Aller au contenu
PcPerf.fr

PcPerf bot

PcPerfonaute
  • Compteur de contenus

    399
  • Inscription

  • Dernière visite

Tout ce qui a été posté par PcPerf bot

  1. <p>For those of you interested in the older 6.41 client, we've updated the download links for the 6.41 GPU client. Please find it in the normal place:</p> <p>http://folding.stanford.edu/English/DownloadWinOther</p> <p>This is likely only of interest to a very small community as we highly recommend most people to use the latest v7 client.</p> <p> </p> <table bgcolor="#FFEEFF" border="0" cellpadding="0" cellspacing="0"> <tbody> <tr> <td valign="top"><strong>Windows XP/2003/Vista/7 GPU3</strong> (required for Fermi) System tray client with special viewer for NVIDIA GPU's (installer msi). <br /><br /></td> <td valign="top">6.41</td> </tr> </tbody> </table> <table bgcolor="#FFEEFF" border="0" cellpadding="0" cellspacing="0"> <tbody> <tr> <td valign="top"><strong>Windows XP/2003/Vista/7 GPU3</strong> (required for Fermi) <em>no-nonsense</em> console client. <br /><br /></td> <td valign="top">6.41</td> </tr> </tbody> </table> <p> </p> Voir l'article complet
  2. We have had an unfilled spot in our GPU programming team for a few months and I'm happy to announce that we recently made a great new hire: Yutong Zhao. Yutong completed his undergraduate degree in Math, Chemistry, and Biochemistry from the University of Toronto, and a Masters degree in Computational Chemistry from HKUST, focusing on GPU-powered clustering algorithms. Previously, he has also worked on projects such as FoldIt! Currently, he is the lead developer of the Folding@Home GPU core and aims to extend functionality while increasing maintainability. He also plays a key role on the OpenMM side, he works on devising parallel algorithms to speed up MD simulations. He also maintains a blog detailing some of his algorithms and problems on www.proteneer.com . Voir l'article complet
  3. PcPerf bot

    GPU update

    We've put a new revised AS up and we think this should fix the recent issues. We will of course continue to monitor the situation. Voir l'article complet
  4. We've been working to fix two issues with GPU clients in FAH. Here's an update: 1) There was an issue with NVIDIA clients with G80 GPUs getting stuck in a core_outdated download infinite loop. This has now been resolved. This was caused when we stopped older projects which used core_11 (which the G80's were primarily running) and now the G80's are directed to more recent projects. However, with the ending of the old projects and the start of the new ones, that means G80 GPUs go to core_15, which has much more strict testing for GPU memory errors. Several donors have reported that these tests are failing with their boards. We are looking into whether this is an issue with our test or potentially that the boards are not capable of running the latest core. So, while the core_outdated loop issue appears now to be fixed, there is another unresolved issue which we are continuing to look into. 2) There was a second issue with ATI clients getting directed to servers running NV WU's. This error is also unresolved at the moment, but we have a handle on what's happening on our end and have a team working on it. I don't have any news here to report other than we can see the issue cause in our logs, which is the first step to fixing it. Voir l'article complet
  5. We've been seeing donor reports of WU assignment issues for GPUs (WU's intended only for G80 GPUs going to Fermi and Kepler). We finished an AS patch yesterday, gave it a day of testing, then rolled it out this morning to address this issue. We generally don't like to roll out new AS code this quickly, but this time it seemed important to address this issue quickly. We hope that as of 7:30am pacific time, this issue is now addressed, but we'll be keeping an eye out during the day to see. Voir l'article complet
  6. The unified GPU/SMP benchmarking scheme will be rolled out today on FAH. All the future GPU projects would now be benchmarked using the new scheme. Most of the existing GPU projects have been re-benchmarked to reflect the changes in the benchmarking scheme. We are still in the process of re-benchmarking some old projects assigned to ATI and G80 GPUs. The uniform benchmarking scheme significantly boosts the base points for all GPU projects. However, Quick Return Bonus for the GPU clients has not been introduced at this stage, but will be introduced once we work out an issue on our side. We would like to thank all the beta testers who participated in testing the unified benchmarking scheme. Voir l'article complet
