Posts by MikeMarsUK

41) Message boards : Number crunching : Claimed -vs- Granted Credit (Message 33393) Posted 25 Dec 2006 by MikeMarsUK Post: It started off life as a 3800, but I upgraded it to an X2 in August when the prices dropped (HostID stayed the same). o/c was about the same each time (2553, 2540), and everything else was the same too (motherboard, memory, etc). The penalty for running two cores rather than one seems to be about 10% (so it's the same as 1.9 3800s. On CPDN/Coupled I was getting 1.7s/ts on the 3800, but 2 x 1.8s/ts on the X2, lower s/ts is better). If you look at my Synergy RAC, that's from two PCs, the X2 and a nonoverclocked Dell 820D. The claimed credit is actually about the same or a little lower as the granted credit on workunits of 3h or more - remember that the benchmarks won't reflect the memory. 1h workunits seemed to be inefficient when I tried that, and the claimed credit was higher than the granted credit. Different Rosetta WUs do give different credit, I think depending on the size of the protein and which algorithms are in use. I was running one core on CPDN:SAP and the other on Rosetta, so I don't know what it would have scored if both were on Rosetta, but the impression I get is that CPDN:SAP is harder on memory bandwidth than Rosetta (i.e., two CPDNs run a little slower than one CPDN:SAP and one Rosetta). CPDN/Coupled model seems to be in the middle - not such a big hit on memory as SAP.
42) Message boards : Number crunching : Claimed -vs- Granted Credit (Message 33382) Posted 24 Dec 2006 by MikeMarsUK Post: I have a similar AMD, gets around 13-15c/h per core. The memory was the expensive part of the setup, rather than the CPU. The clock is similar (2540), but the memory is DDR500 @ 3,3,2,7,1T.
43) Message boards : Number crunching : Claimed -vs- Granted Credit (Message 33374) Posted 24 Dec 2006 by MikeMarsUK Post: The 'Old Mac' would be a PPC machine (Power PC), and the Rosetta code doesn't seem to run very fast on them. This is why it's not getting that much credit compared to the other machines. In addition to that, the Benchmark only measures part of the system performance. So a newer PC with very fast memory and low memory latency will not necessarily get much higher benchmarks, but will run the Rosetta code very well (and hence get more credit).
44) Message boards : Number crunching : ONE ACCOUNT TO RULE THEM ALL.............. (Message 33323) Posted 24 Dec 2006 by MikeMarsUK Post: One question (I already mentioned this under your other thread on ClimatePrediction). Did you download Boinc each time into different locations? (i.e., do you have a climate boinc and a rosetta boinc?) If so, you can attach to both projects via a single manager. The easiest way to do this is use the Attach wizard in version 5.*, and enter your existing account details. You'll need to amalgamate the two projects into one manager before the CPID synchronisation mentioned in the prior posts will work. Are you still having trouble with ClimatePrediction clashing with your PC, or is it working now? Did you ever try Prime95's torture test to help isolate the problem? (If it passes 24h then you know your hardware is fine). Once all the CPIDs are synchronised between the projects, you can pick up a signature from one of the stats sites.
45) Message boards : Number crunching : My PC RAC > 3000 (Message 33272) Posted 23 Dec 2006 by MikeMarsUK Post: Is Clovertown desktop, workstation, or server, or several of these? I'm considering getting a 2 x 4-core box near the end of next year (can't afford it yet), and as far as I can see it the various choices are: * 2 x Clovertown (or Peryn? Yorkfield?) CPUs on a dual socket motherboard * 2 x Opteron 2xxx on a dual socket motherboard once they've got a 4-core offering (Barcelona) * 2 x FX-76 on a 4x4+ motherboard (not 4x4)
46) Message boards : Number crunching : My PC RAC > 3000 (Message 33268) Posted 23 Dec 2006 by MikeMarsUK Post: I didn't say a word about desktop computers, or the 4x4 platform, why are you implying I did?
47) Message boards : Number crunching : My PC RAC > 3000 (Message 33244) Posted 23 Dec 2006 by MikeMarsUK Post: Intel's version of hypertransport is CSI (Common System Interface), delayed from it's original release circa 2007, but now officially announced at the Intel Developer Forum. Now due for release 2008/2009. http://www.theregister.co.uk/2006/09/21/intel_open_chips/ CSI is intended to destroy AMD's dominance in the 4x to 8x server sector, which AMD gained from the advantages hypertransport gives to this class of server.
48) Message boards : Number crunching : My PC RAC > 3000 (Message 33233) Posted 23 Dec 2006 by MikeMarsUK Post: Mostly depends on the memory bandwidth requirements. If you have an application which can run mostly in the cache, and there isn't much core:core communication, hypertransport will have little or no effect, as you're seeing here. Other applications will behave differently, for example, CPDN:SAP is a boinc project with extremely high memory bandwidth + latency requirements, and would probably behave differently. I note that Intel is planning to support hypertransport-like functionality in the future, I doubt they'd do that if it was just a waste of silicon.
49) Message boards : Number crunching : Dual Core (Message 31835) Posted 29 Nov 2006 by MikeMarsUK Post: If you run a long-duration project on one core (say, CPDN), set 'no more work' against that project, and pretty much it'll be one project for one core, and the other for the second.
