Posts by MikeMarsUK

1) Message boards : Cafe Rosetta : Leaving the Project (Message 48269) Posted 2 Nov 2007 by MikeMarsUK Post: As far as I can tell he's posted the same advertisment* on every project he's a member of. If it had included some critical comment and no image, then I don't think anybody would have complained. * An advertisment is still an advertisment even if the thing isn't commercial in itself. I'm not sure, but it's possible the image may violate photobucket's terms and conditions. Is it a 'promotional material'? http://photobucket.com/terms Section 5, restrictions, paragraph d, "Post, upload, email or otherwise transmit, without first obtaining the permission of Photobucket, any unsolicited or unauthorized advertising, promotional materials, junk mail, spam, chain letters, "pyramid schemes," Ponzi schemes or any other form of solicitation;"
2) Message boards : Number crunching : Default No. of CPUs (Message 43070) Posted 2 Jul 2007 by MikeMarsUK Post: The default setting was increased due to increasing core counts. It won't use more cores than you actually have on any given PC.
3) Message boards : Number crunching : Let's stand up to the -1 moron (Message 42684) Posted 26 Jun 2007 by MikeMarsUK Post: There is an alternative here: http://pirates.spy-hill.net/glossary/index.php?title=Forum_rating_points This is the rating code from the pirates project, which introduces things like daily limits and so forth. But I personally prefer the simpler solution of disabling them entirely. For reference the option is set in config.xml. <no_forum_rating/> http://boinc.berkeley.edu/trac/wiki/ProjectOptions
4) Message boards : Number crunching : Let's stand up to the -1 moron (Message 42636) Posted 26 Jun 2007 by MikeMarsUK Post: We disabled this pointless rating feature on the CPDN/boinc forums a while back, it appears to only be used for flame wars and the like, rather than it's intended purpose. If someone is going through and giving dozens of -ve ratings per day to loads of posts and people blindly, they're obviously not interested in the project itself and I don't see what would be wrong in blocking their account in that case. However simply removing the functionality would be more diplomatic. It would be interesting to know who the top -ve rater is ... so, project staff, how about pulling it out of the DB + publishing it? :=)
5) Message boards : Number crunching : Quad Core Intel Price Drops (Message 41938) Posted 7 Jun 2007 by MikeMarsUK Post: "Who?" wrote: how do you like this? who? 8-Core 45nm Penryn box http://www.legitreviews.com/article/521/1/ So when are you putting this box on Rosetta? :-)
6) Message boards : Number crunching : Please remove the .exe extension for work units while sending work units (Message 41199) Posted 20 May 2007 by MikeMarsUK Post: ... My company has a policy that executables shouldnt be downloaded. ... If his company has a policy against running unauthorised executables, then surely he shouldn't be running Boinc there without getting permission for it first?
7) Message boards : Number crunching : Stop or pause BOINC at next checkpoint. (Message 39252) Posted 10 Apr 2007 by MikeMarsUK Post: There is a debug setting you can use to get Boinc to display the checkpoints in the messages tab. 5.8.15 - <task_debug> 5.8.16 - <checkpoint_debug> These are configured via cc_config.xml (see Ageless's FAQ on the main Boinc forum for more details). Be aware that some projects (for example, Einstein) will generate huge amounts of text from some debug settings.
8) Message boards : Number crunching : FAQ revamp - your comments welcome (Message 35816) Posted 31 Jan 2007 by MikeMarsUK Post: I don't know if this helps, but this is where our equivalent is (the 4 README postings): http://www.climateprediction.net/board/viewforum.php?f=36. There are differences of emphasis because the projects are so different of course. The models, vs tasks bit might also explain about the differences between different tasks (many bases, few, different algorithms selected), and what effect this has (why you sometimes only get one decoy done versus many). Perhaps a description of simulated annealing, full atom relax, the physics behind it all, ... etc etc.
9) Message boards : Number crunching : Just moved my V8 to an other project for 2 weeks. (Message 35797) Posted 31 Jan 2007 by MikeMarsUK Post: Vista is supposed to be better than XP on many-cored SMP systems (more than about .. 2, 4, ?? I have no idea). Would be interesting to see if you see an improvement after migrating from XP.
