c*********i
发帖数: 46 | 1 ph.D.老板是搞multicore architecture，比较偏硬件，而自己想搞一些偏软的东西，
说实话，硬件的东西也不是很懂，就是做了一些embedded programming的时候，接触了
一些，但只是皮毛，自己将来也不想做硬件，对软件更感兴趣一些。现在换老板，不是
很好，况且现在这个老板人很nice,对自己也很器重的样子。
想问一下大牛们，multicore 领域，偏软的有哪些研究热点呀，除了搞os scheduling
外？ 有什么现在好找工作，偏application的吗？
多谢了！ |
|
l******9
发帖数: 579 | 2 如何确保多线程程序在 multicore server 上确实使用了 所有的 core
I have a multicore server, which has 24 CPU, each CPU has 6 cores.
It is Intel xzeon X5650 2.67GHz
cpu cores : 6
cpu MHz : 1596.000
totally, I have 24 * 6 = 144 cores.
I designed a multithreaded C++ program with boost/thread.
How to make sure that my program is run by all 144 cores ?
Any help is really appreciated.
thanks |
|
l******9
发帖数: 579 | 3 如何确保多线程程序在 multicore server 上确实使用了 所有的 core
to do computing in parallel physically (not logically)
I have a multicore server, which has 24 CPU, each CPU has 6 cores.
It is Intel xzeon X5650 2.67GHz
cpu cores : 6 (support 6 threads)
cpu MHz : 1596.000
totally, I have 24 * 6 = 144 cores.
I designed a multithreaded C++ program with boost/thread.
How to make sure that my program is run by all 144 cores ?
Any help is really appreciated.
thanks |
|
l******9
发帖数: 579 | 4 如何确保多线程程序在 multicore server 上确实使用了 所有的 core
to do computing in parallel physically (not logically)
I have a multicore server, which has 24 CPU, each CPU has 6 cores.
It is Intel xzeon X5650 2.67GHz
cpu cores : 6 (support 6 threads)
cpu MHz : 1596.000
totally, I have 24 * 6 = 144 cores.
I designed a multithreaded C++ program with boost/thread.
How to make sure that my program is run by all 144 cores ?
Any help is really appreciated.
thanks |
|
l******9
发帖数: 579 | 5 如何确保多线程程序在 multicore server 上确实使用了 所有的 core
to do computing in parallel physically (not logically)
I have a multicore server, which has 24 CPU, each CPU has 6 cores.
It is Intel xzeon X5650 2.67GHz
cpu cores : 6 (support 6 threads)
cpu MHz : 1596.000
totally, I have 24 * 6 = 144 cores.
I designed a multithreaded C++ program with boost/thread.
How to make sure that my program is run by all 144 cores ?
Any help is really appreciated.
thanks |
|
l******9
发帖数: 579 | 6 如何确保多线程程序在 multicore server 上确实使用了 所有的 core
to do computing in parallel physically (not logically)
I have a multicore server, which has 24 CPU, each CPU has 6 cores.
It is Intel xzeon X5650 2.67GHz
cpu cores : 6 (support 6 threads)
cpu MHz : 1596.000
totally, I have 24 * 6 = 144 cores.
I designed a multithreaded C++ program with boost/thread.
How to make sure that my program is run by all 144 cores ?
Any help is really appreciated.
thanks |
|
l******9
发帖数: 579 | 7 如何确保多线程程序在 multicore server 上确实使用了 所有的 core
to do computing in parallel physically (not logically)
I have a multicore server, which has 24 CPU, each CPU has 6 cores.
It is Intel xzeon X5650 2.67GHz
cpu cores : 6 (support 6 threads)
cpu MHz : 1596.000
totally, I have 24 * 6 = 144 cores.
I designed a multithreaded C++ program with boost/thread.
How to make sure that my program is run by all 144 cores ?
Any help is really appreciated.
thanks |
|
l******9
发帖数: 579 | 8 thanks for you reply.