50) Message boards : Number crunching : problem with core 2 duo not running both cores (Message 31205) Posted 15 Nov 2006 by MikeMarsUK Post: Once you get an overcommitted message, Boinc will stop downloading new workunits (regardless if that leaves the second core idle). If you increase the overcommitted project's resource share so that Boinc thinks the result will complete in time, then 'overcommitted' should go away and it'll then download new work.
51) Message boards : Number crunching : Upgrade complete, sort of (Message 25698) Posted 31 Aug 2006 by MikeMarsUK Post: to run two instances of prime95 you have to install it a 2nd time to a different dir iirc :) You can do it via the command-line interface (I have two shortcuts set up, one for each core). Make sure that (unless you have 3 or 4GB ram) you run the smaller memory size test for one or both of the instances, otherwise it'll just be thrashing the hard disk. Thermal throttling can also cause everything to slow down (starts at about 65c though).
52) Message boards : Number crunching : Upgrade complete, sort of (Message 25473) Posted 29 Aug 2006 by MikeMarsUK Post: Have you tried running some other things to try and max out both cpu's. . like that pi calculating program, or another non-boinc dc project? That might help narrow down where the problem is (Rosetta, Boinc, or anything that wants to use 100% of both cores). Prime95's Torture test is very good - http://www.mersenne.org. Use the 'affinity' setting to run it on particular cores, and then run two copies to test both cores simultaneously.
53) Message boards : Number crunching : Thread Delete: Why I am pulling my machines out. (Message 23589) Posted 19 Aug 2006 by MikeMarsUK Post: ... One feature that will not be added is allowing a moderator to edit a post (not my decision, but it is probably a good one). ... Or added back in? I used to be able to edit posts on the BBC/CCE boinc forum, unless that was a custom modification to the code.
54) Message boards : Number crunching : New Crediting system: questions (Message 23555) Posted 19 Aug 2006 by MikeMarsUK Post: RAC is currently based on boinc credit, a second RAC column may appear which is based on work credit. Teraflops will be modified in both directions by the benchmark results - Linux machines will be underreporting for example. If tereflops were calculated from the work credit it would probably return a similar result since it's basically an average of the boinc credit.
55) Message boards : Number crunching : New Crediting system: questions (Message 23522) Posted 19 Aug 2006 by MikeMarsUK Post: It should average out over time - so you get roughly the same credits/hour regardless of whether you pick short or long WUs.
56) Message boards : Number crunching : Removing credits backdated to february. (Message 23520) Posted 19 Aug 2006 by MikeMarsUK Post: It's a reasonable decision which I think everybody can accept, so let's go forwards with it rather than feeling bitter about the past (from whichever side of the fence people come).
57) Message boards : Number crunching : Thread Delete: Why I am pulling my machines out. (Message 23499) Posted 19 Aug 2006 by MikeMarsUK Post: actually mike with php, there are some boards that require you to activate the account thru a link sent to the email address you used to register. that stops a lot of spammers. ... The CPDN PHP forum uses a combination of image verification and email activation. We get about 2-3 spammers a day who go through both phases, and around 10 who only do the initial image verification phase. Without image verification it'd probabably be several dozen a day. The spammers who don't email verify are primarily building google link farms (their account page including website is visible even if not email authenticated, but they can't post). Note that email activation can be automated (image verification can be too, but it's extremely hard). An example of a URL advertised via a forum registration link farm is here: http://www.google.co.uk/search?hl=en&q=%22www.gaa.com%22+interests+location+%22posts+per+day%22&btnG=Search&meta= Spam accounts are usually deleted within an hour or so by the site admins, moderators can't do this. There are many 'questionable' accounts which aren't clear whether they're spammers or legitimate, which we have to leave intact. If there were an additional verification stage where you have to link a climate project together with a PHP forum account, then it would reduce the 10 spammers down to very few indeed (I don't recall any spammers turning up on the boinc forums). As mentioned earlier, PHP forums are targetted by hackers, unlike other forums.
58) Message boards : Number crunching : Removing credits backdated to february. (Message 23483) Posted 19 Aug 2006 by MikeMarsUK Post: Ok. . thread's on notice (insert Colbert photo here). The thread is off topic, I don't want to close it. If you have the ability to lock posts in the same way David's log is locked, then perhaps David could make a statement about whether the project is backdating work credits or not (presumably not), and then lock everything down. That'll answer the original poster's question. (No need to keep this post intact of course since it's a metapost, so feel free to delete). -- Edit: No threads can be locked, you can reply to DB's Journal. . posts that aren't from him just get moved to the comment thread. Ah OK, that explains how the log works, I'd wondered. I won't add any more to this thread, there's more than enough posts already! :-)
59) Message boards : Number crunching : Why are discussions about Rosetta taking place on other boards? (Message 23467) Posted 19 Aug 2006 by MikeMarsUK Post: Thanks David
60) Message boards : Number crunching : Thread Delete: Why I am pulling my machines out. (Message 23408) Posted 19 Aug 2006 by MikeMarsUK Post: That is another bug in the Boinc forum code. If a moderator deletes a post, the reason code always shows 'obscene' regardless of which reason code the moderator actually picked.

Previous 20 · Next 20