10) Message boards : Number crunching : COST comparison: 8-core cruncher (Message 35770) Posted 30 Jan 2007 by MikeMarsUK Post: It's very easy to significantly overclock the 6600 (to around 3.5GHz), on the other hand it's extremely difficult to overclock the Clovertown. You can actually get the 6600 for a little less than the price you found, one place did it for 208 but I found several doing it for 280 (I have to wonder whether the lower price was an error). I don't know if you can get a sufficiently powerful PSU for the clovertown for $30 (Who?'s one was a 1KW unit I think?) On the other hand I'd spend more on the M/B and memory for the 6600 system, which probably balances out the costs mentioned above. Secondly, if you only have 2 DIMMs, you'll experience poor performance, this has been observed at a number of projects (typically 25% less performance with 2 DIMMs). Personally I'd very much like an Octocore system, but from my viewpoint based on prices in the UK it won't be economic until the costs drop in the second half of the year (when there is competition). An alternative at that time would be overclocked quads. In the UK the prices are skewed strongly in favour of the 6600 system - the same clovertown setup would cost around £1200 ($2350), which would get you 3 - 4 overclocked 6600 systems.
11) Message boards : Number crunching : 8-core cruncher? (Message 35605) Posted 27 Jan 2007 by MikeMarsUK Post: Quad Penryn is still just two dies stuck together in the package! :-( Had been hoping for something a bit more sophisticated. Perhaps a better quad is what is coming in August. http://www.reghardware.co.uk/2007/01/27/intel_reveals_penryn/ On the other hand, there are process improvements which the Inq explains better than the Reg: http://www.theinquirer.net/default.aspx?article=37253
12) Message boards : Number crunching : 8-core cruncher? (Message 35547) Posted 26 Jan 2007 by MikeMarsUK Post: ... The applications are not program like this, if you run 4 copies of the same program, you ll get 4 different memory buffers allocated, giving a 100% fit to NUMA, while, in reality, 3DsMax allocate one (AND ONLY ONE) memory space, and all the cores and threads get the data from the 3D world database ... giving NO SPACE to NUMA (Non uniform Memory architecure). ... From what you're saying, it sounds like Barcelona would be perfect for a heavily floating-point Boinc application then, since they run as different processes with different data areas, and hence NUMA will work well. But this is all speculation, and 53xx isn't what Barcelona should be compared to anyway - it's the next generation of Intel quads which is the relevant comparison.
13) Message boards : Number crunching : 8-core cruncher? (Message 35521) Posted 25 Jan 2007 by MikeMarsUK Post: True... obviously getting a bit late for me ! :-)
14) Message boards : Number crunching : 8-core cruncher? (Message 35516) Posted 25 Jan 2007 by MikeMarsUK Post: Squaring is powers of two :-) Each drop in process size gives you approx double the number of transistors (although processes from different manufacturers have different capabilities, and different designs may use the extra transistors in different ways).
15) Message boards : Number crunching : 8-core cruncher? (Message 35507) Posted 25 Jan 2007 by MikeMarsUK Post: Inquirer's good, but they don't have a magic mirror either! :-) Powers of two is a good bet, the main factor is surface area of a core versus how much silicon it's economically practical to put into a single package. So each time the process shrinks, you'll get another maximum number of cores per package. Considering x86 cores only - 4 in 65nm (now), 8 in 45nm (volume in 2008), possibly 12 or 16 in 32nm (volume in 2010??), ... etc etc. Sun has already managed to get a huge number of cores onto a single chip, but it's not x86.
16) Message boards : Number crunching : Best CPU for a silent cruncher? (Message 35494) Posted 25 Jan 2007 by MikeMarsUK Post: Core Duo are also a lot hotter than Core 2 Duo?
17) Message boards : Number crunching : 8-core cruncher? (Message 35449) Posted 24 Jan 2007 by MikeMarsUK Post: How do you get these things to overclock? Something I saw on the web seemed to imply you need a 5000x chipset (workstation), but it wasn't very clear. Bios support? PS I looked at the UK prices for a system with 4 x 1GB FB-Dimms, and 5320 (couldn't find 5310), and it was much more expensive, nearly double the price at current exchange rates.
18) Message boards : Number crunching : Best CPU for a silent cruncher? (Message 35414) Posted 23 Jan 2007 by MikeMarsUK Post: I have a Golden Orb II heatsink, and I can't hear it. Watercooling would be even quieter however. Core2 is a cool chip compared to P4 (and arguably AMD, but beware they're hard to compare).
19) Message boards : Number crunching : Your overall stats are gonna be messed-up today (Message 35335) Posted 22 Jan 2007 by MikeMarsUK Post: That was the plan (which has been talked about publically for months), but it's been reversed. The next XML export will exclude the BBC results again.
20) Message boards : Number crunching : who will be the next king of the hill? (Message 35113) Posted 20 Jan 2007 by MikeMarsUK Post: If your constraint is 'most crunch for $" rather than "most crunch per box", you'd be better off with several overclocked quad core2s boxes instead for the same $4,500 ... i.e., like Who?'s quad system rather than his octo system. PS we're experimenting with an SSE2 version of the climate model using Intel Fortran 9.1, will be a month or two until we know whether it gives the same climate physics as the non-SSE2. The previous time SSE2 was tried it gave different results, but looking better this time round.

Next 20