GPGPU issues for nVidia 296.10 WHQL and 295.73 WHQL drivers

By , the in Drivers - No comment
Many folding users have reported that they encountered issue with GPGPU after installing nVidia 260.10 WHQL and 295.73 WHQL drivers on their machines.

The issue symptoms are GPGPU client crashes after a few minutes/hours when the machine is folding alone whereas it works fine most of the time. The core starts to crash in a loop on an UNSTABLE_MACHINE error, which makes the client to go into sleep mode because of the high number of detected failures.

After investigations, we have discovered that the problem is related to monitor sleep/standby mode. When it turns off, the WU that was running is completed fine, but all the following ones throw the error mentioned above. It looks like with these two versions of the drivers, the GPU is also deactivated when the monitor is in sleep mode. This removes the GPU from the available devices for CUDA or OpenCL applications.

The workaround is quite simple and can be done using one of the following suggestions:

  • keep your older drivers (285.62 for exemple) if you don't need the new features
  • disable monitor standby in power control panel, and turn your monitor of manually using the power switch.

The issue has been reported to the nVidia's forum, so we hope that the manufacturer will take it into account for the next driver releases.

Source: nVidia's Forums

Updated on 04/10/12 à 15H10 UTC:
The latest beta drivers released today (301.24) seem to solve the GPGPU issue encountered with previous WHQL drivers.

Source: Official Forums

New methods for computation drug design

By , the in Folding@Home Project - No comment
A key aspect of the research done by the Folding@Home project has been the use of computational methods to design new drugs, mostly for Alzheimer’s disease. At University of Virginia, Michael Shirts' laboratory is developing these methods to use FAH power to fight diseases.

Generally, the small molecule works as drugs by biding to specific locations of important proteins. For example, an antibiotic works by binding to a protein of bacteria which interfere enough with the pathogen's internal workings to disable it or to kill it. By targeting only protein sites that are unique to the pathogen, drugs can have a very precise effect without risking to hurt the human molecules or desired bacteria that live in our body (like intestinal flora which contributes to digestion). The same principles can turn on or off specific parts of our own protein machinery, allowing the design of drugs that fight diseases related to the breakdown, mutation or malfunction of our own cellular machinery, like Alzheimer disease, heart diseases, diabetes and many other conditions.

It is however very hard to calculate exactly how tightly a given small protein will bind to a target protein, or even exactly where and by what mechanism it will bind. A number of computational methods are used in industry today to estimate the binding affinity of small molecules in the process of drug design, but they mostly rely on approximations that are computationally cheap and very approximate, rather than more expensive methods that have the potential to be much more accurate. With Folding@Home, researchers now have the capability to perform rigorous evaluations of these more complete methods, understand their limits, and make them more efficient and reliable.

Michael Shirts' team has been developing its own method that works mostly on well known systems such as FKBP, a protein on the immune system signaling pathway. Once the methods are well-understood, we will be moving on try to design small molecules to treat AIDS (the HIV reverse transcriptase enzyme, required for DNA to replicate) and influenza (various proteins involved in virus cell entry). Such molecules will still require significant effort to make into drugs, since drugs also have to dissolve easily, penetrate cells, and not be broken down to quickly, but being able to predict more easily which molecules interact tightly with the intended targets will be a huge step in the right direction.

Michael Shirts' team is also contributing to improve Folding@Home infrastructure by working to port new versions of the Gromacs molecular simulation platform to Folding@home and improving the interface and integration between Gromacs and Folding@home.

Source: Vijay's blog

Protein folding and viral infection

By , the in Folding@Home Project - No comment
Understanding protein folding has many application areas in biology and biomedicine. For example, consider one of the major research areas of Doctor Peter Kasson's laboratory at University of Virginia: the study of how the influenza virus infects cells. In the past, Doctor Kasson and Doctor Pande have studied two aspects of this process: how the influenza virus recognizes the receptors on the cell surface to infect the « right » cell type and how small vesicles fuse.

