[独立平台] [生命科学类] Folding@Home

vmzy

April 13, 2008
How the GPU2 core works
There have been some misunderstandings on how the GPU2 core works.  In particular, for small proteins like villin on GPU's with large number of stream processors (SP's) like the 3850 or 3870, the protein is too small to use a larger number of SP's unless the CPU is very fast.  Some people have guessed that there is some internal SP limit.  This is incorrect; the problem is that small proteins can't be parallelized amongst a large number of SP's.  

We are working to release larger proteins (about 2x the number of atoms) as they are more interesting scientifically and use the GPU's (even the high end ones) much closer to 100%.  The exciting part for us is that the larger proteins run at almost the same speed as the slower ones on GPU's (whereas on CPU's, they're 4x slower); this is where the GPU2 code should shine.  In parallel, Mike Houston at AMD is working to optimize CAL such that it has lower CPU overhead.  

For now, we're pushing out villin WU's as a test (good to know that the code is working well), but we expect the larger WU's to be going out soon (say a week or two, pending internal testing).
大意：
当前测试的WU是villin，他的蛋白质分子很小(单位计算量很小)，而且并行度较低，所以出现了cpu占用多，而GPU占用小的现象，这是正常的。我们正在准备放出大蛋白质的WU，不过何时发布取决于当前测试任务的进度。

新的WU肯定能喂饱GPU，与此同时AMD的Mike Houston正在优化CAL，以进一步降低cpu占用。

译者注：如果Nvidia公司能像AMD那样于斯坦福合作，我想GPU客户端早就该支持Nvidia了吧！

April 15, 2008
Updates to the Download page/GPU2 goes live
We've made several changes to the download page.  First and foremost, we've tried to simplify it a bit for donors new to FAH by moving the high performance Windows downloads to their own page.  The goal is to try to simplify the long list of possible clients new donors see.  For those who are already experts at FAH, they will notice the high performance client link on the download page (although it may be nice to highlight that a bit more).

We have also updated the GPU client on the download page, with GPU client downloads pointing to GPU2 rather than GPU1.  This is a major step forward for FAH, as we move to our second generation GPU client ("GPU2").  It's also an important time to note the contributions of the GPU1 client: we've been able to learn a great deal of running GPU's "in the wild," which is very different than in the lab, and what we've learned has gone into GPU2.  GPU2 is much easier to run, more reliable scientifically, and has more advanced algorithms.  Due to these differences, GPU1 has become obsolete.  However, we are working to write a paper with what we've learned from GPU1 and acquiring data for that paper will need some more GPU1 cycles.  

Thus, we will continue the GPU1 projects for a little while longer (likely a month or so), while we encourage people to transition to GPU2.  We would like to take this opportunity to thank everyone who has run the GPU1 client and given us such useful data to advance to the second generation client.  So far, the GPU2 client has become a success, with many more people running it than GPU1, in just two days.  We are also working to improve the performance of GPU2 and one should expect to see core updates on a regular basis as we improve the software.  Finally, we are also considering points modifications to make sure that the compensation is reasonable.
大意：
整理了下载页面。

由于GPU1客户端支持1xxx显卡是DX9的，不支持很多功能，所以我们抛弃了GPU1客户端。不过我们打算就GPU1的数据，写一个总结报告，所以暂时还不会停止它（继续约1个月左右）。

April 17, 2008
Visualization on GPU2
Several people have asked about visualization with GPU2.  That is in the works and should be pretty close (not months, but hopefully weeks, maybe faster).  The visualization will be real time like the PS3 (although perhaps not quite as slick, as the Sony team put a lot of effort into lots of nice small touches in the PS3 client visualization).
大意：
几周内，会完成客户端的图形界面，但是效果肯定没PS3那么好。

[ 本帖最后由 vmzy 于 2008-4-19 20:33 编辑 ]

sakeki

没PS的效果也无所谓，反正有就行

sakeki

当前测试的WU是villin，他的蛋白质分子很小(单位计算量很小)，而且并行度较低，所以出现了cpu占用多，而GPU占用小的现象，这是正常的。我们正在准备放出大蛋白质的WU，不过何时发布取决于当前测试任务的进度。

