|
楼主 |
发表于 2013-3-6 12:24:27
|
显示全部楼层
MARCH 05, 2013
Introducing Folding@Home Core 17 – GPU zeta core
As also announced on OpenMM/Folding@home programmer Yutong "proteneer" Zhao's web site, we are happy to announce that Folding@Home Core 17 has entered Beta. Externally, you probably won’t notice too much of a difference. Internally, this is a complete overhaul that brings many new features, and sets a strong foundation for the future of GPU core development. In addition, the restructuring brings much tighter integration of the core with the rest of the development within Folding@Home.
We’re also introducing an explicit solvent project (p7661) as part of the Beta. To keep the credit assigned to these projects consistent with previous explicit solvent work units run on CPUs, we are also awarding a quick return bonus with a k-factor of 0.75. This reflects the additional scientific value of these units, and keeps the Folding@Home credit awards consistent across different architectures.
Usage:
This is a still a very new core, a lot of the features have yet to be fully tested. Thus, as is the beta policy, no official support is given. You must enable the -beta flag on FAHClient, ie. set client-type=beta. If you’re using client 7.2.x or earlier, there are two options:
1. Specify -gpu-vendor=XXX, where XXX is either nvidia or ati
2. If -gpu-vendor is not specified, the core will automatically guess the platformId.
Otherwise, 7.3.6 lets you specify the particular -gpu-vendor as an option
Supported NVIDIA cards: Fermi or better (Titan does not work atm, as NVIDIA needs to publish new OpenCL drivers)
Supported ATI cards: HD5000 or better
As always, please use the latest drivers (Win XP is NOT supported due to super-old AMD drivers).
Key Features:
Cleaner Code
We have deprecated several layers of GROMACs and other wrappers as the old architecture severely limited the types of simulations that can be run. Much of the work on the new core has been to replace existing features. The resulting code is now more streamlined and integrated. We also anticipate that this major re-design will allow us to introduce new features into the Folding@Home much faster.
Serialization
We have introduced a new serialization mechanism that allows Pande Group researchers to setup significantly more diverse simulations. Pande Group researchers can now easily setup jobs and projects using Python (with a much richer and easier to use set of libraries), while the core maintains its speed by virtue of being written in C++. We achieved this using a serialization technique, whereby all details of a simulation are encapsulated using a standardized format that is then be passed around between language barriers. This also drastically reduces the dependencies needed by the Work Server and other parts of Folding@Home.
A single unified core now runs both NVIDIA and AMD cards
Before we had two development branches for NVIDIA and AMD cards. It was a difficult and cumbersome task to debug and maintain. We couldn’t easily mix runs and gens produced by different GPU types. Now, using OpenCL, a single core supports not only AMD and NVIDIA, but theoretically any OpenCL-capable device.
Improved Stability
By reducing the amount of boilerplate code, we’ve also increased the robustness and stability of the core. The log files should also now be much more informative. There are also a lot of useful debugging features built right into the core to help PG developers nail down hard to find bugs.
Extra special thanks to our testers:
Jesse_V, k1wi, artoar_11, bollix, ChelseaOilman, bruce, Demonfang, Grandpa_01, EXT64, Flaschie, HayesK, jimerickson, mmonnin_, P5-133XL, Patriot, rhavern, sc0tty8, SodaAnt, SolidSteel144, EvilPenguin, art_l_j_PlanetAMD64, thews, cam51037, Pin, muziqaz, baz657, PantherX, QINSP, Schro, and hootis.
大意:
FAH开始测试core17
core17几乎重写了代码,为将来的各项改进打下基础,测试项目为p7661,此项目有奖励分,奖励系数为0.75。
参与方法:由于此内核还在测试,所以可能不稳定。
首先,你要给客户端加beta参数(client-type=beta),如果客户端显卡识别有误,你也可以用-gpu-vendor=XXX参数强行指定显卡类型。
其次,你需要支持的显卡。N卡:fermi及以上(Titan由于opencl驱动问题,暂不支持)。A卡:HD5000及以上(由于A卡的xp驱动太老,可能无法支持)
再次,最好升级到最新的驱动,以便对opencl提供最佳支持。
亮点:
1、代码重构。有些老代码很低效,也不支持新功能。所以完全重写了。
2、新的序列化技术。PG小组现在可以使用python客户端直接设置发布新任务,做更多的模拟,而计算部分也用c++重写了,尽量保持计算高效的同时,降低了科学家的任务发布复杂度。
3、N、A内核合体。以前维护2套内核的成本很高,现在由于采用了opencl技术,2个客户端合体了。而且理论上所有支持opencl的设备都可以用。
4、提高稳定性。代码重构的同时,加入了更多有用的log输出和debug功能,方便PG小组找bug。
5、致谢内测人员。
OpenMM Youtube channel
OpenMM is a key part of Folding@home, powering its GPU cores. You can learn more about OpenMM at its youtube page, which includes technical videos on how you can incorporate OpenMM into your code. It also includes an introduction to Markov State Models (MSMs), which is a key technology used in Folding@home.
http://www.youtube.com/user/SimbiosOpenMM
大意:
FAH的gpu主要使用的是OpenMM库,官方做了几个介绍视频,有兴趣的可以翻墙去看下。
译注:既然显卡奖励出来了,估计地主们的显卡升级潮也会到来,需要淘汰卡(尤其是免费用电)的朋友请常来看看,密切关注论坛的捐赠贴子。 |
评分
-
查看全部评分
|