I can generate these threads.
But, how to make sure that these threads are really scheduled to
these cores by OS ?
I try yo use "top" command on linux but it is not helpful.
On the multicore server, there are 24 CPUs, each of them has 6 cores.
Sometimes, only one CPU is used to run a multithread program
no matter how many threads are generated.
I run some test program, which create 4 (boost) threads but only one
core is used.
I cannot see all 4 threads are run in 4 cores in... 阅读全帖 |
|
f*********y
发帖数: 43 | 9 前段时间看了一篇文章讲multicore的killer app少。
我怎么觉得network的主要工作像rounter什么的，
都是可以简单并行的呢？
有谁看过这方面的文章给科普一下。 |
|
f******y
发帖数: 2971 | 10 Is this a multicore server? I think it should be called a cluster. |
|
s*****t
发帖数: 737 | 11 当前最热的是multicore systems.
形式语义,自动机,计算模型都是非常基础的知识,
对应programming language来说还远远不够,
PL和computer architecture是紧密相关的.
正因为multicore的出现,concurrent software 的编写将是一个非常大的挑战.
因为以往软件很多都是随着单cpu的主频的提高而运行速度得到提高.
但是在multicore systems上,这个free lunch已经不存在了.
当前有很多基于multicore system的PL research. |
|
S***d
发帖数: 1802 | 12 From mozilla forum:
If I understand it correctly multithreading can be achieved on single core
or multicore setups, but Firefox multithreading(present since Firefox 3.0)
isn't optimized for multicore setups. Multiprocessing which a lot people
confuse with multithreading, can benefit from both multicore setups and
multithreading. I think with the fact that a lot of Firefox users still use
single core setups, Mozilla next step is to implement multiprocessing(
Electrolysis Project), which is curren... 阅读全帖 |
|
发帖数: 1 | 13 Open Source RISC-V Postdoc Position
University of Washington, CS & EE Department
Location: Seattle, CA
Postdoctoral positions are available immediately to join the productive
Bespoke Silicon Group (BSG) at the University of Washington http://darksilicon.net ).
BSG Leader Professor Michael Taylor received his PhD at MIT and was lead
architect of the 16-core MIT Raw tiled multicore processor, one of the
earliest multicore processors, which predates Intel’s multicore chips. He
co-authored the earl... 阅读全帖 |
|
a***c
发帖数: 315 | 14 multicore 在PL上没啥新颖的. 都是研究了几十年的东西. 只不过又给了一群搞
parallel programming/compiler 的人骗房顶的机会. multicore真正要解决的问题基
本都是IC,体系结构level的. On-chip Network, 3D IC/3D Packaging, architecture
啥的. |
|
d**y
发帖数: 32 | 15 两篇文章
1：Optimizing ontology and semantic search usinggenetic and greedy
algorithms approach
abstract：
The content is extracted by means of semantic relevancy.
The semantic relevancies relate the content of videos based on a
certain parameter. The parameter varies between system to system
(implementation). The parameter will improve the performance of
semantic relevancy and accuracy. This accuracy is obtained after
various random experiments. Here a method called concept, sub
concept graph method ... 阅读全帖 |
|
l******d
发帖数: 530 | 16 要做这么个multi-core的程序，在每个core上有个线程，这些线程每隔一个固定的相同
的周期，都要协同执行一些操作。
如果不通过各个core之间握手来保证“相同的周期”，是不是得要求multicore有个
global clock，每个core都以这个global clock计时。请问现在multicore都有global
clock吗？能不能举几个现实的例子。谢谢。 |
|
l*****0
发帖数: 299 | 17 不知道这样说对不对： 既然matlab和OpenCV都自动利用multicore以提高运算速度，
做
vision的人大可不必学习了解multicore方面的技术了！ |
|
k**********g
发帖数: 989 | 18
In the past, multicore programming focuses on fully utilizing CPU. This can
be seen in the High-Performance Computing (HPC) benchmark rankings.