新的WU肯定能喂饱GPU，与此同时AMD的Mike Houston正在优化CAL，以进一步降低cpu占用。

译者注：如果Nvidia公司能像AMD那样于斯坦福合作，我想GPU客户端早就该支持Nvidia了吧！


＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝＝

令人振奋的消息，如果GPU2发展得顺利，我考虑买ATI来玩游戏，不用NV了

JUST

原来是这样，太好了！

shouldbe

原帖由 dlxwj 于 2008-4-19 17:13 发表
1950gt算分 也挺猛的

从PPD来讲1950GT不如超频的E4400。所以有钱还是升级CPU好。当然玩游戏的话，再配个NV的显卡更好。玩游戏用显卡，不玩游戏CPU算FAH。两全其美

vmzy

April 22, 2008
New rules for advanced methods (-adv flag)
In the near future, we will be releasing some new projects which require a very rapid turn-around time.  These are peptide fragment simulations which we are interested in simulating for a time-sensitive collaborative project involving protein structure prediction.

These WUs will go directly to the classic clients running with -advmethods.  Non-classic clients (eg SMP, GPU, PS3) will not be affected, as all of these calculations will be run via the AMBER core and only the classic client supports the AMBER core.

To reward users for participating in this exciting project, we will be giving a x1.5 bonus in the points awarded.  What's the catch?  These projects will be less rigorously beta tested, so there will be an increased risk of Early_Unit_End errors.  We believe the risk of this is minor (there will likely be a higher rate of early unit ends, especially very early in the WU, but we do not expect client machines to become significantly less stable). However, if you do not wish to participate in this project, just remove the -advmethods setting from your client.  

This will not last forever and will likely go back to normal -advmethods usage in a few months.  With that said, we do plan other uses for -adv in the future.

Note: the text was updated to clarify that this is classic clients only.
大意：
接下来的几个月我们会发布一个测试子项目--模拟缩氨酸片段，它们的过期期限很短，而且由于没有经过详细的测试，很可能会出问题，不过为了补偿用户，我会给这些任务以1.5倍的积分。

如果你想参加这些子项目，请打开-advmethods参数。只有标准客户端能参加（SMP, GPU, PS3都不行）。以后还会有类似的任务。

April 23, 2008
Folding@home and Simbios

We've teamed up with Simbios (an NIH Roadmap center at Stanford) to start making key technologies in Folding@home available for others to use.  Our protein folding work has been named one of Simbios' key Driving Biology Problems (DBPs) and there's now a protein folding DBP page.  Our hope is that our developments which have been so useful in Folding@home can help accelerate the work of others as well.  Right now, there are two key areas we've been working to distribute, with more coming along the way.

First, we're working to make our GPU code available to others.  This code will be distributed in a couple of forms.  First, we'll give out a GPU-enabled version of Gromacs (basically, a standalone version of the GPU2 core), which will enable others to get major speed increases from GPUs.  Next, we are working to release a GPU-enabled library (OpenMM), which will allow others to integrate GPU code into their programs.  OpenMM is special in that it is a place for integrating both application developers as well as GPU vendors; much like OpenGL, our hope is that hardware acceleration vendors will now have a single API to accelerate and people who want to write applications will have a single, hardware accelerated API to use that would work on a variety of platforms.

Second, we're also starting to make large data sets from Folding@home available to others.  You can see some of the first data sets on this project page, and we expect to put more data up as time goes on.  Folding@home donors have generated wonderful data sets that aren't possible to generate by other means, and our hope is to publish them so that other scientists can data mine them for other purposes as well.

Finally, we expect to release more technologies from FAH.  In particular, we will be releasing some of the key server-side algorithms which allow FAH to use lots of processors to tackle single complex problems.  This will allow people to run massively parallel code on large clusters to do tasks similar to FAH (albeit on a smaller scale).
大意：
FAH与Simbios合作，扩大FAH的应用范围。

首先，会扩大GPU代码应用范围。会把GPU代码提取出来做一个单独的API库OpenMM。

其次，公开结果数据，供科研用。

最后，会开放某些算法，尤其是服务器端的并行算法。

shouldbe

看了后一部分的英文原文。感觉这回答了“这些项目都很久了啊，有什么成果了吗？（http://www.equn.com/forum/thread-18017-1-1.html）”的问题。虽然我不明白这种计算成果的开放程度到底有多大，但我感觉FAH项目组这次是真的开放了一些数据。而且我能感觉到他们是因为出于严谨，现在才放出这些数据的。这是典型的搞科研的人的态度。

