21)
Message boards :
Number crunching :
CPU performance
(Message 57987)
Posted 17 Dec 2008 by AlphaLaser Post: AFAIK, Rosetta is bound by FPU performance. As for using both, yes it is possible, as most modern CPUs are Superscalar. That means that it will try to use all of the execution units available (ALU, FPU, SIMD). However, programs don't always have a good mix of instructions that utilize all the units fully. That's where SMT (hyperthreading) steps in because there is a better chance of using all the resources when there are more instructions to pick from. Since Rosetta contains mostly FPU instructions and is not highly bandwidth constrained, it does not benefit as much from SMT as would other applications. |
22)
Message boards :
Number crunching :
I this sounds silly...but...
(Message 57609)
Posted 4 Dec 2008 by AlphaLaser Post: Possibly related is BOINCoid, the port of BOINC to the Google android phone. They have a similar idea of having BOINC run while the phone is charging, and hope that the computational power can be realized through installing it on enough phones. I'm not sure what the current status is, you'll have to check the website. |
23)
Message boards :
Number crunching :
Intel i7 CPU
(Message 57608)
Posted 4 Dec 2008 by AlphaLaser Post: We will most likely see the L2 cache size problem addressed in Westmere. AnandTech published an article where Intel recognized that the current cache situation is not ideal. A larger L2 size on the current process is hard to justify because that would decrease the effective size of the L3 cache, due to its inclusiveness as Mastergee pointed out. A larger L2 would have to come with a larger L3, which can be made possible with the extra transistors in 32 nm. What would be interesting is that if HT for Rosetta is limited due to the shared use of the FPU, then would there be less of a performance drop when running an integer-heavy BOINC project on the other logical core. If this were true, then the ideal case is for BOINC or the OS to set process affinity such that each physical CPU runs an ideal mix of work. |
24)
Message boards :
Number crunching :
RAC Total Flawed
(Message 54859)
Posted 3 Aug 2008 by AlphaLaser Post: AFAIK, BS-RAC is based on a 30-day moving average of daily credit. They do provide a separate RAC figure based on exponential decay that BOINC uses, though. Its not shown on the signatures but is provided on the detailed stats page. |
25)
Message boards :
Number crunching :
F@H adds nVidia GPUs to crunch on
(Message 53857)
Posted 19 Jun 2008 by AlphaLaser Post: nVidia also has a new series of GPU's (GeForce GTX 260/280) coming whose performance rivals the SLI setups of today. The die size is enormous; 1.4 billion transistors on a 65 nm process! AnandTech posted a lengthy review which has a short section on Folding@home. |
26)
Message boards :
Number crunching :
BOINC in Java code
(Message 53501)
Posted 1 Jun 2008 by AlphaLaser Post: Something that may be of interest is Intel's recently launched Atom CPU, an x86-compatible chip designed for smart phones, PDAs, and other ultra-mobile devices. An early SuperPI benchmark shows it edging out a 1.13 GHz PIII. Here's a CrystalMark 2004 benchmark (website translated from German), showing it competing against Via's C7 and Isaiah and Intel's own 1.6 GHz Celeron M. Also note that it has two-way multithreading ("HyperThreading"). SPECint2KRate (2.5 MB PDF, page 17) shows scaling numbers for HT. This might enable BOINC to be easily ported over to a mobile device, attributing to the x86 compatibility. The downside is that Intel is just beginning to enter the embedded market that's dominated by other architectures such as ARM, so we'll have to wait and see whether they'll be able to compete. Another thing is it doesn't help put BOINC on existing, non x86 devices. My friends and I ported the C++ BOINC & SETI to Java - sf.net/projects/boincoid. However this seems promising, throughput may not be great but it would still be very neat if my phone was doing something! ;) |
27)
Message boards :
Number crunching :
xml to html
(Message 52006)
Posted 18 Mar 2008 by AlphaLaser Post: Depends on what you want to do, but you may want to look into XSLT stylesheets. |
28)
Message boards :
Number crunching :
Switch from W.XP to Linux.
(Message 50940)
Posted 24 Jan 2008 by AlphaLaser Post: I have a few linux varieties set up on a virtual machine. It is a nice way to try out different distros and learn about things without the fear of messing up anything or installing to a physical disk. |
29)
Message boards :
Number crunching :
How can I minimize memory usage?
(Message 50849)
Posted 20 Jan 2008 by AlphaLaser Post: If anything, the runtime preference should be set as high as possible. This means your computers contact Rosetta less often, reducing server load. The long runtimes do not impact the significance of the science being done in any way. |
30)
Message boards :
Number crunching :
Bloomfield worth the wate?