Doctor Kasson's group is now studying the function of the viral protein that controls cell entry, a protein called hemagglutinin. The hemagglutinin protein interacts with cell membranes: one piece inserts in the cell membrane, refolds, and alters the membrane in some unknown manner to promote the virus entry in the cell. Another piece links the virus and the cell membranes and refolds to keep the two together. Some simulations are running on Folding@Home to study each of these pieces of the hemagglutinin. Doctor Kasson's laboratory also looks at these processes experimentally.

Hemagglutinin protein

Both of these problems involve protein folding. This extends the problem of understanding protein folding beyond the « canonical » model of an unstructured protein in water taking on a final shape but instead in the first case it is about a small protein inserting into a lipid membrane and changing shape in response to its environment, and in the second case, it is a large protein changing shape in response to physiological cues.

Future news will address methods that Doctor Kasson's team has developed to assist in these studies as well as other important problems they work on. The team is also contributing to improve the efficiency of running Folding@Home simulations and analyzing the results. The Folding@Home community has made an importent contribution in providing the computing power for these studies (as you can see on the project results page) and the researchers are grateful to all donors involved.

Source: Vijay's blog

Protein folding and molecular recognition

By , the in Folding@Home Project - No comment
Here is message from Professor Xuhui Huang’s laboratory at Honk Kong University of Science and Technology, another research group that collaborates with Folding@Home project. Professor Huang and his laboratory have created many methodological applications to FAH (refer to this publication) as well as important research on the molecular nature of Huntington’s Disease. Here is an update from Professor Huang:

In the past couple of years, Folding@Home has greatly contributed to the team’s research on understanding the molecular recognition processes. Molecular recognition, such as enzymes that need to recognize their substrate and drugs that have to be designed to bind to specific receptors, is crucial to biology and medicine. Experimentally probing the chemical details of molecular recognition events is challenging, while computer simulations have the potential to provide a detailed picture of such events. With FAH donors’ help, the team is running large scale simulations on a group of Periplasmic Binding Proteins aiming to reveal the general relationships between protein structures, its intrinsic dynamics, and mechanism of recognition process.

Folding@Home projects involved in this experiment are in the 7700-7712 range. Professor Huang’s team greatly appreciates the help of all the FAH donors, the beta testers and the rest of the FAH team to make their research on molecular recognition possible.

Source: Vijay’s blog

LTMD: A new key technology to speed up Folding@Home simulations

By , the in Folding@Home Project - No comment
This news is coming from one of the key researches of the Folding@Home project which is studying proteins folding in collaboration with Prof. Jesus Izaguirre lab at the University of Notre Dame.

The Izagurre's lab at University of Notre Dame is working with Vijay Pande's one at Stanford to produce a new GPU core that leverages the amazing speed of OpenMM implicit-solvent force calculations (the heart of Folding@Home GPU core) and adds a new method called LTMD (Long Timestep Molecular Dynamics). This combination allows an about 10 times simulation speedup over what OpenMM is capable of for systems from the WW domain (35 residues, 544 atoms) to the Lambda repressor (80 residues, 2000 atoms). This speedup allows about 10 milliseconds of simulation per day of computation, which get Folding@Home platform closer to the ability to simulate a single trajectory in the millisecond scale.

In collaboration with Cauldron Development (lead by Joseph Coffland, main developer of Folding@Home v7 client and some cores), researchers are hoping to build a GPU core that might be the first CPU/GPU hybrid core. There are still some technical questions that need to be sorted out about how to best achieve this and a talk with beta testers should be engaged when the core will be closer to entering in production.

To go further, researchers will continue to improve LTMD GPU technology to achieve greatest performance improvements on larger systems that will have a better biomedical interest. A very interesting improvement would be to extend this technology to explicit solvent simulations.

As far as scientific simulations, we simulate folding of about 80 mutants of the Pin1 WW domain, a protein involved in some cancers and the Alzheimer's disease. Understanding the role of these mutations in protein misfolding might have great biomedical consequences since many diseases have at least some components linked to protein misfolding. Another project that is about to start will simulate the dimerization during folding of proinsulin and its mutants, which results in some Type IA diabetes.