  7. Guest post from Dr. Gregory Bowman, UC Berkeley Ken Dill, a long-standing leader in protein folding, justpublished a beautiful review article about progress in the field. The main point is that what began as aspecific research question has now evolved into an entire field with numerousresearch directions. It’s would bedifficult to provide a more eloquent explanation than Ken’s, so here are a fewexcerpts from the article “The protein-folding problem was firstposed about one half-century ago. The term refers to three broad questions: (i)What is the physical code by which an amino acid sequence dictates a protein’s native structure? (ii) How can proteins fold so fast? (iii) Can we devise acomputer algorithm to predict protein structures from their sequences?... computer simulations of the physical forcesin chemically detailed models have now achieved the accurate folding of smallproteins… What began as three questionsof basic science one half-century ago has now grown into the full-fledged research field of protein physical science.” Insights and achievements fromFolding@home are highlighted on a number of occasions throughout the article. So, congrats to all for helping to establishthis ever-growing field. Also, there's a nice video about this online for those who are interested Voir l'article complet
  8. We have been updating our internal accounting of stats for FLOPs for FAH. The major revamp has come from a better handling of the difference between a CPU and a core. Modern CPUs have multiple cores. Before the v7 client, it was common for donors to fold with multiple uni-processor clients, one for each core. Now, with v7 making SMP more easy to run, SMP is much more common. This means we need to keep track of CPUs and cores more explicitly. Our new serverstats page takes care of this and also updates for estimates for FLOPs across the board, but especially from GPUs. Our goal in general is to be conservative with this reporting, but the old page was likely too conservative. The end result of the update is more accurate representation of the power of FAH today. The new page is located at: http://fah-web.stanford.edu/cgi-bin/main.py?qtype=osstats2 Voir l'article complet
  9. One of our servers (VSP12) which runs multiple virtual servers (VSP12a through g) will be take down for maintenance in the afternoon (Nov 20, around 1:00 pm) and is planned to be back up later in the evening. All the interfaces on this machine (Vsp12a,b,c,d,e,f,g with server addresses 171.67.108.58, 59, 60, 141, 142, 143, 144) hosting projects with ID in the range 8001-8067 will be affected. This includes several GPU3 and SMP A4 (multi+uniprocessor) projects. This server hosts a large number of current FAH projects but donors would be able to get work units from other servers with similar work units. Voir l'article complet
  10. One of our key server rooms will undergo network maintenance on Saturday, November 10th, from 5AM to 8AM pacific time. During the maintenance window, we expect that the servers in that room will be unreachable, hopefully for only 30 minutes each, but potentially for the full time range. We note that this is only one of our server rooms, so the FAH backend should still be primarily operational, but some donors will see some issues with returning work during this time. We also stress that the FAH server backend is architected such that even when servers are down, the points for donors will not be lost, and just the accounting for those points will be delayed until the servers are back up. Voir l'article complet
  11. We have several on-going software development efforts and I'd like to give donors an update. v7 client. Joe Coffland and his team have been working hard on new client releases. 7.2.9 has just been released and a new version will be undergoing beta testing soon. Moreover, we are continuing work on improving the v7 client for windows and squashing the remaining bugs. Moreover, there's additional effort in OSX due to the hiring of a programmer (Kevin Bernhagen) just for the OSX client, as well as additional work for smoother OSX and linux installs. Gromacs core. The Gromacs core team (Prof. Michael Shirts and Prof. Peter Kasson and their labs, at the University of Virginia) are working on the new cores based on the new version of gromacs (4.6). New OpenMM core. The OpenMM team at Stanford (Dr. Peter Eastman and Yutong Zhao) are working on speed improvements for OpenMM (the basis of the FAH GPU core) in general, but in particular optimizations for Kepler and AMD (in coordination with engineers at NVIDIA and AMD, respectively). Yutong has a new FAH GPU core working in the lab and we are doing internal testing on it. Since openMM is full open source, you can see more details, including a commit and change log, at the openMM web site (https://simtk.org/home/openmm). New FAH viral ad campaign. We're also working on a new landing page for FAH and a new video to advertise FAH. This new web/video campaign is coordinated with new client/installer changes to make FAH easier to install and run, especially for those new to Folding@home. We understand that donors don't get to see all of what's going behind the scenes, so we'll try to post these sorts of updates more frequently. Voir l'article complet
  12. For several years, we have worked closely with Sony to bring Folding@home to the PS3. We're excited about what we've been able to do. Since the PS3 started folding in 2007, we've done some really amazing things, with several announcements this year acknowledging advancements in Alzheimer's Disease, Cancer (and this link), Influenza, Type II Diabetes, and other new drug targets. We've come a long way in the last 5 years and we have a lot going on to continue our tradition of pushing the envelope into new technologies. Voir l'article complet
