News

New uniprocessor projects: 10200-10205

By , the in Folding@Home Project - No comment
Greg Bowman has just released a new series of projects utilising the latest A4 core.

According to the project description, this series of projects focuses on sampling the accessible conformations for b2AR, a membrane-bound G-Protein-binding receptor (GPCR) with seven trans-membrane helices. b2AR is one of only a few (there are 1000 total candidates) GPCRs for which the structure has recently been determined.



The beta-2 adrenergic receptor surrounded by solvent (H2O)


GPCRs are of great importance because they are involved in most of the exchange process signals across the cell membrane and, moreover, as a therapeutic target, they represent nearly half the drugs currently on the market.

The team hopes that through the identification of accessible conformations for b2AR, it will be able to identify the protein's important conformational states, and the corresponding thermodynamic properties. The team also hopes to apply the knowledge they acquire to other GPCRs whose structures have not yet been resolved.

These simulations are conducted using the Amber03 force field with Berger parameters for the lipid molecules, solvated using the TIP3P water model.

These projects are distributed by the server at 171.67.108.39. Each WU has 58,724 atoms, and are 0.5ns-scale simulations. They will yield 493 points per unit, and should preferably should be completed in less than 92 days, but no later than 136 days.

Source: Project Description

Folding@Home and Ubuntu 10.10: a rocky start

By , the in OS - 3 Comments
If you plan on upgrading your version of Ubuntu to 10.10, then beware. The SMP2 client is currently experiencing technical difficulties on Maverick, in the form of a segmentation fault upon startup:

fah6: ../sysdeps/unix/sysv/linux/getpagesize.c:32: __getpagesize: Assertion `_rtld_global_ro._dl_pagesize != 0' failed.
Segmentation fault


There doesn't yet seem to be any working fix available, but when it comes it will likely be similar to the library hack we reported on the last time F@H and Ubuntu had a spat. One promising potential fix that we found involved manually initiating the nscd daemon before starting up the client, but alas it proved not to work.

The output given by the client would seem to indicate the client cannot work when there is no swap/page partition available, however this error is thrown up even on systems with an available swap partition.

Feel free to provide your own feedback, and if you discover a fix before we do, we'd appreciate if you share it with us!

Edited 04/11/10:

A workaround has been discovered; following are the console commands and instructions:

Execute sudo apt-get install nscd
Open /etc/nscd.conf with a text editor
Find the line enable-cache hosts no and change "no" to "yes"
Save the file and quit the editor
Execute sudo service nscd restart

Your folding client should now function correctly. Hopefully the v7 client won't require any such intervention when it is released!

Folding with a Radeon HD6800? Possible, but underwhelming!

By , the in Hardware - 3 Comments
Folding with a Radeon HD6800? Possible, but underwhelming!
It seems we were wrong when we announced in our recent news that the Radeon HD6000 would need a new client and core in order to fold.

Though we were wrong about the current client being incapable of folding with the cards, we were right that the performance would not be of the first order.

This has been confirmed by two users on the official forum, Tobit and mdk777. The current client (6.23) is perfectly capable of folding with AMD's new cards if one simply starts it up with the -forcegpu ati_r700 flag, just as one would with a Radeon HD5000.

Unfortunately, the performance figures are disappointing; the current core is unable to efficiently take advantage of these chips' additional calculation units.

mdk777 got 3,153 PPD (p5733) on his non-overclocked HD6850. With the fan speed manually set at 70%, he was able to keep the GPU at 50°C. The difference in power consumption between the card idling and folding was 115W.

For his part, Tobit experimented on a HD6870. He received 3,618 PPD, but unfortunately didn't provide any information on the operating temperature or power consumption of the card.

By way of comparison, we invite you to read (or re-read) Radeon HD5770 test.

Meanwhile, we'll just have to wait for the Radeon HD6900's release at the end of November, and the new OpenCL core, and hope to see some performance increase from the red side of the force.

Source: Official Forum

GTX 580 rumours accumulate

By , the in Hardware - 4 Comments
GTX 580 rumours accumulate
Following a brief posting on their website, nVidia has discreetly let slip the possible existence of a GeForce GTX 580 in its latest 261 beta drivers. Don't expect a great revolution from the possible new card, however. We told you last week about the new "tick-tock" strategy adopted by nVidia, which means that GTX 580 might end up essentially being a die-shrunk version of the current GTX 480.

Since nVidia doesn't tend to die-shrink solely to reduce chips' power consumption, we can suppose that the new manufacturing process will come alongside either a clock frequency rise for the GPU - which would, we have to admit, be very welcome - or an increase in the number of shader processors.

Source: PCINpact (in French)

Round-up of new projects from the last two weeks

By , the in Folding@Home Project - No comment
Since we have been a bit quiet recently we missed some project announcements. The following are the projects released since our last such news on October 10th:

Classic client
Project 10412 was recently released from adv methods to general folding@home, so will be assigned to clients with no special flags set. Further details can be found in our news about the a4 core.

GPU2 client
Greg Bowman has launched project 10111 for the GPU2 client on nVidia. These units complete the study of ubiquitin from project 10109, and are distributed from the server at 171.64.65.71. They contain 1174 atoms and are worth 494 points, with a preferred deadline of 2 days and a final deadline of 3 days.

GPU3 client
Yu-Shan Lin has released two series of projects for non-Fermi cards running the GPU3 client: projects 11161-11168 and 11169-11181. These projects simulate the beta-amyloid peptide and utilise the generalised born method for modelling implicit solvent. They are distributed from the server at 171.64.65.31, and consist of between 600 and 630 atoms each. For your time you will get around 1300 points, assuming they are returned before the preferred deadline of 20 days or final deadline of 29 days.