至于分享源代码和服务器端程序，这是个很好的步骤。至少从科学计算的角度来看，可以节省其他科研机构在编程方面的重复工作。而“OpenMM”，这是个很有意思的概念。让人看到了ATI的影子。虽然仍旧能够看到商业利益的影子，但想想如果真的象OPENGL那样形成开放标准，这也是一件好事。这比ATI自己原来的CLOSE TO METAL（从名字上看就很原始）或者NVIDIA的CUDA都是一个进步。避免了一个封闭标准让掌握标准的厂家控制整个行业的问题。让未来拥有更先进技术的小厂家也有机会进来竞争。

vmzy

官方发布GPU2客户端6.11 beta 5版。
主要改进，加入图形显示功能，但是对计算性能影响极大，请慎用。按F1~F6键可以看不同的显示模式。
下载页面：
http://folding.stanford.edu/English/DownloadWinOther
下载地址：
http://www.stanford.edu/~beberg/Folding@home-R600-systray.zip

vmzy

May 23, 2008
GPU news (about GPU1, GPU2, & NVIDIA support)

There are several new developments regarding the GPU cores and clients.

We've been working with NVIDIA to develop a GPU2 core for NVIDIA hardware.  So far, the  code is progressing well and the new GPU2/NVIDIA core is now in closed beta testing.  It's hard to tell if there will be any show stoppers (there's lots of things that could go wrong in distributed computing on GPU's), but so far so good.  We're very excited about the performance (more details on that later).  We hope to have a public beta in the next few weeks.


We are nearing the end for the GPU1 project.  Our plan is to deactivate the GPU1 client on June 2.  We would like to thank everyone who has contributed to that project.  FAH GPU1 was a landmark in computing, being the first distributed computing project on GPU's as well as the first major molecular dynamics calculation to be performed on GPU's.  We have learned a lot from GPU1, and those lessons have been used to architect GPU2, which will be faster, more reliable, and much more scientifically useful.

We are beta testing a new visualization for GPU2/ATI.  This new code provides real time visualization for the GPU2 core, similar to what we have for the PS3.  More information can be found on our FAQ.  We plan to release a similar visualizer for the GPU2/NVIDIA  core as well when it's ready.
So, we've definitely had our hands full on the GPU front.  We look forward to taking these next steps forward!
大意：
GPU的消息

NVIDIA版GPU2客户端开发还算顺利，现已进入封闭内测阶段，如果一切顺利，几周内将开始公测。它性能非常强劲。

GPU1项目将于6月2日正式结束。感谢大家的贡献。通过GPU1我们学到了很多，也为GPU2的顺利开发打下了坚实的基础。

我们正在测试GPU2/ATI新的图形界面，它将和PS3客户端一样，是根据计算内容实时更新的。GPU2/NVIDIA的图形界面也在开发中。

译者注：
这里可以看下介绍，FAH老大Vijay Pande在做演讲。
http://www.pcper.com/comments.php?nid=5679

[ 本帖最后由 vmzy 于 2008-5-24 12:16 编辑 ]

zglloo

wow  NVIDIA就要上线啊！

2287732

等待A和N的新卡发布，然后对比下决定升级A还是N，CPU肯定是上4核肉了。

Bismarck

强烈建议采用原装主板跑24×7

shouldbe

当初买8800GTS还是对的。至少魔兽从来没有让我失望。这回不魔兽的时候就FAH。希望PPD能够重新回到4000以上。不过电费会很恐怖。我住到新家刚半年电表度数已经是住进来快2年的邻居的2倍了。夏天一来，再开空调就更恐怖了。白天不在家总不能开着空调吧。可显卡能不能挺住还是个问题。

[ 本帖最后由 shouldbe 于 2008-5-27 21:05 编辑 ]

2287732

NV在芯片工艺上还是太保守点。

Bismarck

NV芯片工艺不是NV说的算的，而是台积电的问题，台湾的工艺要比Intel和IBM至少落后半代