News

New Fermi OpenMM and Gromacs 3 projects: p6806 and p6881

By , the in Folding@Home Project - No comment
Yu-Shan Lin, who is now becoming quite prolific, has put two new projects online.

p6806
This GPU3 project is geared towards Fermi-generation GPUs. This project comprises part of the second phase of study on the Abeta amyloid peptide, the results of which will be used to build Markov states for the Abeta monomer.

This project also serves to evaluate the performance of the new servers and the latest version (2.15) of Core 15 for the Fermi. The units are distributed by the server at 171.64.65.64 and yield 1,348 points each. Each unit is comprised of 634 atoms, and should preferably be completed and returned within 5 days, but no later than 13 days.

p6881
This Gromacs Core 3 project is part of the same series of studies as p6806, but focuses on a mutant version of the Abeta amyloid peptide, simulated in water.

These units are distributed by the server 171.67.108.33, and are comprised of 12,233 atoms. They yield 69 points, and should preferably be completed and returned within 14 days, and no later than 21 days.

Sources: p6806 / p6881

New Gromacs 3 project: p6885, and a milestone for FAH-Addict EN

By , the in Folding@Home Project - 2 Comments
This project will complete the studies of Yu-Shan Lin on the Abeta amyloid peptide, which is a component of plaques that appear in the brains of people suffering from Alzheimer's disease.

This project uses the Gromacs 3.1 core (FahCore_78) and simulates the peptide in an explicit water solution. The project is distributed from the server at 171.67.108.33. Each unit contains 12,193 atoms and is worth 69 points. The preferred deadline is 14 days, with the final deadline a week later.

Source: Official Forums

In other news, FAH-Addict's English version recently passed 100,000 visits, as you will be able to see at the top right of the page. The team thanks you for your continued readership and we hope we can continue to improve the site and make it worth your while to visit :)

GeForce GTX 570 specs leaked

By , the in Hardware - No comment
GeForce GTX 570 specs leaked
While the announcement is expected next week, manufacturer Palit has leaked the specifications of nVidia's new GeForce GTX 570.

This chip is a derivative of the GF110, which is used by the GTX 580. As we know that this card will only have 480 shaders, it is probably safe to assume that these are made from chips with one defective group of shaders which could therefore not be sold as GTX580s.

The shader units should be clocked at 1464MHz for the GTX 570, compared to the 1900MHz memory and 732MHz of the main chip, while the TDP stands at 225W.

The GTX500 range marks a shift in the highest-performing card, with the GTX570 being the closest replacement to the old GTX480, and the GTX580 positioned above.

The price of this new model remains to be seen, but initial estimates put it at somewhere above €250, and performance can reasonably be expected to be somewhere between that of the GTX480 and GTX580.

Is the Sandy Bridge GPU compatible with OpenCL?

By , the in Hardware - No comment
Is the Sandy Bridge GPU compatible with OpenCL?
Two weeks ago, to general indifference, Intel released the first beta of its OpenCL 1.1 SDK for the Sandy Bridge GPU. This means that the GPU in Sandy Bridge is definitely compatible with DirectX 10 or higher, with unified shaders capable of complex processing.


Sandy Bridge architecture



OpenCL is normally a standard independent of hardware (that is, it can run on CPUs and GPUs of any variety), so the future OpenCL version of the GPU cores could work right out of the box with the Sandy Bridge GPU. Of course we should not expect performance comparable to the best from nVidia and AMD, but we remain curious as to the potential performance of this chip. We should know next month... assuming it arrives on time...

Source : PCInpact

Three new uniprocessor projects: p11289, p6880, p6886

By , the in Folding@Home Project - No comment
Although it has been a relatively uneventful fortnight for Folding@Home, Yu-Shan Lin has nonetheless been quietly continuing his research around the amyloid-beta peptide and has released three new projects for uniprocessor clients.

p11289
In this project, the peptide is simulated in water using the ffamber99sbe force field.

The project is distributed by the server at 171.67.108.33. Each unit contains 12,219 atoms, and yields 69 points. These units should preferably be returned within 14 days, but no later than 21 days.

p6880
In this project, the peptide is simulated in explicit water.

The units have the same characteristics as project 11289, except the number of atoms stands at 12,222.

p6886
Finally, project 6886 simulates a a full-length amyloid-beta monomer mutant in explicit water.

Each unit contains 12,229 atoms, and is otherwise identical to the two previous projects.

All three projects utilise the Gromacs 3 core (Fahcore_78).

New BigAdv Gromacs A3 project: p6900

By , the in Folding@Home Project - 5 Comments
After a long wait - and a unit shortage - Peter Kasson has finally announced the arrival of a new BigAdv project. Designated project 6900, it is a Windows-only continuation of the studies carried out by previous projects in the 268x series.

The performance and characteristics will be more or less the same as that of its predecessors: 1,098,185 atoms, 8,955 point yield, and a k-factor of 26.4. This project is geared towards octo-core and higher machines; deadlines are then reduced accordingly: a preferred deadline of 4 days (to take advantage of the bonus), and a final deadline of 6 days.