ATI Catalyst 10.10 drivers out with OpenCL support

By , the in Drivers - 4 Comments
ATI Catalyst 10.10 drivers out with OpenCL support
In addition to announcing the 6850 and 6870, AMD has released a new version of its drivers: Catalyst 10.10.

This marks the first time AMD has promptly released drivers compatible with the latest generation of their graphics cards; previously, buyers would expect to wait for up to a month, if not longer, before they could enjoy full support for their equipment.

Another important milestone for AMD and F@h is that this driver is the first to include the functionality required to run OpenCL applications. This version has the quaint subtitle of Accelerated Parallel Processing (APP) Technology Edition.

The driver can be obtained from AMD's driver download page.

AMD Radeon 6850 and 6870 announced

By , the in Hardware - 1 Comment
AMD Radeon 6850 and 6870 announced
Today, the new generation of ATI graphics cards was announced. Or should we say AMD graphics cards, the first time this designation has been officially applied. The cards in question are the Radeon 6850 and 6870.



The two cards, featuring a classic design.



While being classified as x8xx cards, you'll quickly see that the cards would fit better in the 6700 range; while these new chips have an increased number of stream processors compared to the HD 5750 and 5770, yet remain below what is offered with the 5850 and 5870.

In addition, as with the 5700 range cards, AMD has opted not to include support for double precision operations with the 6800s. This won't pose any problems for GPU folding, but may prove disappointing for users of other GPGPU calculation programs.

On the architecture side, AMD retains its 5-dimensional vector calculation units; only one component is able to handle complex functions (such as sine, cosine, division, logarithms, etc.). The other four are limited to simple functions such as addition, multiplication, and multiplication combined with addition (MAD). This architectural flaw adversely affects the performance of complex calculations such as those used by Folding@home. Fortunately, this type of architecture is otherwise very effective for 3D rendering, some BOINC projects, and simple mathematical operations.

The GPU powering the 6850 will feature 960 computation units, while the 6870's will feature upwards of 1120. Clock frequencies seem set at around the 775MHz mark, and will of course vary from manufacturer to manufacturer.

A strong point for the cards are their reasonable prices; around €160 for the 6850 and €210 for the 6870, in addition to a reduced power consumption over their predecessors. A third pro is an efficient heat-management system, which allows for high utilisation without the noise to match.

ATI should be announcing their plans for new high-end cards in late November.

GPU folding with these cards will not be possible without a core - and probably also client - update. Core 16 and the v7 client, perhaps?

GTS 450 and GTX 460; great performance, and more to come?

By , the in Hardware - 1 Comment
GTS 450 and GTX 460; great performance, and more to come?
The success of nVidia's GTX 460 and GTS 450 in folding is becoming more evident. In particular, the GTS 450 can easily produce 14,000 PPD at a very low rate of energy consumption.

However, it seems even now, they are not yet being fully exploited; the GTX 460 still possesses hidden reserves of strength that existing units are not exploiting. In our hardware test, we told you that we achieved less than 9,000 PPD, but the newer units are grossing up to 11,700 PPD. In this post on the Folding Forum, Imran Haque talks about his hopes for performance gains on the card, comparing PPD gains with the GTX 480.

So, will this run of form continue?

Yes and no, as PC-INpact's news announces the release of a successor to the current cards, due for release in 2011. The novelty is that this chip will not be brand new, as nVidia seek to move to a tick-tock strategy similar to that used to great effect by Intel. This will entail alternating improvements between chips, as one generation will bring a new architecture and then the next will bring a die shrink. The launch of Fermi was a fiasco as nVidia attempted to produce a new architecture on a new manufacturing process at the same time, leading to low yields and lots of chips needing to be binned as useless.

The new Fermi chips with simplified design (such as that used on the GTX460) are an indication of how the process is now much more controlled and simplified. The design will remain in 2011, but we should see a die-shrink for GF100, GF104 and GF106 that should allow faster clock speeds than are currently possible.

Fermi will be replaced with a new architecture in 2012, using the manufacturing process that will be introduced in 2011. In passing we note that nVidia continues to name its chips after great physicists, with the next generation cards to be named Kepler and Maxwell. We anticipate the nVidia Pande at some point.

Sources: Official Forum and PC INpact

-smp 128 inside

By , the in Hardware - No comment
Oracle (née Sun) has announced the SPARC T3, featuring 16 cores, with the ability to handle eight threads each, for a total of 128 threads. The 40nm CPU has a limited 6MB L2 cache, a maximum clock speed of 1.65GHz, and a TDP of 139W, in addition to SMT (simultaneous multi-threading) capabilities.

It is true that these processors are nothing like the Phenom and Core processors we are used to today, given their different target market (servers), however it may give us a clue to the way desktop processors will evolve in future.

The potential of this CPU for folding BigAdv units isn't in question, but unfortunately there are currently no cores available for the SPARC architecture.



Source: Tom's Hardware (in French)

New nVidia drivers released: 260.89 WHQL

By , the in Drivers - No comment
New nVidia drivers released: 260.89 WHQL
A major driver update for nVidia graphics cards has just been released.

Version 260.89, geared primarily towards Windows users, features a new and much simpler installer; the predecessor of which was beginning to seriously date, despite the pretty cosmetic improvements.

This version of the driver supports all GeForce cards beyond the 6 series, including the new Fermi-based GT430, and carries over all the innovations introduced in previous versions.

The driver can be obtained through the driver selection page over at nVidia's official website.