  13. The current benchmarking calculations for SMP and GPU projects are performed on different machines since originally the SMP cores could not perform the calculations that the GPUs cores could and vice versa (GPUs were only for implicit solvent calculations and SMP only for explicit solvent calculations). With recent advances in both cores and completion of our testing of these capabilities to ensure agreement, we are now confident we can do the same work on both cores. Thus, we feel that it is time to unify GPU and SMP benchmarking, both for simplicity and fairness. In order to complete the move towards this plan of "equal points for equal work," new GPU projects will be benchmarked using the existing SMP benchmarking scheme. Based on our internal tests, the end effect of this new, unified benchmarking scheme would boost the points for the GPU projects, both in terms of base points but also by bringing Quick Return Bonuses to GPU clients. In order to test the new scheme, we have started a GPU3 project (Project ID: 8057) and released it for beta testing. Once the benchmarking scheme has been tested, all the current GPU projects will be re-benchmarked to reflect the changes in the benchmarking scheme. Voir l'article complet
  14. Guest post from Profs Kasson and Shirts, UVA and Mr. Coffland A new version of Gromacs (4.6) is coming, and we’re working to bring it to Folding@home. The new code contains a number of improvements (more than you’d expect for a minor version number!), and we’ll post about some of the individual features as we go. Not all of them will be available on F@h immediately, as some will require substantial development work over the next few months. But some of the basics are new free energy methods from our very own Prof. Michael Shirts, new and slightly faster inner-loop code, and some important tweaks to parallelization. Free energy calculations allow us to calculate things like how tightly drugs bind to proteins and the strength of attraction between protein components when pulled apart. And you, of course, know what faster inner-loop code and better parallelization mean! Gromacs is an interesting piece of simulation software in that it’s heavily optimized both for single-computer performance (part of why we chose it for F@h in the first place) and for parallel scaling. A lot of codes choose to emphasize one or the other. But Gromacs tries to do both. That will have some interesting and useful implications for F@h particularly as we look at more and more cores on donor CPU’s (and things like GPU integration). That’s all for now; we’ll keep you posted on progress. Thanks! The 4.6 Core Team (Profs. Kasson, Shirts, and the indefatigable Mr. Coffland) Voir l'article complet
  15. This video is a year old (and we've previously posted the audio), but looking at this again, a lot is still relevant (and we didn't post the video url), so here it is in case people are curious: Host: Marc Pelletier Dr. Vijay Pande, Stanford's Director of Folding@home, details how the World's most powerful system models Alzheimer's and other human diseases. Guest: Dr. Vijay Pande We invite you to read, add to, and amend our show notes. Comments and suggestions on Futures in Biotech. Also thanks to Phil Pelletier and Will Hall for the great themes. Thanks to Cachefly for providing the bandwidth for this netcast. Running time: Voir l'article complet
  16. Guest post from Dr. Gregory Bowman, UC Berkeley Two general objectives of the Folding@home project are (1)to explain the molecular origins of existing experimental data and (2) toprovide new insights that will inspire the next generation of cutting edgeexperiments. We have made tremendousprogress in both areas, but particularly in the first area. Obtaining new insight is even more of an artand, therefore, less automatable. To help facilitate new insights, I recently developed aBayesian algorithm for coarse-graining our models. To explain, when we are studying someprocess—like the folding of a particular protein—we typically start by drawingon the computing resources you share with us to run extensive simulations ofthe process. Next, we build a Markovmodel from this data. As I’ve explainedpreviously, these models are something like maps of the conformational space aprotein explores. Specifically, theyenumerate conformations the protein can adopt, how likely the protein is toform each of these structures, and how long it takes to morph from onestructure to another. Typically, ourinitial models have tens of thousands of parameters and are capable ofcapturing fine details of the process at hand. Such models are superb for making a connection with experiments becausewe can capture all the little details that contribute to particularexperimental observations. However, theyare extremely hard to understand. Therefore, it is to our advantage to coarse-grain them. That is, we attempt to build a model withvery few parameters that is as close as possible to the original, complicatedmodel. If done properly, the new modelcan capture the essence of the phenomenon in a way that is easier for us towrap our