This approach neglects an issue that is ubiquitous: Most "processes" (
including computations, communications, and I/O) do not generate (yield) a
block of "result" of size N at once. Instead, almost all real-world
processes yield results incrementally.
To visualize, one can imagine this at a microscopic timing level:
After spending 500 microseconds of i... 阅读全帖 |
|
k**********g
发帖数: 989 | 19
Start with Java Executors.
Learn how to use it just like the Runnable and Callable class.
http://docs.oracle.com/javase/7/docs/api/java/util/concurrent/E
For example, if your goal is to write Android programs, (1) Executors is
basically all you need - the only other things you need are (2) thread
safety on Android, and (3) understand when and why you need to call Activity
.runOnUiThread() in some situations
C++ didn't have anything in the standard library comparable to Java
Executors. Design-by... 阅读全帖 |
|
j****x
发帖数: 943 | 20 芯片还是有很大进步的 (supposedly follow Moore's law)，从hyper threading到
multicore。量产工艺已经到32nm (22nm is the next step?). 越小的话power
consumption越小(就不会那么烫了), higher potential frequencies, transistors
密度越高.最关键的是便宜了,想想十年前一个pentium多少钱, 而且现在你pc里的cpu和
十年前586的cpu性能相比是什么概念。
现在自己弄个multicore的机器，就可以玩mpi了，以前得要多少钱。更关键的是
scientific programming的paradigm的shift，现在不写个parallel的code都不好意思
给人看。 |
|
发帖数: 1 | 21 Multiple PhD positions with full financial support are now available (
starting from Fall 2018) with Dr. Cong Liu in the department of Computer
Science at UT-Dallas.
Dr. Liu’s research focuses on the areas of Real-Time Systems, GPGPU, and
Data-driven Cyber-Physical Systems. According to www.csrankings.org, UT-
Dallas is ranked No. 8 nationally in the general area of embedded & real-
time systems due to Dr. Liu’s contribution. His supervised PhD students are
publishing at top conferences such as ... 阅读全帖 |
|
发帖数: 1 | 22 Multiple PhD positions with full financial support are now available (
starting from Fall 2018) with Dr. Cong Liu in the department of Computer
Science at UT-Dallas.
Dr. Liu’s research focuses on the areas of Real-Time Systems, GPGPU, and
Data-driven Cyber-Physical Systems. His research is funded by both NSF (
including the prestigious NSF CAREER award) and industry. For more
information, please visit: http://www.utdallas.edu/~cong
There are currently three general projects to be worked on:
1. P... 阅读全帖 |
|
s*****r
发帖数: 43070 | 23 为啥没人想到multicore-CPU，俺觉得那才是超级计算机的发展方向，跟攒一大堆机器
的技术含量相比，完全不是一个量级的。 |
|
t**********8
发帖数: 1683 | 24 中国研制出世界最快计算机 运算能力比美快一倍 2013-06-04 00:45:23
2013年06月04日09:59 来源：TechWeb.com.cn
【TechWeb报道】6月4日消息，据国外媒体报道，科技网站Computerworld从一位上周在
华访问的美国学者处获悉，中国研制出了一台具有每秒54.9千万亿次浮点运算(
Petaflops)能力的超级计算机，比美国目前任何一台超级计算机的速度都快一倍。
据悉，中国这台最新的超级计算机名为“天河-2”，配置了3.2万个多核英特尔至
强Ivy Bridge处理器芯片，4.8万个英特尔至强Phi芯片。Phi芯片是基于英特尔MIC（多
集成内核）架构的协处理器。中国政府研发投入为大约2.9亿美元。
根据去年11月发布的超级计算机500强排行榜，美国最快的超级计算机名为“Cray
”，运行速度接近每秒18千万亿次浮点计算。这个排行榜将在今年6月中旬更新。
link address: http://www.computerworld.com/s/article/9239710/China_surpassing_U.S.... 阅读全帖 |
|
s*******y
发帖数: 4173 | 25 现在cpu都是multicore
to
reach
while |
|
s********i
发帖数: 17328 | 26 There are several "open-source hardware" CPUs, typically implemented as a
soft microprocessor.
