CUDA輕鬆上手 新世代GPU應用技術 台灣第一本CUDA中文書籍

本書主要希望讓對於CUDA技術有興趣的讀者能在短時間內建置相關的軟硬體環境，並進一步嘗試CUDA平行程式的設計


此外、本書亦可以做為CUDA技術的入門書籍，可以讓初入門者了解圖形處理器以及CUDA技術的發展；對於業界而言，則可以做為CUDA技術的使用手冊，讓工程師能在短時間內建置CUDA的軟硬體環境以從事後續的開發工作

 

林俊淵、周嘉奕、林郁翔
李昇達、陳昱蓉、黃宣穎、李天齡   |著
松崗出版



    
第一章 | 筆者淺談了平行計算的背景，包含了多電腦系統架構以及多核心處理器架構，也介紹了MPI、Pthrad以及OpenMP等規範，接著筆者介紹本書的主角--圖形處理器以及CUDA，並說明了圖形處理器與中央處理器的差異以及CUDA與MPI、OpenMP的比較，最後說明了CUDA的演進、現況與未來發展。

第二章 | 筆者介紹了支援CUDA的相關圖形處理器產品，讓讀者們得以了解自身的需求所需購置的硬體設備，接著筆者介紹了CUDA技術相關的Driver、Runtime API及編輯器，也說明了CUDA技術的不同計算能力版本，讓讀者們了解所購置的硬體設備所需的軟體工具版本。最後，筆者也介紹了其它支援CUDA技術的套件及軟體工具

第三章 | 筆者首先先介紹如何安裝NVIDIA公司的顯示卡裝置，並提醒讀者們在挑選周邊硬體設備所需注意的事項以及相關的網路資源網站，後來，筆者介紹如何取得CUDA技術相關的Driver、toolkit以及SDK，並說明作業系統上的注意事項，最後筆者介紹了CUDA Driver及CUDA Toolkit的安裝步驟。

第四章 | 介紹CUDA SDK的安裝步驟，操作和展示CUDA SDK的範例程式，並詳細介紹常用的"DeviceQuery", "Bandwidth test"程式的使用方法

第五章 | 介紹如何開始撰寫CUDA平行程式，並說明如何編譯及執行CUDA範例程式，最後告訴讀者們如何利用除錯工具來檢視所撰寫的CUDA平行程式。

2011台灣CUDA程式設計比賽報名網站上線(2011 Taiwan CUDA Contest) http://nvidia.ithome.com.tw/cuda/index.html

紐約GPU圖形處理器高速運算”鄉民聚會

紐約GPU圖形處理器高速運算”鄉民聚會”

GPU MEETUP IN THE BIG APPLE

(Amdrew Sheppard的文章)
BY ANDREW SHEPPARD Posted on Feb 3 2011 at 04:22:13 PM in Software,Supercomputing

GPU圖形處理器高速運算真的是越來越有趣了.

These are interesting times for the GPU.

在紐約這個大都會中, 身兼金融顧問與CUDA開發者的我, 看到了GPU圖形處理器在高效能運算領域與日俱進的應用. 因此, 上個月我決定為致力於GPU圖形處理器運算的紐約人, 發起一場GPU聚會.

As a financial consultant and CUDA developer in metro New York, I see increasing momentum in the area of GPUs for high-performance computing. Last month I decided to launch a “meet-up” event for people interested in this topic.

因為這是第一次發起聚會, 我不能確定能達成如何的結果. 我們把微軟要幫我們舉辦聚會活動訊息發布出去, 在一月中舉辦的首次會議很快地獲得超過25位來自各領域GPU圖形運算人員的出席 – 這包括在GPU圖形處理器運算上, 跨越財務, 醫學, 科學與其他各領域的人員, 從專家到新手都有. 藉由GPU圖形處理器高效能運算PC把大家串聯在一起, 更快更好的解決學術與產業最尖端的問題與應用. 直到二月初, 這個團體的人數已經增加到50人.

Since this was my first time using meetup, I wasn’t sure what to expect. We put the word out in mid-January and at our first meeting, held at Microsoft, we had more than 25 attendees representing a wide range of people, from GPU experts to newbies, across finance, medical, science and other industries.  What drew all of us together is an enthusiasm for HPC and GPU technologies and how they can be applied to cutting-edge problems – doing things better and faster for more challenging applications. Today, after just four weeks, the group has grown to over 50.

在我們的首次會議, 我們的演講者是Dídac Artés, 也是首次會議的共同發起人. Dídac討論會員的權益與責任, 並很快的交換了團隊的方向與組織大綱.

這次會議也討論GPU圖形處理器高速運算的領域範圍及其應用, 舉例來說, 像是財務上知名的智慧高頻率交易(High Frequency Trading, HFT)應用.

