convore.json/groups/pycon-2011/solvcon-a-new-python-based-software-framework-for-massively-parallelized-numerical-simulations-yung-yu-chen/messages.json

[{"user_id": 10411, "stars": [], "topic_id": 12600, "date_created": 1299946627.546381, "message": "I wish we had that awesome spinning bar from the superboard talk yesterday.", "group_id": 373, "id": 333128}, {"user_id": 10411, "stars": [], "topic_id": 12600, "date_created": 1299946912.000293, "message": "Typical scaling issues with unstructured meshes...", "group_id": 373, "id": 333176}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299946366.0546451, "message": "me too", "group_id": 373, "id": 333052}, {"user_id": 8649, "stars": [], "topic_id": 12600, "date_created": 1299947667.554554, "message": "what is diff between \"multi-core\" and \"many-core\"? :)", "group_id": 373, "id": 333315}, {"user_id": 20975, "stars": [], "topic_id": 12600, "date_created": 1299947795.9494481, "message": "it would likely be better to use compression rather than skipping timeslices", "group_id": 373, "id": 333346}, {"user_id": 10411, "stars": [], "topic_id": 12600, "date_created": 1299946352.0178101, "message": "Waiting...", "group_id": 373, "id": 333046}, {"user_id": 8649, "stars": [{"date_created": 1299947643.2943909, "user_id": 10411}], "topic_id": 12600, "date_created": 1299947614.5022891, "message": "Or use Mayavi/TVTK to throw numpy arrays at VTK", "group_id": 373, "id": 333302}, {"user_id": 10411, "stars": [], "topic_id": 12600, "date_created": 1299947302.619153, "message": "Maybe I missed something here.  If you are using finite element model, why put it in integral form.  Wouldn't monte carlo work better?", "group_id": 373, "id": 333245}, {"user_id": 10411, "stars": [], "topic_id": 12600, "date_created": 1299947484.4440761, "message": "I think the resolution that you want is based on what problem you are trying to solve.", "group_id": 373, "id": 333271}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299947793.998771, "message": "maybe its an arbitrary distinction then", "group_id": 373, "id": 333344}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299947919.7382829, "message": "well, a core (multiprocessor) will generally run one warp at a time (or 2 half warps) using CUDA terminology", "group_id": 373, "id": 333376}, {"user_id": 10411, "stars": [], "topic_id": 12600, "date_created": 1299947569.7944601, "message": "Wouldn't a binary format be smaller than VTK XML?", "group_id": 373, "id": 333288}, {"user_id": 10411, "stars": [], "topic_id": 12600, "date_created": 1299947685.8667541, "message": "@pwang spelling? ;)", "group_id": 373, "id": 333318}, {"user_id": 20406, "stars": [], "topic_id": 12600, "date_created": 1299947358.9560881, "message": "I have no idea on what he's talking about, but the figures are surely pretty", "group_id": 373, "id": 333251}, {"user_id": 8649, "stars": [], "topic_id": 12600, "date_created": 1299947629.0815229, "message": "http://code.enthought.com/projects/mayavi/", "group_id": 373, "id": 333305}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299947689.0635519, "message": "i believe multi-core is on one machine, many-core is across many machines", "group_id": 373, "id": 333322}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299947766.606204, "message": "ah, i can't really read the slides from back here", "group_id": 373, "id": 333337}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299947818.2887061, "message": "GPUs have ~15 cores", "group_id": 373, "id": 333351}, {"user_id": 8649, "stars": [], "topic_id": 12600, "date_created": 1299947740.879447, "message": "but he indicated GPU for \"many-core\".. what if I stick a couple of nvidia tesla cards in SLI config in the same box? :)", "group_id": 373, "id": 333334}, {"user_id": 8649, "stars": [], "topic_id": 12600, "date_created": 1299947786.097935, "message": "i can sympathize - the font is not great", "group_id": 373, "id": 333341}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299947942.6550231, "message": "how many threads that entails depends on the card", "group_id": 373, "id": 333384}, {"user_id": 8649, "stars": [], "topic_id": 12600, "date_created": 1299947794.048481, "message": "he needs to bump up the courier by 2 pts or make it bold", "group_id": 373, "id": 333345}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299947827.0111549, "message": "i7 has 8?", "group_id": 373, "id": 333352}, {"user_id": 8649, "stars": [], "topic_id": 12600, "date_created": 1299947879.374326, "message": "i don't even know the gpu manufacturer terminology anymore... how many warps in a core?", "group_id": 373, "id": 333365}, {"user_id": 20705, "stars": [], "topic_id": 12600, "date_created": 1299947852.83233, "message": "er GPU has ~15 multiprocessors, each one can run 100s of threads", "group_id": 373, "id": 333360}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300112235.7764809, "message": "@scopatz @pwang Just as a record.  I have written my own writers for legacy sequential VTK format, XML sequential VTK format, and XML parallel VTK format.  Only XML formats are documented to have parallel I/O support.  Also, VTK XML format support compressing (but only gz IIRC).  But if you have a fast parallel file system, compressing actually could result in more time in I/O.", "group_id": 373, "id": 347538}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300112009.940439, "message": "@scopatz I am not using finite-element model.  The integral form is necessary for capturing extremely strong nonlinear behavior, e.g., shock waves, expansion waves, contact discontinuities, etc.  We are focusing on generic, linear/nonlinear, hyperbolic PDEs because too many complex and challenging transient phenomena are governed by them.  I actually am \"thinking\" about FEM for elliptic type equations, but it cannot be materialized soon (say, in a couple of years).  MC works well for steady state solutions, but not very good for transient problems.", "group_id": 373, "id": 347510}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300111597.026248, "message": "Ah I finally got a chance to the thread.", "group_id": 373, "id": 347471}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300112528.055948, "message": "@pwang \"Many-core\" is how GPGPU guys (aka NVIDIA :p) documents their technology in literatures.  I guess they chose the term because they want (i) to show GPU is good and (ii) to distinguish GPU from CPU.  I first heard this term two years ago in a workshop given by Wen-mei Hwu http://impact.crhc.illinois.edu/people/current/hwu.php .  Basically many-core means the architecture of using many many stream processors in parallel.", "group_id": 373, "id": 347568}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300112785.5386691, "message": "For those who wants to know more about GPGPU computing, a book by Hwu could be a good starting point http://impact.crhc.illinois.edu/people/current/hwu/hwu_pub.php", "group_id": 373, "id": 347595}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300112853.6733711, "message": "But this field changes so quickly.  I would like to raise a warning that the content could be not so up-to-date.", "group_id": 373, "id": 347601}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300112710.7040529, "message": "@enjalot @pwang Thanks for the comments on fonts.  That size works very good in a common academic conference in which we have a much smaller room.  Next time I will make it larger for a much larger room.", "group_id": 373, "id": 347590}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300112918.182802, "message": "Thank you guys for many many great comments :)", "group_id": 373, "id": 347614}, {"user_id": 7688, "stars": [], "topic_id": 12600, "date_created": 1300113287.5681801, "message": "@camarajm Ooops I missed this one.  Compression is important (and has been implemented in SOLVCON), but it is not likely to be able to reduce data size by orders.  I can reduce the data by several times, but that's the best it can do.  Compression is not the top kill.", "group_id": 373, "id": 347667}, {"user_id": 10411, "stars": [], "topic_id": 12600, "date_created": 1300119523.4983981, "message": "@yungyuc Thanks for the clarification here.  I agree that MC does not work particularly well for transient problems.", "group_id": 373, "id": 348817}]