Looking for CUDA application engineer in Shenzhen

dougie 2010-01-21 11:03:19
Hi CUDA expert here,

I am sorry to disturb everybody here. I have been searching for CUDA developers in China without success. I am working in a company that has more than 20000 employees globally, and risk analysis is a very important business element. Why this is important to everyone is because 18 months ago, I made 2 bold forecasts to my global managers. First, NVIDIA hardware with CUDA library will beat Larrabee with Intel Math Kernel library (IMKL). Second, many good CUDA developers with English skills will be available in China, and thus my company should form a CUDA development team in China. Not all the managers are convinced but at the end, they agreed to let me to lead this global effort. My first forecast is true now. But my second forecast is not accurate yet.

Without being able to find high caliber CUDA engineer in China, I will fail. And those who were not convinced 1.5 years ago would say China can never produce English speaking high tech software engineers. So I need your help. We are looking for people with skills of :
- Experience in CUDA 2.x or later
- Have in depth multithread programming experience in Linux and Windows
- have strong experience in Visual Studio with C++
- Have experience in usage of NVIDIA integrated tools like NEXUS
- Able to communicate in English.

For those who are interested, pls send your resume to douglaswoocc@yahoo.com.cn. Please send me email also, if you want further details.

regards
Douglas
...全文
428 31 打赏 收藏 转发到动态 举报
写回复
用AI写文章
31 条回复
切换为时间正序
请发表友善的回复…
发表回复
Maschinenwissen 2010-03-05
  • 打赏
  • 举报
回复
为什么要用英语?中文难道不够表达么?
  • 打赏
  • 举报
回复
引用 30 楼 cyrosly 的回复:
那时GPU实际上就是变种的多核CPU了


如果真的如此的话,也就不好玩了.
本来没有cache的话当然不太好,但挖掘了一定的局部性后,再想提高一点都要复出巨大的代价.
与其"浪费"那么多晶体管的资源,还不如用在其它地方.呵呵.

Cyrosly 2010-02-28
  • 打赏
  • 举报
回复
两者本来就不是同样的体系架构,提高方式当然不一样了.你这不也说了要看应用类型,所以CPU未必比不上SIMD架构,或者说本身这个说法就有问题。现在的GPU内部很多标量处理器阵列,与存储控制器,线程调度器等前段构成许多“弱逻辑”处理器,但看GPU发展的历史,内部的流处理单元的结构越来越复杂,将来如果想取代CPU,那么必须需要更复杂的内部逻辑才能运行诸如操作系统之类的软件,那么也会和CPU遇到一样或类似的问题,那时GPU实际上就是变种的多核CPU了
  • 打赏
  • 举报
回复
引用 28 楼 cyrosly 的回复:
很同意这个观点,现在的多核处理器很多价位和高端GPU差不多甚至更便宜,比如四核,每个核都使用SSE* 加速,同时考虑超线程实际至少有4x4x2=32个线程并发,限制更少以及运行在更高的频率上,如果程序充分优化,CUDA的优势绝对没有那么大.测试过四核CPU,没有使用SSE,效率大约提升了3.7倍,如果考虑代码和算法优化有些程序大约可以缩短CUDA 5到一个数量级的差距,有些程序还比CUDA快很多


对计算密集型应用SSE该是不错的了.但如果是带宽限制哪?忘了哪里看的资料,CPU的内存借口带宽最高也就是10G左右吧.还是感觉两者的性能提高方式不同,要看应用的类型了.
Cyrosly 2010-02-28
  • 打赏
  • 举报
回复
引用 27 楼 opengpu2010 的回复:
引用 26 楼 l7331014 的回复:引用 25 楼 gogdizzy 的回复: 我觉得如果Intel的计算性能赶上CUDA,那么价格和功耗就非常大了。 个人认为X86构架就算堆积多个cpu搞成了MIMD,本身也是不可能赶上SIMD构架的. 毕竟cpu靠的是挖掘程序和数据的局部性来提升性能.而gpu靠的是挖掘数据的"有序性"来提升性能. 出发点就根本不一样.