Amber is an ARM-compatible 32-bit RISC processor. Amber implements the ARMv2
instruction set.
LEON is an open source 32-bit SPARC-like CPU created by the ESA. It's the
standard CPU for the European Space Industry.
Milkymist SoC, featuring the LatticeMico32 microprocessor core, has a
complete set of independently developed open source system-on-chip cores
such as a SDRAM controller, bus components, grap... 阅读全帖 |
|
F*****O
发帖数: 338 | 27 你是专家啊。
网上看了一下， 多数人都指出那个1/10跑分是因为jit没有优化。
编译路漫漫
程序语言 | 编译技术 | 开源软件
Home IDE Research Security SpiderMonkey Tricks Unix/Linux V8 Web 编译 闲谈
Home > 闲谈 > 龙芯CPU的JavaScript性能
龙芯CPU的JavaScript性能
June 14th, 2015Leave a commentGo to comments
这是4月份的新闻的评论，一直没有时间来写，拖了两个月了。
四月份的时候，类似“龙芯性能不到iPhone6 A8性能的1/10？”的新闻[1][2][3]又一
次把龙芯拉出来被人吐槽。其中出现 1/10 对比的一项是 SunSpider 测试，SPEC
CINT2000 的测试差异没有那么大。另外没有贴出来 SPEC CFP2000 的分数对比。
作为一度天天跑SPEC和SunSpider等Benchmark的人来说，新闻中的测试报告是没有可信
度的。性能测评需要有一套完整的流程，才能够确认自己获得的测评分数是正确的。不
同的... 阅读全帖 |
|
b**t
发帖数: 102 | 28 (有意请先PM, thanks!)
Principal Logic Design Engineer - SATA Expert
The Processor Division of Broadcom Corporation is well known in the
semiconductor industry for providing super high-performance multi-core
network microprocessors. We are staffing up in critical R&D areas in the
high-rise facility at Highway 101 & Great America Parkway
As a senior Logic Design Engineer you will be responsible for performing
micro-architecture and logic design in the development of I/O devices in
high performance mu... 阅读全帖 |
|
b**t
发帖数: 102 | 29 (有意请PM!)
Job Description
The Processor Division of Broadcom Corporation is well known in the
semiconductor industry for providing super high-performance multi-core
network microprocessors. We are staffing up in critical R&D areas in the
high-rise facility at Highway 101 & Great America Parkway in Santa Clara, CA
.
As a senior system Logic Design Engineer you will be responsible for
performing micro-architecture and logic design in the development of system
logic in high performance multi-threade... 阅读全帖 |
|
h******u
发帖数: 155 | 30 不完全这样...multicore architecture 和相应起来的parallelism，concurrency也是
一个大坑，吸引无数人往里面跳 |
|
r**u
发帖数: 1567 | 31 上周面的，intern，挂掉。做OS kernel, 在multicore NUMA系统上performance
optimization什么的。
面试官是小印，group manager(三哥还是挺强的）, 上来先问有没有做过kernel的东西
，答曰：NO。估计就定下基调了，面了30分钟吧（suppose 40),问题不难，都是OS的基
本知识。
1. 我的research，real-time scheduling .......
2. 指针初始话，意思就是必须malloc让pointer point to valid address才能使用它
。malloc到哪. heap.
3. 如果physical mem 2GB, 俩program都要2GB，可不可以同时运行。NO problem,
virtual address space.
4. process migrates from on core to another, what will affect performance?