Izaguirre lab thanks the Folding@Home donors, the beta testers and Vijay's lab for their generosity and their leadership that made such advanced simulations achievable.

WW domain

Adapted from: Vijay's blog

Progressive roll out of "BigAdv-16"

By , the in Folding@Home Project - No comment
As already announced, some changes are ongoing in the "BigAdv" project: the minimum number of required cores is increasing and the deadlines will be shortened accordingly to reflect the need of faster return of these new projects. Vijay Pande posted some explanations about these changes:
The "BigAdv" project has been developed on purpose to address the fastest machines available, which is of course a moving target. This project uses the highest-end machines involved in Folding@Home on projects that are unusually big (related to the memory needs or the upload/download requirements) and which require a large amount of computing power. Luckily over time, the CPU computing power is increasing and the high-end systems follow the same dynamics. Scientific researches enabled by the folding donors are exciting and have an impact at both computing power and fast return needs. As a result, having 50% of computing power running "BigAdv" projects wouldn't help so as if older or bandwidth limited machines kept running these projects where fast return of result is very sensitive.

As announced previously, Pande Group objective is to tighten the "BigAdv" projects deadlines. As a result, only machines with a minimum of 16 cores will be able to receive these projects. New projects are being designed for this new phase of the project that we will call "BigAdv-16". This design took a bit more time than expected, but researchers have completed internal tests and they are now starting the beta test phase. A new server has been set up to deliver these projects. Since it is an early release, points and deadline are monitored closely. You may expect some adjustments during the test phase.

Once the new "BigAdv-16" projects will be stable, older "BigAdv-12" projects will be converted as "BigAdv-16". The schedule for these changes remains unknown. On the other side, the old "BigAdv-8" projects will be definitely abandoned (even if they are almost gone already).

To conclude, the Pande Group is well aware that the number of cores is not a perfect measure for system performances. In the future, the team will attempt to design a better measurement. We will keep you informed about the changes when they will be announced.

Happy folding and thanks for your continuous support to the project.

Adapted from: Vijay's Blog

Linux 3.2 Cross Memory Attach : our SMP1 performances would have been better!

By , the in OS - No comment
Linux 3.2 Cross Memory Attach : our SMP1 performances would have been better!
This news is coming a bit too late but reminds us a little bit of nostalgia … or anger … :D

The cross memory attach patch developed by Christopher Yeoh has been incomporated in 3.2 Linux Kernel.
The objective of this work is to improve the performances of the processus that rely on MPI (Message Passing Interface) layer. This norm is often used for high performance computing because it allows many processors in clusters to exchange messages to process data.

Yes you read it accurately, if we folders now have wonderful multithread implementations of Gromacs core, MPI no longer being used for Folding@Home since the switch to SMP2, the rest of the world is still using it ! To summerize the optimisation by Christophe Yeoh, MPI used to generate two copy operations for each message exchanged. The new implementation only requires one, which greatly reduces the amount of data exchanged between processes.

Christopher Yeoh modified the OpenMPI library in order to use the new cross memory attach feature and he ran a few benchmarks on a 64 cores machine using POWER6 CPUs. As expected when removing a costly copy operation, the tests show a big performance improvement:

Number of processus 4 8 16 32
MB/s without pach 1235 935 622 419
MB/s with CMA feature 4741 3769 1977 703

Those gains would have helped a lot our Core 2 Quad CPUs in the past ! Congratulations to Christopher Yeoh and to the other contributors to the Linux kernels to keep going further !

Even of we’re no longer using this technology in Folding@Home cores, some BOINC projects are working on SMP implementations of their scientific application using OpenMPI. We hope this technology will help to speed up their computations.

Source : Linux FR (in French)

Happy new year! 2011 results and 2012 perspectives.