大牛!X86的SSE还是很强大的,如果多SSE协处理器也是可以做到很高性能的:>


---------------------------------------------------------------------------------

OpenGPU论坛http://www.opengpu.org/bbs

OpenGPU Graphics Open Source community(图形开源社区),聚焦领域(focus domain)包括:
* GPU Architecture(图形处理器体系结构)
* Graphics Algorithm(图形算法)
* GPGPU Programming (面向通用的图形处理器编程)
* Open Source Rendering Engine(开源渲染器)
* Open Source GPU Simulator/RTL Implement(开源GPU模拟器 )


很同意这个观点,现在的多核处理器很多价位和高端GPU差不多甚至更便宜,比如四核,每个核都使用SSE* 加速,同时考虑超线程实际至少有4x4x2=32个线程并发,限制更少以及运行在更高的频率上,如果程序充分优化,CUDA的优势绝对没有那么大.测试过四核CPU,没有使用SSE,效率大约提升了3.7倍,如果考虑代码和算法优化有些程序大约可以缩短CUDA 5到一个数量级的差距,有些程序还比CUDA快很多
OpenGPU2010 2010-02-27
  • 打赏
  • 举报
回复
引用 26 楼 l7331014 的回复:
引用 25 楼 gogdizzy 的回复:
我觉得如果Intel的计算性能赶上CUDA,那么价格和功耗就非常大了。

个人认为X86构架就算堆积多个cpu搞成了MIMD,本身也是不可能赶上SIMD构架的.
毕竟cpu靠的是挖掘程序和数据的局部性来提升性能.而gpu靠的是挖掘数据的"有序性"来提升性能.
出发点就根本不一样.


大牛!X86的SSE还是很强大的,如果多SSE协处理器也是可以做到很高性能的:>


---------------------------------------------------------------------------------

OpenGPU论坛http://www.opengpu.org/bbs

OpenGPU Graphics Open Source community(图形开源社区),聚焦领域(focus domain)包括:
* GPU Architecture(图形处理器体系结构)
* Graphics Algorithm(图形算法)
* GPGPU Programming (面向通用的图形处理器编程)
* Open Source Rendering Engine(开源渲染器)
* Open Source GPU Simulator/RTL Implement(开源GPU模拟器 )
  • 打赏
  • 举报
回复
引用 25 楼 gogdizzy 的回复:
我觉得如果Intel的计算性能赶上CUDA,那么价格和功耗就非常大了。


个人认为X86构架就算堆积多个cpu搞成了MIMD,本身也是不可能赶上SIMD构架的.
毕竟cpu靠的是挖掘程序和数据的局部性来提升性能.而gpu靠的是挖掘数据的"有序性"来提升性能.
出发点就根本不一样.
  • 打赏
  • 举报
回复
我觉得如果Intel的计算性能赶上CUDA,那么价格和功耗就非常大了。
dougie 2010-02-24
  • 打赏
  • 举报
回复
2个月前, Intel 的代表到我公司开会, Intel 游说我们不要用cuda, 他们的立场是虽然Intel 在12月11日己经公开宣布Larrabee 计划要抻后, 短期内Intel 无意推出高性能计算的特别新硬件,但他们认为Intel 现有的XEON CPU, 提升到16核心, 32核心, 是可达到, 顾客应继续用XEON CPU, 这样可以不用更改现有程式, 他们说, 用IMKL, 便可以确保多线程的架构优化, 我听完后, 觉得无什么新意, 现有的Intel XEON, 很难有大突破
cuda2010 2010-02-24
  • 打赏
  • 举报
回复
#22有个笔误更正一下,"好像还见过"应为"好像没见过",csdn不能修改自己的帖子真是不方便。
Larrabee矩阵乘法性能测试见:

http://forums.gametrailers.com/thread/larrabee-hits-1-tflops---2-7x-/975813

号称达到了GT200的2.7x。另外这2个回帖内容和主题无关,还请楼主谅解。
cuda2010 2010-02-24
  • 打赏
  • 举报
回复
请教一下,您文中提到的"Larrabee with Intel Math Kernel library (IMKL)"有消息来源吗?好像还见过IMKL已经支持Larrabee的消息。

本来对Larrabee也挺期待的,前段时间还看到有文章说Larrabee矩阵乘法测出极高的flops,可惜Larrabee比fermi更加遥遥无期啊。
dougie 2010-02-24
  • 打赏
  • 举报
回复
http://job.dajie.com/position_90474.html 是這個深圳的cuda 招聘广告 .
有兴趣cuda开发者请联络卓银公司, 深圳卓银公司是在帮我工作的公司做软件开发.