At the first meeting, our speaker was Dídac Artés, a co-organizer. Dídac talked about opportunities for members to both contribute to and benefit from the group. Then the discussion got down to brass tacks and a freewheeling exchange of ideas on the aims of the group and how it should be organized.

The discussion touched upon topical aspects related to the domain specific issues relating to GPU technology also bubbled up, such as the application of the GPU to such hot areas as High Frequency Trading (or HFT, as it is known).

這次聚會也鞭策大家討論未來的主題, 例如”改用平行運算來思考與解決問題”與”GPU圖形處理器遇上無限帶寬”. 接下來, 聚會將以每月挑定某天晚上6點開始, 前半小時讓與會者交融與回顧上次會議內容, 接著到晚上8點半的時間由當月選定的演講者來講解大家有興趣的內容.

The meetup spurred discussion of future topics, such as “Thinking in Parallel” and “Infiniband+GPU.” Moving forward, the format will be monthly meetups starting at 6:00 pm, with the first half hour set aside to mingle and catch up, followed by a speaker talking about a topic of common interest.

這個團體邀請把GPU圖形處理器高速運算應用於真實世界難題的專家 – 真正在前線的實務應用者, 這樣明確的學習方向能吸引與會者勤於出席. 這晚上6點半到8點半的兩小時進行實際操作, 程式碼教學與分享, 白板教學/討論, 意見交流, 最後在8點半進行總結.

The group has a distinct leaning towards inviting practitioners to present, in the sense of people who are applying GPU technology to real world problems – who are “in the trenches”, so to speak. Time will be set aside for hands-on practice, code sharing, white boarding and social networking, with things wrapping up around 8:30 pm.

我希望所有的GPU圖形處理器運算專家們, 不論你是住在紐約大都會地區, 或者只是途經紐約, 能夠留意一下這個GPU圖形處理器高速運算團體, 並撥空參與聚會. http://www.meetup.com/HPC-GPU-Supercomputing-Group-of-New-York.

I hope any GPU enthusiasts who live in the New York metro area, or are just visiting New York, will check out the group and consider attending a future meetup: http://www.meetup.com/HPC-GPU-Supercomputing-Group-of-New-York.

GPU圖形處理器高速運算真的是越來越有趣了. 抱著愉悅的心情來學習與分享吧.

Interesting times for the GPU. Come share in the fun!

註: 我們剛剛在波士頓也成立了一個類似的GPU圖形處理器運算活動團體, 首次會議將於3月1日舉行. http://www.meetup.com/HPC-GPU-Supercomputing-Group-of-Boston.

PS: We’ve just launched a similar group in Boston, which will have its first meeting on March 1:  http://www.meetup.com/HPC-GPU-Supercomputing-Group-of-Boston.

匹茲堡大學研究員致力於”GPU圖形處理器加速達數百倍快” (SEVERAL HUNDREDFOLD)的分子動力學研究

匹茲堡大學研究員致力於”GPU圖形處理器加速達數百倍快” (Several hundredfold)的分子動力學研究

UNIVERSITY OF PITTSBURGH RESEARCHER STUDIES MOLECULAR DYNAMICS WITH HELP OF GPUS

BY CALISA COLE Posted on Jan 13 2011 at 03:56:24 PM

約書亞研究藉由GPU圖形處理器高速運算的威力, 加速進一步了解例如糖尿病與癲癇等常見疾病的生物現象. 我們是從匹茲堡大學的Grabe Lab知道了他的研究, 並特別感興趣於他的研究居然能獲得FundScience, 一個專門挹注資金於獲得同行互評為前瞻研究以期達到從大眾取經效果的非營利性組織, 這樣一流的公眾科研投資單位挹注資金, 真不簡單. 我們找到了約書亞, 並做了個簡短的訪問:

Joshua Adelman is studying biological phenomena to better understand common diseases such as diabetes and epilepsy, leveraging the computational power of GPUs. We recently discovered his work through the Grabe Lab at the University of Pittsburgh, and were particularly interested when we learned he’s funding his project through peer-reviewed, “crowd-source” funding organization, FundScience. We caught up with Joshua, and had a chance to interview him to learn more:

NVIDIA: 約書亞, 能告訴我們你的研究方向嗎?

NVIDIA: Joshua, tell us about your research.

約書亞: 我研究細胞內蛋白質如何有選擇性的通過細胞膜傳遞微小分子. 特別來說, 我研究兩種物質傳遞 – 一種是神經遞質(神經傳遞物質), 這是透過突觸傳遞對於特定神經系統的關鍵作用; 而另一種是腸道與腎臟如何進行吸收糖分子的吸收. 這兩種研究都與肌肉萎縮症(ALS), 癲癇, 與第二型糖尿病未來的新治療方向有關係.