minds around. Based on theunderstanding this new model provides, we can start to generate new hypothesesand then test them with our more complicated models and, ultimately, viaexperiment. Statistical uncertainty is a major hurdle in performing thissort of coarse-graining. For example, ifwe observe 100 transitions between a pair of conformations and each of thesetransitions is slow, then we can be pretty sure this is really a slowtransition. However, if we only observeanother transition once and it happens to occur slowly, who knows? It could be that it is really a slowtransition. On the other hand, it couldbe we just got unlucky. Existing methods for coarse-graining our Markov modelsassume we have enough data to accurately describe each transition. Therefore, they often pick up these poorlycharacterized transitions as being important (for protein folding, we typicallycare most about the slow steps, so slow and important are synonymous). The new method I’ve developed (describedhere) explicitly takes into account how many times a transition wasobserved. Therefore, it canappropriately place emphasis on the transitions we observed enough times totrust while disregarding the transitions we don’t trust. To accomplish this, I draw on Bayesianstatistics. I can’t do this subjectjustice here, but if you’re ever trying to make sense of data that you havevarying degrees of faith in, I highly recommend you look into Bayesian statistics. Voir l'article complet
  17. Guest post from Dr. Gregory Bowman, UC Berkeley We’ve been making a lot of progress with developing Markov state model (MSM) methods for analyzing the data we generate with the help of the FAH community. For those of you with a theory background, MSMs are just discrete-time master equation models. For everyone else, MSMs are a way of describing the conformational space a protein (or other biomolecule for that matter) explores as a set of states (i.e. distinct structures) and the transition rates between them. Much of the theory underlying these methods is quite old but their use has been limited by the challenges inherent to identifying a reasonable set of states. During my time in the Pande lab, I worked with Xuhui Huang (now at the Hong Kong University of Science and Technology) to develop new methods for building MSMs from the large data sets we generate with FAH. Together, we started an open source software package called MSMBuilder (here) to automate the process of building MSMs. Now a number of more recent additions to the Pande lab are helping Xuhui, Vijay, and me in continuing to develop the software. As we just released an update to MSMBuilder, I was looking back at some of our user statistics and was pleased to see how quickly our project is gaining traction. Since its initial release in 2009, there have been over 1,600 unique downloads of MSMBuilder. One cute feature of our webpage—provided by the SimTk software consortium at Stanford—is that you can go look where all of our users are (here). Its fun to see that MSMBuilder is being used on 5 continents. Maybe most importantly, MSMBuilder has been used in at least 40 publications to date. MSMBuilder is coming up at conferences with increasing frequency too, so I look forward to reporting back on our growth in another year or so. Voir l'article complet
  18. Guest post from Dr. Gregory Bowman, UC Berkeley Most rational drug design efforts assume the target protein exists in a single structure and that the structure of one region of the protein--called the active site--allows the protein to perform some function. Once this assumption is made, the only way to manipulate a protein’s activity is with inhibitors that bind the active site tightly enough to block it from performing its intended function. Unfortunately, this strategy only works for ~15% of proteins, greatly limiting the number of proteins we can manipulate for therapeutic purposes. In a recent article published in the Proceedings of the National Academy of Sciences (link), I showed that simulations run on Folding@home can reveal new ways of manipulating a protein's activity. Specifically, I start off by recognizing that proteins are actually flexible and then use Folding@home to enumerate the different conformations a protein adopts. I then use statistical analysis to find parts of the protein that can communicate with the active site through a process called allostery. These regions--called allosteric sites--are attractive drug targets as the binding of small molecules to them can be communicated to the active site, ultimately affecting activity. As a proof of principle, I showed that my approach can identify a known allosteric site in Beta-lactamase (see figure below). This protein is an important drug target because it can confer bacteria with antibiotic resistance by breaking down antibiotics like penicillin. I also use my approach to predict new allosteric sites in Beta-lactamase and two other proteins that play important roles in immune deficiencies and HIV. Now I'm performing experiments to test my predictions. It will require a lot more of your WUs, but I hope this type of approach can eventually lead to new pharmaceuticals. On the left is a structure of Beta-lactamase that most people would think of as the structure of this protein. However, the right shows a different structure with a drug (cyan) bound in a pocket that isn’t visible in the structure on the left. Binding of this drug somehow affects the structure near the active site (green). Using my approach, I’m able to start with the structure on the left and then predict the existence of the structure on the right and the allosteric site the drug is bound to. Voir l'article complet
