Welcome to FAH-Addict
Welcome to FAH-Addict, the primary source of information about the Folding@Home project. Our goal is to provide:
- latest project news
- tutorials for installing or optimizing clients and OS's
- hardware reviews
Please enjoy your visit, and feel free to send us your suggestions and/or report errors.
- latest project news
- tutorials for installing or optimizing clients and OS's
- hardware reviews
Please enjoy your visit, and feel free to send us your suggestions and/or report errors.
Recently, many people have noticed that there have been periods where they have been unable to get work, apparently due to server downtime. However, this has not been down to the servers themselves.The problem is that if the server has run out of units the clients display the same error as if the server was down and they were unable to connect to it. This is incorrect, as the servers have been online without a break since February 25th, when a group of GPU servers went down, despite what is reported by the Server Status page (now if it reports a server as down, it could mean that it has merely run out of units). However in the past week there have been shortages of units which have appeared (to users) to be server downtime.
To overcome this situation, the Pande Group have been working to increase the numbers of projects and work units available on the network, for both classic and SMP clients. A large quantity of units are in the pipeline for the Protomol (b4) and Gromacs 4 (a4) uniprocessor clients, the latter being a new core that will roll out properly for the first time shortly. New units for the a3 SMP core are also in development, but some issues were found in the science code that precluded their release. The last few issues are currently being worked out of the core, and these units should make their way into the wild in the near future.
We are currently in a transition phase (the beginnings and ends of a large number of projects happening at the same time) which will occasionally lead to shortages for some client types. The target set by Vijay Pande is that there should never be such an interruption in the availability of units for contributors, and they are working to achieve that goal once again.
We will post updates as these new projects come online.
Adapted from: Vijay Pande's blog
jmn
On: 03/12/10
Vincent Voelz has put into production the latest series of GPU projects: 6600 to 6606. These projects each contain 1047 atoms, are worth 450 points, have a preferred deadline of 6 days, and a mandatory one of 9 days. Project 6606 is a bit different and is worht 550 points, while keeping the other characteristics of the series. The projects are distributed through a new server at 171.64.65.61, which is still using v4 of the server code.
The first benchmarks gives :
p6600-6605
- 4520 PPD on a GT 240 with shaders at 1700 MHz
- 4800 PPD on a 9800 GX2 core
p6606
- 5525 PPD on a GT 240 with shaders at 1700 MHz
- 5800 PPD on a 9800 GX2 core
These simulations involve the native and unfolded states of the protein CspTm (a protein consisting of 66 thermophilic icebreaker residues), whose denatured state has been studied experimentally using a single molecule, and the method of energy transfer by resonance (FRET - "Förster resonance energy transfer"), also known as energy transfer between fluorescent molecules. There is a Wikipedia page available in German, if you wish to learn more about Theodor Förster.
While FRET experiments have shown fast intramolecular rearrangement, other studies (notably Trp-Cys quenching studies) show very slow intrachain diffusivity. The dynamics of the chain in the unfolded state under native folding conditions is crucial to understanding how the chain can rearrange to find the folded state in a tractable manner. Vincent hopes to use this simulation to shed some light on what molecular relaxation experiments are measuring.
Updated 03/11/10 at 8:30am UTC:
550 points for projects 6606 was a mistake... each project in this series is worth 450 points per WU. All upcoming WUs from project 6606 will also be valued at 450 points each.
Some of you may have noticed the lack of points for these projects in the statistics; they are not lost and have only been delayed. This bug is now fixed and all WUs will be properly credited.
KaySL
On: 03/09/10
A persistent rumour has been spreading on the nVidia forums about the 196.75 drivers. Following reports from several users (specifically here, here and here), these drivers have a problem with the automatic fan speed control.The problem manifests itself in the fan failing to speed up when the card is under load (and therefore warmer). Some people have had cards burn themselves out due to overheating.
nVidia has not officially confirmed the bug but the fact that the 196.75 drivers have been withdrawn from the download site seems a tacit admission of a serious problem with these drivers.