(Message 50825)
Posted 20 Jan 2008 by AlphaLaser Post: The initial estimates for Bloomfield's release are sometime late this year. It'll be the 'Extreme'/enthusiast brand for the Nehalem architecture. Nehalem |
31)
Message boards :
Number crunching :
new to rosetta using q6600
(Message 50656)
Posted 13 Jan 2008 by AlphaLaser Post: im using boinc, and only rosetta ;) im new to rosetta and i was wondering: everest ultimate says all 4 cores are 100%, so if there any additional setup in order to use 4 cores fully ? also, does this program continue the simulation after i shut off the computer and restart it ? thanks, any other suggestions or comments or issues i should know about ? thanks If BOINC is installed as a service, then BOINC will automatically start crunching after the computer is restarted, even if no one is logged in. Under any other install type, BOINC starts crunching only when logged in and BOINC manager is started. |
32)
Message boards :
Number crunching :
Sponsored Computer for anyone interested
(Message 50603)
Posted 12 Jan 2008 by AlphaLaser Post: If you were buying computers just for grid computing, the 40GB version of the Play Station 3 would seem to be the most efficient, but unfortunately not all projects have PS3 clients written for them. These are theoretical capabilities but on the other hand it is not so easy to create an application that will fully utilize the PS3's computing power. An application must be very parallelizable and be able to fit in the PS3's memory capacity and bandwidth constraints. The software difficulties are a big reason why PS3's aren't being harnessed very well by BOINC projects. |
33)
Message boards :
Number crunching :
Difference between GHZ and FSB?
(Message 50602)
Posted 12 Jan 2008 by AlphaLaser Post: When comparing CPU's between two different architectures, making a speed comparison based only on clock speeds is no longer valid. Different CPU architectures can have different instructions per clock (IPC) among other factors that make that architecture fast or slow irrespective of the clock. For example, the Pentium M/Yonah architecture has a short, efficient instruction pipeline whereas NetBurst's is quite long, resulting in slow throughput clock-for-clock. NetBurst was originally designed to overcome this problem by clocking to extremely high frequencies (5 GHz+). Unfortunately NetBurst hits a thermal and power "wall" long before that happens. |
34)
Message boards :
Number crunching :
Better Graphics and resolution in Rosetta?
(Message 50154)
Posted 28 Dec 2007 by AlphaLaser Post: The screensaver mentions something about being "low resolution". But I can't find a way of altering to "high res". Am I missing something? I'd wager that the "low resolution" you are referring to has to do with the ab initio search, where the proteins are modeled as a highly simplified structure. This simplification allows a wide variety of protein conformations to be tested with fewer computations. The other type of search Rosetta uses is the full relax or high resolution search where the protein is modeled with much more complexity, which helps to pin down a more precise prediction for the protein. In that sense the resolution only has to do with Rosetta's internal algorithms for computation and nothing to do with the screensaver graphics resolution. |
35)
Message boards :
Number crunching :
Longer target CPU settings??
(Message 42164)
Posted 14 Jun 2007 by AlphaLaser Post: Unless you want to be really picky and count the second or less it takes BOINC to end one result and start the next ;) |
36)
Message boards :
Number crunching :
Effect of Memory on performance
(Message 42074)
Posted 11 Jun 2007 by AlphaLaser Post: I believe DIMMs, which refers to a general construction of the RAM module, is being confused with DDR, a more specific type of RAM which has higher throughput but requires matched pairs of modules. |
37)
Message boards :
Number crunching :
rosetta not using as much ram as i have to give on mac pro
(Message 41886)
Posted 6 Jun 2007 by AlphaLaser Post: There is really no benefit to extra RAM once you have enough to run Rosetta, your OS, and any background tasks you have running. Basically, below a certain amount of RAM you are slowing down Rosetta because your apps are swapping pages, which involves slow disk access. However, once you've gotten to the point of not needing to swap as you are now, there no longer is a performance penalty nor a performance gain. Memory bandwidth/latency and particularly processor will have a much bigger impact on speeding up Rosetta rather than memory size at this point. Good news is, you can still benefit from all that memory if you start a memory-heavy app while BOINC is running. Also if you run other BOINC projects they can stay in physical RAM when they are not running. |
38)
Message boards :
Number crunching :
Lost Results
(Message 30137)
Posted 27 Oct 2006 by AlphaLaser Post: currently running version 5.34, computer is running fine, no errors, total credit and average credit both going up but the results keep dropping. If you are referring to the number of results listed on your Account Pages, then a dropping number is possible since results are purged from the BOINC database at regular intervals. This is done to keep the database as small/fast as possible. |
39)
Message boards :
Number crunching :
Please help
(Message 29846)
Posted 23 Oct 2006 by AlphaLaser Post: The % complete tends to stay the same when in the middle of calculating a decoy. It jumps when the decoy is finished. If it is going to take longer than your runtime pref, Rosetta will still try to make one prediction, but you will not see much progress until it finishes. There is a watchdog thread which I think stops a WU if no step has been calculated in 1 hr. |
40)
Message boards :
Number crunching :
No Rosetta Wu's being downloaded
(Message 29608)
Posted 19 Oct 2006 by AlphaLaser Post: Visit your Rosetta Preferences and change 'Target CPU run time' |
©2024 University of Washington
https://www.bakerlab.org