By , the in FAH-Addict - No comment
Happy new year! 2011 results and 2012 perspectives.
Greetings noble visitor! The FAH-Addict team wish you a happy new year 2012, maybe the last one according to some predictions, but … let’s enjoy it :D

2011 has not really been a good year for FAH-Addict, the administrators being overworked for good, or very good reasons. Congratulations to Frodo The Hobbit who got married in the end of 2011 ! In 2012, we’ll try to do better, we owe it to you because of your continuous support to the site !

Feedbacks on what we announced

2011 Folding switch to v7 : TRUE !
Finally, it’s here, and it works pretty well. Although it is not mandatory yet, the v7 client are doing well though open beta and it’s way easier to use for the new donors !

2011 BigAdv returns on Linux : TRUE !
Thanks to the A5 core released this year, Linux users are able to fold BigAdv WUs again.

2011 ATI cards are able to fold more efficiently : TRUE !
With GPU3 (which require v7 client), ATI cards are back to the business. All issues are not sorted out yet and performance improvements are expected, but we hope the future ATI drivers and core releases will help.

This year has also seen the following changes : the extension of the quick return bonus to uniprocessor projects, the A4 core gaining the possibility run both in uniprocessor and SMP modes, some updates to the stat system and the opening of the closed beta.

Consulting the FAH-Addict’s crystal ball

For 2012, nothing is clear yet. Last week, BigAdv projects have been repositioned to aim 16 core (or better) machines, but excepted this obviousness there’s not a lot to speculate on. The project has gained in maturity, so it’s up to us to keep folding ;)

So for 2012, are we going to see GPUs go back in the competition (with ATI’s Southern Island and nVidia’s Kepler) or a Folding core for ARM CPUs (which are growing in number and processing power) ?

Feel free to share your opinions on the Folding@Home project future …

16 cores minimum requirement for BigAdv on January 2012.

By , the in Folding@Home Project - 1 Comment
This is like a storm in the most devoted folders world : starting January 16th, 2012 the BigAdv projects will require at least 16 cores (16 physical cores, or 8 physical core supporting HyperThreading technology).

Stanford is aware that this decision will result in many disappointment and frustration for those who won't be able to get BigAdv anymore, since the BigAdv projects are earning a substantial bonus compared to regular SMP projects. But this decision was made to focus BigAdv program back to its initial goals: to provide a group of machine with higher computing power than average ones to work on a limited number of big projects that require fast return rates.

Since the launch of BigAdv program, 8 cores have become the norm thanks to Intel’s i7 CPUs providing 4 physical cores with HyperThreading technology. The massive enthusiasm for these projects has created shortages, forcing researchers to open up more projects in parallel, which is not always possible.

So, on January 16th, the « -bigadv » flag on machine with 8 or 12 cores will no longer provide BigAdv projects. It will automatically fall back to regular SMP projects.

Source : Vijay’s blog

Dust, our old enemy finally defeated by MSI?

By , the in Hardware - No comment
Dust, our beloved machines tend to accumulate it forever … at a variable pace depending on their situation in the house and the time they spent powered on …
Faced with this plague, we all fight it with our own proven weapons: can of compressed air, tooth brush, paintbrush, everyone has his own style.

Yet MSI seems to have discovered an innovation that could revolutionize our struggle, this innovation is ...

... the fan that removes dust from itself!

The idea is simple: during the first 30s of boot, your graphic card is cold and we are sure that it won't perform any intensive computation (you're still booting your OS).
The spin of the fan is then reversed compared to nominal functioning and the speed it runs is pushed at its maximum value. The intended effect is to get out some of the dust that began to take up residence on the blades and the bearings of the fan.After 30s of reverse spin, the fan spin is reversed again to enter its nominal functioning behavior.

This will never replace a good clean up, but that’s probably something that will help the graphic cards to handle the folding torture longer.

The first MSI card to use this new feature is based on a GTX 580, the GeForce N580GTX Lightning Xtreme Edition.

Source : Tom's Hardware France