If you have installed this set of drivers we urge you to remove them and replace them with the 196.21 WHQL or 196.34 beta drivers, which do not have this issue.
Thanks to chhosni for reporting this information.
Updated on 05/03/2010 at 22:46 UTC :
nVidia acknowledged the issue, and their recommendations are similar to ours:
nVidia have said:
196.75 Alert!
We are aware that some customers have reported fan speed issues with the latest 196.75 WHQL drivers on NVIDIA.com. Until we can verify and root cause this issue, we recommend that customers do not download this driver. Instead, please stay with, or return to 196.21 WHQL drivers. Release 196.75 drivers have been temporarily removed from our website and we also are asking our partners and others to remove temporarily this 196.75 WHQL driver as well.
We are aware that some customers have reported fan speed issues with the latest 196.75 WHQL drivers on NVIDIA.com. Until we can verify and root cause this issue, we recommend that customers do not download this driver. Instead, please stay with, or return to 196.21 WHQL drivers. Release 196.75 drivers have been temporarily removed from our website and we also are asking our partners and others to remove temporarily this 196.75 WHQL driver as well.
jmn
On: 03/05/10
Yellow Dog Linux, which already provides extensive support for PowerPC platforms (G4, G5 and Cell) has seen a new incarnation with a version dedicated to the nVidia CUDA API.Based on Red Hat Enterprise, this version requires an x86-64 machine, and (naturally) a CUDA-compatible nVidia graphics card; GeForce 8xxx or upwards.
The distribution comes complete with CUDA development tools and nVidia drivers pre-installed. It integrates the previous three CUDA SDKs (2.2, 2.3 and 3.0) and the tools needed to develop OpenCL-based code. Finally, for Eclipse lovers, that IDE is also included, featuring a plugin which fully supports CUDA development.
Yellow Dog Enterprise Linux for CUDA is free for educational use, but using it to develop commercial applications does require a license, which costs $400 per year.
This distribution could be an ideal candidate for the quick setup and running of dedicated GPU and Linux folding clients (only the CUDA wrapper needs to be installed manually). The terms of use make it very accessible for student use - you must be a student in order to use the free version, otherwise you must pay for access to the commercial version.
The question, however, is whether this distribution is actually faster than other distributions with the execution of CUDA applications.
Source: Fixstars
KaySL
On: 03/04/10
Imran Haque has announced a new series of projects for nVidia GPU clients using FahCore 14. These projects range from 5907 through to 5909.These projects are similar to p5905, and should behave in the same manner. As with any project using Core 14, the GPU load varies a fair bit, and the time per frame can also vary slightly. Imran has warned that these new projects use a different Run/Clone/Gen management that can cause duplicate WUs to be assigned to several machines. Don't worry if this happens, as the results of your calculations will still be used, and you will still be credited points.
These new projects are distributed by the server at 171.64.122.70. They have 973 atoms, weigh in at 1888 points each, and should preferably be returned within three days, but no later than six.
KaySL
On: 03/03/10
If you have the Release Candidate version of Windows 7 (build 7100) installed on your computer, you've probably recently noticed a message requesting you to update it to the full release version.The RC version's tenure is now coming to an end, and it will soon become unusable for Folding@Home usage, as well as any extended running.
As a reminder, there are three major phases in the RC's life-cycle:
From February 15, 2010:
- Users will receive a daily notification in the taskbar, warning of the impending expiration of the RC
- Users will be similarly warned via a wizard after four hours, then once every hour from thereon
From 1 March 2010:
- Taskbar-based notifications will continue
- The system will restart itself every two hours
From 1 June 2010:
- The desktop will appear black
- Windows 7 will warn that the installation is not authentic, and will present the activation wizard
- The system will continue to restart itself every two hours
If you are still using the RC, it would be wise to either update to the final release version... or migrate to a certain elegant operating system for a more civilised age.