European folders will be pleased to learn that this project is distributed by the server at 130.237.232.141, located at the Royal Institute of Technology in Sweden.

Source: Official Forum

New Gromacs A4 project: p10425

By , the in Folding@Home Project - 1 Comment
Kyle Beauchamp has just released project 10425, which makes use of the brand new A4 core.

The simulation is performed at 278K, sans denaturant, using the recently-developed Amber99SB-ILDN force field.

These Windows-only units are distributed by the server at 171.64.65.84, and are comprised of 576 atoms. They yields 193 points total, have preferred deadlines of 13 days, and final deadlines of 19 days.

Source: Official Forum

New GPU3 project for Fermi: the serious work begins

By , the in Folding@Home Project - 1 Comment
Yu-Shan Lin continues to fuel the GPU3 client with new projects, and has just added her latest: p6811. This is only available to clients running -advmethods, but marks a turning point for what we can expect from GPU client work units.

This project constitutes the first phase of a study on the prion protein monomer. Prion is a proteinaceous infectious particle that is responsible for prion diseases such as scrapie in sheep, Bovine spongiform encephalopathy (BSE) in cattle and Creutzfeldt-Jakob disease (CJD) in humans. These simulations are the Pande Group's first attempt to sample the configurations of the peptide and the results will be used to build Markov states for further sampling.

The project is distributed from the server at 171.64.65.64 and has 2150 atoms, making it the biggest system yet released for the GPU client. The WU is worth 7202 points, and with a preferred deadline of 13 days and a final deadline of 18 days.

As a result of the number of atoms in these units, their behaviour may be different to what is expected. The extra size means that they work newer GPUs much harder than before, and there have been reports of previously stable cards overheating, so watch your temperatures carefully. On "lower-end" Fermi cards (which for the purposes of this simulation means anything less than a GTX460), this unit will saturate the GPU and may lead to a significant loss of performance.

Tobit on the official forums has run performance tests for the most common Fermi-based cards, and his results are thus:

  • GTX480 (700 core/1400 shaders): approximately 14,142 PPD
  • GTX470 (770 core/1540 shaders): about 14 500 PPD (overclocking is important here as it compensates for the loss of some shaders).
  • GTX460 (675 core/1350 shaders): about 8550 PPD (9371 PPD on Tobit's card, but he did not specify the frequency)
  • GTS450 (850 core/1900 shaders): about 6914 PPD

As you can see, the cards with fewer shaders perform significantly less well than their more powerful brothers, this was not shown previously as other units do not saturate the cards so much. Only the most powerful cards show little performance loss (the GTX470's reduced number of shaders compensated for by its higher clock speed).

Sources: Project Description and the Official Forum

New GPU3 Fermi project: p6800

By , the in Folding@Home Project - No comment
Yu-Shan Lin has released a new project for the GPU3 Fermi client: project 6800. It is currently available only to users running their client with the -advmethods flag enabled.

This project is comprised of second-stage sampling simulations of a single full-length amyloid beta (Abeta) monomer in water. This peptide is a pathological marker of Alzheimer's disease in the form of plaques in patients' brain. The simulated system is comprised of a full-length Abeta in water. The results will be used to build models of Markov states for the Abeta monomer, and will serve as a starting point for future oligomerization studies.

These units are distributed by the server at 171.64.65.64 and is comprised of 626 atoms. They each yield 1,298 points, and should preferably be completed and returned within 4 days, but no later than 10 days.

Source: Project Description

Further information about the GTX580

By , the in Hardware - 3 Comments
Further information about the GTX580
If you saw our previous news... ignore it, we have been duped! The GTX580 is not a die-shrink after all, it is in fact a new chip codenamed GF110, which retains many of the characteristics of the GTX480, and (finally!) activates all 512 SPs available on the chip.

The changes are mainly intended to reduce the power draw/amount of heat produced, as the GTX480s are well known for their hot-running nature. In this respect, the new design seems to have been successful. It has a maximum power draw 6W lower than the GTX480, despite having the additional 32 SPs and running at a stock speed of 772MHz (GTX480 had a default of 700MHz). This operating speed approaches that of the GTS450, the current folders' dream card.


Click image to enlarge.


For cooling, nVidia has developed a clever phase-change cooling system, which in practice works the same way as heatpipes in other more standard coolers.


It is a very promising GPU for units with large systems (yes, such units have been spoken of for a long time and we are still waiting) but the price may put potential buyers off, as it is set at about €500.

On the other hand many sources said that this chip has had some HPC (High Performance Computing) functions removed, to allow the removal of 300 million transistors, so we were curious to see what impact these changes would have on Folding@home.

Only a practical test would allow us to see what the effect is. Fortunately, Hardware Canucks have obliged with some numbers on three classes of GPU WU:


The average gain is 17%.
From the raw numbers, the gain we would have expected would be: (512/480 * 772/700) = 17.6%
So we can conclude that the removal of the HPC functions has not negatively affected Folding@home performance. Great news!

Sources: PC INpact, Hardware Canucks