other affiliated applications.

The authors describe an optimized version of a Gaussian-mixture-based acoustic model likelihood evaluation algorithm for GPUs. The approach used by the authors is unique compared to previous studies because they use a GPU programming framework in a much more efficient way. According to the author, NVIDIA and AMD are quite similar, although both NVIDIA and AMD consist of multiple Processing Elements (PE), AMD PEs internal architecture is different compared to that of NVIDIA. The AMD PE contains 16 stream core that are each equipped with five stream processors, making it have a superior computational performance compared the NVIDIA GPUs. The article concluded that the AMD GPUs outperformed the NVIDIA GPUs in implementation of acoustic model likelihoods computation. Therefore, based on the results, the article recommended that during speech recognition, it is probable to use any larger acoustic models that can be receive constant training.

Vallero, Alessandro, Dimitris Gazopoulos, and Stefano Di Carlo. "SIFI: AMD southern islands GPU microarchitectural level fault injector". https://ieeexplore.ieee.org/abstract/document/8046209/ IEEE 23rd International Symposium on On-Line Testing and Robust System Design (IOLTS). (2017): pp.138-144,

The article demonstrates how to test for the reliability of GPUs in several application domains. The authors observe that GPGPU speed up data parallel workloads, thereby leveraging the computational power of GPUs. Using SIFI (Southern Islands Fault Injector) as a reliability evaluation framework for the soft-errors found in AMD GPUs built on microarchitectural design, Micro2Sim, the authors prove that SIFI can be used to support decisions regarding the best architectural parameters for any given application.

What is important from the article for the current paper is its reference to the different reliability metrics. That is, the article basically provides metrics for the reliability of GPUs, making it easier for the author to point out with certainty performance indicators of GPUs from the two firms of AMD and NVidia. In particular, the art...