KaySL
On: 03/02/10
In February, Greg Bowman launched several new projects for the nVidia GPU client: projects 10102 through to 10106.The projects are complementary simulations to project 10101, which announced in late December 2009, and have some experimental extensions of that project. The simulations are geared towards studying the lambda repressor of the head piece of the villin protein.
These projects have the same parameters (deadlines, number of atoms, etc.) as that of project 10101. You can read our announcement of project 10101 again to learn more about these projects.
KaySL
On: 03/02/10
Often, when trying to gauge the power consumption of GPUs, overall consumption of the machine ends up being measured in the process.It is important to know exactly how much power our GPUs are consuming, and consumption can range from just a few watts to several hundred, in the case of multi-GPU cards. Since GPU folding places a great demand on graphics cards, it would be both interesting and useful to get an idea of how much power is consumed in the process.
Hardware.fr has just updated its testing methods, and is able to give precise consumption information on about 79 cards, thanks to an ingenious new system that takes power readings directly from both the PCI Express 16x port, and the PCI Express power connectors.
This should help folders better decide which card is right for their needs, power-wise, and will also give a good PPD/W measurement. There is one caveat, however: since GPU folding does not utilise the GPU in the same way that a traditional 3D application does, the actual measurements may differ somewhat from the values given by Hardware.fr.
Read the test on Hardware.fr.
KaySL
On: 03/01/10
So, you have a desire for some... bling? No, you don't have to go round the back of the Flaming Noob Inn and ask a man named Nick "raZor" Rock for "a little something to take the edge off". Instead, all you have to do is visit ASUS, and you'll get something even better!nVidia is still holding out against the competition with its Fermi GPGPU, but ATI is nonetheless set to impress. The Asus Ares is the dual-GPU version of the AMD 5870, but is differentiated by way of 4GB RAM, as opposed to the regular 2GB. It doesn't increase 3D performance by a great deal; instead the main benefit here is the number of displays it can output to, and the resolution it's capable of doing this at, namely 3x 2560x1600. Yes, three displays is a rather unrealistic resolution for most of us, though it may appeal to the sci-fi lovers amongst us.
On the folding front, it unfortunately suffers from the performance issues we've all come to expect from ATI graphics cards. Stanford has halted development on the GPU2 ATI core, which is coded in such a way to take particular advantage of HD4xxx cards, so cards in the HD5xxx series will not see any performance bonuses.
You will be able to jack up your PCI-Express plugs to the Asus Ares this March.
Source: Inpai
KaySL
On: 02/25/10
AMD has released pricing information for the first processors in the Opteron 61xx series, codenamed Magny-Cours.The chips are manufactured on a 45nm process, are compatible with socket G34 (1974 pins), and will come in 8- and 12-core versions. They are primarily geared toward the server and multiprocessor workstation market. Each processor has 4 Hypertransport 3.0 links, quad-channel DDR3 memory support and 12 MB of L3 cache.
The initial prices/specs are:
- Opteron 6128 (8 cores) | 1.5 GHz | 12MB L3 cache | TDP of 115W - €253.49
- Opteron 6134 (8 cores) | 1.7 GHz | 12MB L3 cache | TDP of 115W - €489
- Opteron 6136 (8 cores) | 2.4 GHz | 12MB L3 cache | TDP of 115W - €692
- Opteron 6168 (12 cores) | 1.9 GHz | 12MB L3 cache | TDP of 115W - €692
- Opteron 6172 (12 cores) | 2.1 GHz | 12MB L3 cache | TDP of 115W - €917
- Opteron 6174 (12 cores) | 2.2 GHz | 12MB L3 cache | TDP of 115W - €1078
We don't yet have exact performance figures in relation to Folding@Home, but these processors are nonetheless ideal candidates for BigAdv projects.
Source: TechConnect Magazine
KaySL
On: 02/23/10