Joshua: I am interested in understanding how proteins that sit in the cell membrane selectively transport small molecules across the membrane. Specifically, I study two transporters – one removes a neurotransmitter from the synapse and is critical in proper nervous system function; the other is responsible for absorption of sugar in the intestines and kidneys. Both are potential targets for treating a number of diseases including ALS, epilepsy and type 2 diabetes.

 Joshua Adelman (seated) with advisor Michael Grabe

NVIDIA: 你如何使用GPU圖形處理器高速運算?

NVIDIA: How are you leveraging GPU computing?

約書亞: 我用分子動力學進行蛋白質模擬, 以了解每個幫助物質傳遞方案所產生的結構性變化. 已經有很多用GPU圖性高速運算於蛋白質研究的例子, 證實了這個方法前途有望. 我只是站在這個基礎上, 用了OpenMM API, 這個用NVIDIA CUDA與OpenCL高速平行運算的應用程式介面, 以GPU圖性高速運算做到分子動力學模擬所需要的各種演算.

Joshua: I use molecular dynamics to simulate these proteins to understand how structural changes in each facilitate transport. There has been considerable effort over the last couple of years by several groups to use GPUs to accelerate these types of calculations. The initial results are quite promising. I’m piggybacking off of one of these efforts, the OpenMM API which provides CUDA and OpenCL implementations of key algorithms necessary to run molecular dynamics simulations.

 Molecular model of a glutamate transporter

NVIDIA: 那麼你達成了什麼結果?

NVIDIA: What kind of results are you seeing?

約書亞: 使用GPU高速平行運算, 即使在簡單的蛋白質表現的重建模擬上, 我們也能獲得驚人的, 數百倍的加速 – 相較於使用單核CPU的序列運算(*Note: CPU是序列運算, 再多核也不能平行處理, 所以才說與單核比較). 在這點上, NVIDIA CUDA GPUs高速平行運算已經是一種賦能技術(*Note: 能因此種技術而達成終端研究的一種中介的關鍵促成技術). 它讓我們可以做到前幾年的科技所無法做到的高速運算量.

Joshua: For simple representations of the protein, we typically get a several hundredfold increase in simulation throughput compared to a CPU implementation running on a single core. In this regard, GPUs running CUDA have been an enabling technology. They have allowed us to perform calculations that would have been completely unfeasible just a couple of years ago.

NVIDIA: 請告訴我們你的研究如何獲得資金的挹注?

NVIDIA: Tell us about how your project is funded.

約書亞: 我們部分資金來自於FundScience, 一個專門挹注資金於獲得同行互評為前瞻研究以期達到從大眾取經效果的非營利性組織. 我的研究在FunScience主持研究計畫中排名前三大重點項目, 而這筆先期資金挹注的確有效的幫助, 並開始與維持我這個研究計畫.

Joshua: We are funded in part through FundScience, a non-profit organization that financially supports peer-reviewed pilot projects using crowd-sourced funding. My work was selected as one of the first three projects hosted on their site and the preliminary support has been a real boon in getting this project up and running.

NVIDIA: 在你的研究計畫上, 誰提供了貢獻與成為你的研究夥伴?

NVIDIA: Who are some of your contributors and partners?

約書亞: CDW與Paragon Micro兩家公司慷慨的捐獻以CUDA為基礎的NVIDIA GPU圖形處理器, 並且本計畫也獲益於許多對我們感興趣的學術個體. 我也獲得匹茲堡大學模擬與建模中心(University of Pittsburgh’s Center for Simulation and Modeling)莫大的協助, 匹茲堡大學模擬與建模中心的策略方向是, 成為一個致力於使用高等通用型圖形處理器運算(Advanced GPGPU)以幫助與訓練各種科學領域獲得GPU高速運算能力的專門中心.

Joshua: CDW and Paragon Micro have generously donated CUDA-based GPUs and the project has been supported by individuals with an interest in the science. I also have worked extensively with the University of Pittsburgh’s Center for Simulation and Modeling, which has made a strategic commitment to advancing GPGPU computing in various scientific disciplines on campus.

NVIDIA: 人們如何更進一步知道你的研究內容, 與如何提供協助?

NVIDIA: How can people learn more about your research and how to donate?

約書亞: 我的FundScience專案計畫網頁就在這裡, 更多關於我的研究成果可以在http://mgrabe1.bio.pitt.edu看得到.

Joshua: My FundScience project page can be found here and more information about our lab’s efforts can be found at http://mgrabe1.bio.pitt.edu.