5. what is the problem of implementing a schedule |
|
x***y
发帖数: 633 | 32 来自主题: JobHunting版 - 请教个题目 CPU utilization? Multicore? |
|
a****n
发帖数: 1887 | 33 GPGPU(cuda), FPGA, MultiCore, 这三个东西以后在金融领域应该都不错
如果做EE, 我觉得FPGA发展前景很不错. |
|
o******e
发帖数: 81 | 34 btw这个是俺的马甲
今天第一轮电面，我是experienced的不是fresh
感觉不怎么好，就问了一道题，太多detail了
电话来迟了8分钟，加拿大人。先问了问最challenge/favorite的project，说了几分钟
后他说我知道了，考你个题吧
很简单，Unicode的charactor array/string，找出现次数最多的char
我给了3个解，基本没有用任何考虑时间
1. double loop, O(n^2)
2. hash，他问我worse case，我一不留神说了O(n)，忘了hash的陷阱。他说恩perfect
hashing是不存在的，blablabla，我说我完全同意。我说如果space足够的话效率比较
好的char hashing是不太难的，他说anyway那也是O(n)，好吧。。
3. int[65536]的array，因为是unicode，我跟他确认了一下是2 bytes。我一开始说复
杂度的时候原来的solution是iterate string一遍，然后iterator count的array一遍
，突然想到可以用一个temp varia... 阅读全帖 |
|
B*******1
发帖数: 2454 | 35 Agree. I think for multicore cpu with hardware thread, multithread will run
faster. For single core cpu, single thread is faster. |
|
c******g
发帖数: 69 | 36 单线程单进程程序没有context switch和multicore的问题
我猜想是cache performance的问题，假设L1 cache四路组相联（这个是一下推论的假
设，但应该属于cpu常规配置，可以搜搜主流cpu的配置）
#define N 64*1024
// 每个数组64KB
a[i] = b[i] + c[i] + d[i] + e[i];
// 每次循环读四个mem location,写一个mem location
那么在L1 cache来看, 由于每个数组64KB, a[0]-e[0]映射到同一组4路cache line，5
个memory address在4个cache line上冲突，每次循环都会因此出现cache miss
当数组size变成70时，a[0]-e[0]映射到不同的组上，5个mem address在L1 cache共存
，performance就上去了
运行都多次performance不一样，还没有想到解释。。。 |
|
k********h
发帖数: 22 | 37 About Us:
ROBLOX is a leading user-generated gaming site that makes players the
architects of their own 3D worlds, drawing over one billion page views and
21 million in-game hours each month. The site is ranked #1 in the US for
total engagement time within the 8-14 year old segment, and is rapidly
growing in popularity amongst players aged 15+. Our vision is to become the
leader in user-generated gaming for all ages, around the world. On ROBLOX,
players use core building components to create ela... 阅读全帖 |
|
c**z
发帖数: 669 | 38 is this company's name multicore |
|
b**t
发帖数: 102 | 39 (有意请先PM, thanks!)
Principal Logic Design Engineer - SATA Expert
The Processor Division of Broadcom Corporation is well known in the
semiconductor industry for providing super high-performance multi-core
network microprocessors. We are staffing up in critical R&D areas in the
high-rise facility at Highway 101 & Great America Parkway
As a senior Logic Design Engineer you will be responsible for performing
micro-architecture and logic design in the development of I/O devices in
high performance mu... 阅读全帖 |
|
b**t
发帖数: 102 | 40 (有意请PM!)
Job Description
The Processor Division of Broadcom Corporation is well known in the
semiconductor industry for providing super high-performance multi-core
network microprocessors. We are staffing up in critical R&D areas in the
high-rise facility at Highway 101 & Great America Parkway in Santa Clara, CA
.