dougie 2010-02-10
  • 打赏
  • 举报
回复
I start to receive resumes, but not entirely matching.
Two resumes match our requirement. The first person has worked in NVIDIA for 2 years, but unfortunately his experience is more on the graphics programming, e.g. OpenGL, DirectX , rather than the core CUDA programming. One person is taking a PhD course about Accelerated computing research in University. His 3 years master degree is on physics modelling that requires computer graphics to illustraet models. 2009, he started to change the image modelling from CPU to GPGPU. Again, unfortunately he spent most time in the graphics part. Both persons English is good, graduated with good academic results.
Why I shared the experience here, is because for those who feel they are 懷才不遇, who feel incapable people are underestimating their capability, why don't send your resume so that you can tell your experience and skills are more matched.

frog_skywalker 2010-02-05
  • 打赏
  • 举报
回复
fact is fact
i have to admit those problems do exist here,in china

u wanna offer a chance,huh? that's definitely good thing
i respect the passion, for our motherland, deeply
i had some misunderstanding that u stick to the english problem.sorry for that,
but u do sounds like so a little

last few words
i'm not qualified for the job for i only have experience in C for CUDA, and i also havn't tried the BETA version of nexus yet.
but i still wanna know what u r offering exactly, and...maybe a little about the passion u have,u know, help those younglings,like me
how about a interview, in cafe? just kidding
  • 打赏
  • 举报
回复
20多年前,确如LZ所说的,大陆的毕业生到一流公司工作的机会几乎没有.记得当时还有研究生应聘酒店大堂领班的现象出现.
20多年过去了,难道还是和当年一样?呵呵.

建议LZ还是再多了解了解吧....
Cyrosly 2010-02-04
  • 打赏
  • 举报
回复
我这么一位GPU大牛放在这里,LZ都无视偶的存在,还招什么贤,纳什么士
frog_skywalker 2010-02-04
  • 打赏
  • 举报
回复
呃~~~写着写着我也跑题了
frog_skywalker 2010-02-04
  • 打赏
  • 举报
回复
怎么越讨论越乱~~~

就事论事
lz的定位既然是想找人才,为啥不提供多一点的信息呢?
费尽周折招不到人,有可能是没有符合条件的人,也可能是招的方式不对。如果提供的信息让符合条件的人对lz的公司产生了错误的第一映像,那结果会如何呢?


u want candidates to write a mail in PROPER english
like a entrance test,uh?
then u will provide further informations

i may or may not be qualified according to the skills u requied,
but i don't like the...(how to say that) the culture behind ur annoucement...of ur company

lots of people here may have the same perspective, and they may choose don't bother to send their resume

i feel very weird that 2 chinese people talking english.
"Able to communicate in English"
yeah,u r damn right, it's all about communication
so what's the problem if i can understand what u r talking about and make myself understand with some grammar and spelling mistakes?

i agree with u about the 3 folds, most of it
now what, u wanna help us make some differences with the advanced knowledge u process?
or u just point out our problem, and come back 10 years later?


"China is still behind India in software development"
yeah,i also agree with that.
so why not go and find a indian guy, after the failed searching in china
wjk98550328 2010-02-03
  • 打赏
  • 举报
回复
GPU出现不到10年,CUDA概念出来应该不超过3,4年,作为一个新的技术,国内应用本来就不多,要找senior 开发人员,难。。。。。。。。。。。。。。。
楼主不愿培养新人,直接想摘果子,呵呵,笑话了
  • 打赏
  • 举报
回复
[Quote=引用 11 楼 gogdizzy 的回复:]
" Project managers themselves have no management experience that don't know or have no interest to coach junior developers "

大部分公司确实有这个问题。
[/Quote]

这只是一个引导的问题.如果,一个组织从下到上都是在"头"的帮助指点下成长起来的,就不存在这样的问题了.关键还是在"头".呵呵.
加载更多回复(11)

357

社区成员

发帖
与我相关
我的任务
社区描述
CUDA高性能计算讨论
社区管理员
  • CUDA高性能计算讨论社区
加入社区
  • 近7日
  • 近30日
  • 至今
社区公告
暂无公告

试试用AI创作助手写篇文章吧