  19. A brief update to our previous blog post on the A4 bonus: the bonus is now in effect. Voir l'article complet
  20. We've noticed a significant number of high priority projects are trailing behind existing projects. Newer projects are aimed at interpreting and guiding experiments where the full power of Folding@home (F@h) is essential to continue pushing the boundaries of scientific and medical discoveries. The main cause of this issue is the core version needed to run these simulations. Many of our newer SMP projects use the A4 core, which has several scientific advancements, while existing projects use the still important A3 core. The A4 core is not compatible with Clients below version 6.34 and many donors are still running these older Client versions. This presents an opportunity to encourage people to donate their cycles towards these vital A4 projects. To emphasize the scientific importance of these work units, we are boosting the base points of all A4 work units by 10% when uploaded (Note that this bonus will not be reported by V7 or by the 3rd party applications which project PPD but the points will appear when your statistics are credited). The quick return bonuses will be calculated on top of the increased base points. This will start on Monday July 23, 2012, and we will keep this 10% bonus in effect for at least 3 months as a trial period, but plan to keep it longer, as needed. To participate, donors should be running a recent version of the F@h Client. We strongly encourage Windows users to update to the much improved V7 Client. Although F@h Client v6.34 or newer is sufficient to participate for any supported operating system. Please note the Linux and OSX V7 Clients are a work in progress and feedback is welcomed. v7 Clients: Windows/Linux: Visit our home page, http://folding.stanford.edu/English/HomePage Mac OSX: v7 for OSX is still in testing. For a beta copy: https://fah-web.stanford.edu/projects/FAHClient/wiki/BetaRelease Old v6.34+ Clients Windows/Linux: http://folding.stanford.edu/English/DownloadWinOther Voir l'article complet
  21. Guest post from Dr. Xuhui Huang, Hong Kong University of Science and Technology In this post, I want to introduce a new GPU-powered clustering algorithm we recently developed to analyze the large molecular dynamics simulation datasets generated by Folding@home. Folding@home can generate enormous sets of protein structures. A critical step in analyzing these large datasets involves some form of reduction in the dataset, usually in the form of clustering. We recently developed a GPU powered clustering algorithm using the intrinsic properties of a metric space to rapidly accelerate the clustering. Overall, our algorithm is up to two orders of magnitude faster than the CPU implementation, and holds even more promise with the ever increasing performance in GPU hardware. This algorithm should facilitate numerous applications. For example, one of the systems we tested our code on is the human islet amyloid polypeptide (hIAPP) peptide, whose aggregation is implicated in Type 2 diabetes. We hope further analysis of this data will provide insights that will inform the development of treatments for diabetes. Voir l'article complet
  22. Guest post from Dr. Vincent Voelz, Temple University Using protein folding simulations alongside experiments remains challenging because the two techniques often "see" very different things. Simulation trajectories "see" every atom in a single protein in microscopic detail, while experiments often "see" only bulk properties averaged over large ensembles of molecules. For example, in the last few years, we have built kinetic network models of ever larger and slower-folding proteins. These models can have huge numbers of states and many possible folding pathways, yet experimental folding kinetics can be fit to models having only two or three states. In a new paper, we try to bridge these two levels of detail using a combination of simulation and experiment to study the early folding events of ACBP, a 86-residue helix-bundle protein that folds on the ~10 millisecond timescale, one of the largest, slowest-folding proteins we have studied to date. Previous experiments suggested that ACBP folds via a "three-state" mechanism, with an intermediate forming on the ~100 µs timescale. To understand the molecular events underlying the formation of this intermediate, we used Folding@Home to generate tens of thousands of GPU-accelerated trajectories, and stitched these together to build a kinetic network model of the complete folding reaction (see figure below). By comparing our model to the results of state-of-the-art experiments (single-molecule FRET, Trp-Cys quenching, and time-resolved FRET) we found something surprising -- the folding relaxation timescale around ~100 µs corresponds to the heterogeneous formation of unfolded-state structure, rather than some discrete structural state. This work is exciting because it shows that our models can predict atomically detailed mechanistic information about folding (currently very difficult to obtain experimentally) while simultaneously providing accurate predictions of quantities seen in bulk folding experiments. Voir l'article complet