As a senior system Logic Design Engineer you will be responsible for
performing micro-architecture and logic design in the development of system
logic in high performance multi-threade... 阅读全帖 |
|
o****d
发帖数: 2835 | 41 难得看到有人讨论这个
我也学的architecture 如果要专业对口，的确职位较少
不过architecture compiler OS 都属于计算机的system 这几个方面联系比较紧密
公司里也有类似system engineer的职位 有做simulation，performance analysis之类的
如果你以后想呆industry
建议你也多了解些OS compiler，不要只局限于architecture
另外 根据你做的architecture的东西 提炼出你的技能
比如我做的有关multicore，涉及到dynamic analysis，scalability，runtime等
找工作时 往需要这些东西的地方靠就是了。。。
如果是去做码工 那到时好好做题吧
跟平时学啥就关系不大了
的？ |
|
o****d
发帖数: 2835 | 42 你说的很有道理
architecture的有些研究 纯粹是为了发paper
其实架构的设计 需要的就是 “简单并有效”
设计的太复杂 功耗也大 还容易出错
比如cache的研究，人家许多CPU还就用简单的pseudo LRU
但每年那么多cache的paper 都是加这加那 然后show performance，其实也并不一定有
很大提高 关键是hardware overhead也都上去了
现在multicore，manycore出来之后
pipeline也都偏向从out of order 回到 in order
不过大家phd不容易啊 总得想出一些idea来发paper。。。
的. |
|
b**h
发帖数: 109 | 43 非常感谢，
确实一直没怎么看到讨论architecture的，所以提出来。我现在做的project向下延伸
较多，甚至做一些circuit的小模块，然后写verilog，做synthesis，再往上是c，c++
用软件simularion，不知道哪些技能是重要的？
另外你说的scalability，dynamic analysis属于什么技能？你说的是比如从single
core到multicore这种scalability？
类的 |
|
|
l*n
发帖数: 529 | 45 hadoop就是个map-reduce的实现，讨论ML/DM跟hadoop的关系，实际上就是MR跟ML/DM的
关系。google下"Map-Reduce for Machine Learning on Multicore"这个paper，感觉
是最能体现二者关系的结论。
ML算是DM的超集吧，DM倾向于结论需要human readable，ML的结果只要机器懂不用管人
看不看得明白。 |
|
|
x*****n
发帖数: 65 | 47 https://sjobs.brassring.com/TGWEbHost/jobdetails.aspx?partnerid=20085&siteid
=5109&Areq=126488BR&codes=IJMON
Title Software Engineer
Location(s) US - Massachusetts - Hopkinton
Business EMC
Functional Area(s) Engineering - Software
Requisition ID 126488BR
Job Description
GENERAL SUMMARY
The EMC Global Platform Engineering (GPE) organization is looking for a
highly motivated technical individual to contribute to the development, test
, and integration of SCSI Enclosure Services (... 阅读全帖 |
|
f*****q
发帖数: 35 | 48 PhD new grad, 明年毕业，主要研究multicore OS，有kernel的设计和开发经验。
面试完后已经和Google的allocation team进行了一段时间的选组，到目前为止已经给
了七八个组选择，跟其中两个谈完感觉和自己的兴趣还有背景不是完全match。其中有
几个是android的组，只是做的方向并非我专长。
十分犹豫是否去其中一个相对贴边的组，或者继续让HR找其它组。大家给点建议？有了
解Google里面跟OS / kernel相关的组的话，麻烦还请给个提示。多谢！ |
|
j*******n
发帖数: 10 | 49 上周四hiring committee meeting 完，HR 回信说 committee 建议找到组。
本人方向 system，multicore，large－scale，infrastructure
有相关组的联系，非常感谢 |
|
k********h
发帖数: 22 | 50 About Us:
ROBLOX is a leading user-generated gaming site that makes players the
architects of their own 3D worlds, drawing over one billion page views and
21 million in-game hours each month. The site is ranked #1 in the US for
total engagement time within the 8-14 year old segment, and is rapidly
growing in popularity amongst players aged 15+. Our vision is to become the
leader in user-generated gaming for all ages, around the world. On ROBLOX,
players use core building components to create ela... 阅读全帖 |
|