  23. Guest post from Dr. John Chodera, UC Berkeley Kinases [http://en.wikipedia.org/wiki/Kinase] are the molecular logic gates of the cell. These important proteins integrate critical signaling information in every cell of our bodies, becoming active only when specific upstream signals are received. However, in many kinds of cancer, mutations can emerge in one or more kinases that cause them to ignore these regulatory signals and become active all the time. If these kinases are involved in cell division, this can erroneously cause cells to keep dividing even when they shouldn't, potentially resulting in a form of cancer. Our group [http://choderalab.org] is using Folding@Home to understand how some successful anti-cancer therapeutics (like imatinib [http://en.wikipedia.org/wiki/Imatinib]) are able to selectively target the targeted disease-causing kinases while minimally interfering with other normally-functioning kinases. A deeper understanding of this selectivity would help recapitulate the success seen in treating some cancers by aiding the design of novel therapeutics targeting other cancers. Up to now, the origin of this selectivity has been elusive because it appears that highly selective drugs like imatinib can bind in essentially the same way to the highly similar Abl and Src kinases, despite the fact that it binds Abl well and Src poorly (see Figure). It is now believed these differences in binding are due to conformational preferences of the kinase for different geometries, something that had been traditionally hard to study but is well-suited to techniques we originally developed to study protein folding problems on Folding@Home. Stay tuned for future updates on how Folding@Home is helping our study of kinase inhibitors and cancer! Voir l'article complet
  24. Guest post: Dr. Greg Bowman, UC Berkeley We just had a protein folding conference at Stony Brook University in New York that was extremely encouraging. Both the experimental and theoretical communities are very excited about the results we are generating with Folding@home. In particular, they are excited about (i) our increasing ability to make quantitative connections with experiments and (ii) the long timescale dynamics for large proteins we are now able to capture. For example, we recently succeeded in folding an 80-residue protein on 10 millisecond timescales (paper is here). For reference, that’s about twice as many residues and about 1,000 times longer timescales than what most anybody else is able to achieve! There are now multiple experimental groups who are asking us to make predictions for them to test. So, we appreciate all your help and have plenty of new calculations for you to contribute to. Voir l'article complet
  25. To start off FAHcon2012, I gave a talk which included a review of how far Folding@home has come in the last decade. I showed a slide from the very first talk I gave about Folding@home results. That talk was given at Columbia University in August of 2000, and I talked about results from our paper in Science entitled "Screen savers of the world, unite!". That work described the folding of a very small protein (16 amino acids) on a very short timescale (10ns = 10 x 10^-9 seconds!), but still was a major accomplishment for the time. It's exciting to see how far we've come. One way to think about it is in terms of how long of a time scale and length scale we can simulate for protein folding and protein misfolding diseases (such as Aß aggregation in Alzheimer's Disease): Time scales: advancing roughly 1000x every 5 years 2000: 1 to 10ns (Fs peptide) 2005: 1 to 10µs (villin, Aß aggregation of 4 chains) 2010: 1 to 10ms (NTL9, Lambda repressor) 2015: 1 to 10s? Just breaking past a microsecond was a big deal. The fact that we can simulate 10's of milliseconds is very exciting, but I'm really excited about where this appears to be leading, allowing us to tackle really challenging and important problems. It would also mean that through a combination of new methods, algorithms, and hardware advances, we've already increased our capabilities by a million fold in just 10 years (2000 to 2010). We're looking forward to hopefully making it a billion fold in 2015! Length scales: advancing roughly 2x every 5 years 2000: 16 amino acids (Fs) 2005: 35 amino acids (villin) 2010: 80 amino acids (lambda, ACBP) 2015: 160 amino acids? It's also important to note that these are sizes for protein folding. For other problems, such as protein conformational change, we've already tackled much bigger systems. I'm really excited to see what the next 5 years will bring! Voir l'article complet
×
×
  • Créer...