Posts by Venkatesh Srinivas

1) Message boards : climateprediction.net Science : Anyone know if there have been any recent (2019 or 2020) publications helped by CPDN? (Message 62546) Posted 30 May 2020 by Venkatesh Srinivas Post: The most recent on https://www.climateprediction.net/publications/ are from April 2018. Thanks!
2) Questions and Answers : Unix/Linux : Trying to get tasks to not crash Linux client, now not receiving tasks (Message 57251) Posted 29 Oct 2017 by Venkatesh Srinivas Post: Any news on Linux/Mac tasks? (Do we have any idea how much compute capacity is idled by the lack of Linux/Mac workers? Hopefully not that much? Though the backlog of tasks seems pretty high now...)
3) Message boards : climateprediction.net Science : Publications - 2017 (Message 57093) Posted 11 Oct 2017 by Venkatesh Srinivas Post: Came across two 2017 publications that used work run on climateprediction.net -- "Western US high June 2015 temperatures and their relation to global warming and soil moisture" - only have access to abstract, but even it notes large ensemble of w@h models was used to understand 2015 pacific northwest. "weather@home 2: validation of an improved global–regional climate modelling system" Guillod et al. (Geoscientific Model Development). Provides background on why w@h2 was created and a summary of how it evolved from HadCM3. Also discusses use of volunteer compute & notes the increasing capability of boinc computers allowed for higher-resolution models and correcting for known biases in the Had* models.
4) Questions and Answers : Unix/Linux : Trying to get tasks to not crash Linux client, now not receiving tasks (Message 57062) Posted 6 Oct 2017 by Venkatesh Srinivas Post: Any news on Linux/Mac tasks?
5) Message boards : Number crunching : New work Discussion (Message 57011) Posted 1 Oct 2017 by Venkatesh Srinivas Post: Of some interest - the owner of batch 664-666 (the PNW FMEC project), Linnia Hawkins, talked about climate modeling on the August 25th episode of Earth Report Radio. (https://itunes.apple.com/us/podcast/earth-repair-radio/id1259504838) climateprediction.net didn't come up, but discussion of downscaling models from 50 km -> 2/5 km did.
6) Message boards : Number crunching : New work Discussion (Message 56996) Posted 29 Sep 2017 by Venkatesh Srinivas Post: Is the new PNW 25m work available for Mac or Linux?
7) Message boards : climateprediction.net Science : CPDN helped w/ "Emission budgets and pathways consistent with limiting warming to 1.5 °C" ? (Message 56876) Posted 19 Sep 2017 by Venkatesh Srinivas Post: Hi, Recently a news site highlighted Prof Myles Allen's paper "Emission budgets and pathways consistent with limiting warming to 1.5 °C"; was any of the modeling work for this paper done via CPDN? IIUC Allen was one of the founders of CPDN... (Or more broadly, have there been any 2017 publications based on cpdn work?) Thanks!
8) Message boards : Number crunching : New work Discussion (Message 56838) Posted 13 Sep 2017 by Venkatesh Srinivas Post: Thanks. For the FMEC project, the link on the climateprediction.net site to Oregon State's page seems wrong - the page links to http://terraweb.forestry.oregonstate.edu/FMEC.htm, the lab + project page seems to be http://terraweb.forestry.oregonstate.edu/forest-mortality-economics-and-climate-change-western-north-america now?
9) Message boards : Number crunching : WU energy vs value? (Message 56744) Posted 29 Aug 2017 by Venkatesh Srinivas Post: Sorry, I asked my question poorly earlier - I wasn't asking if CPDN is worth running overall. I was trying to ask if CPDN is worth running at a particular FLOPS/W (when a particular WU runs for a 1 million seconds at +30 W over baseline). A poster above noted that his workstation was able to complete a similar WU in 300,000 seconds at +12 W, nearly an order of magnitude fewer watt-seconds. (Do all pnw25 WUs have comparable amounts of work?) If we have some kind of idea of the value of one WU, we can decide between running CPDN locally or (say) renting time on more power efficient computers. Even if we don't have a dollar-based value function, an idea of how many trajectories go into a given project and what avg performance of computers delivering results would help with those kinds of decisions.
10) Message boards : Number crunching : WU energy vs value? (Message 56730) Posted 27 Aug 2017 by Venkatesh Srinivas Post: Anyone have any thoughts as to when (energy-wise and emissions-wise) it is worth running a CPDN workunit? In June my computer ran a wah2_pnw25_c25q_200312_49_588 task, taking 1,077,019 CPU-seconds. Dynamic power of one core working on a CPDN WU is 30 W above idle. This is ~30,000,000 watt-seconds => ~9 kWh. Where I live average lb CO2e/mWh is 1,498, so this workunit generated ~13 lb CO2e. At this rate, is it worth running CPDN WUs here? Or should we prefer to only run them where we are more power-efficient (FLOPS/W) or cleaner (lb CO2e/mWh)?
11) Message boards : climateprediction.net Science : CPDN helped with "FORESTS AND EMISSIONS" (2008)? (Message 56601) Posted 30 Jul 2017 by Venkatesh Srinivas Post: Were any of the simulations for the 2008 Met report "FORESTS AND EMISSIONS: A CONTRIBUTION TO THE ELIASCH REVIEW" done with CPDN?
12) Message boards : Number crunching : Replanca Error/Sigseg fault. (Message 56390) Posted 16 Jun 2017 by Venkatesh Srinivas Post: How can you tell what batch a task is from?
13) Message boards : Number crunching : AVX and AVX2; Is it used at CPDN? (Message 56381) Posted 14 Jun 2017 by Venkatesh Srinivas Post: It looked like a single-precision powf() was used; SSE2 at least can match precision trivially. I suspect that no one has optimized 32-bit libm for recent processors. 64-bit libm uses MULSS (SSE) on my system instead of x87.
14) Message boards : Number crunching : AVX and AVX2; Is it used at CPDN? (Message 56320) Posted 1 Jun 2017 by Venkatesh Srinivas Post: I profiled the instruction mix of the wah2rm3m2t_um_8.25_i686-pc-linux-gnu model on a platform with SSE* and AVX. As far as I can tell, it uses a mix of x87 and SSE instructions only. Substantial time (~5%) is spent in libm's powf(), which uses legacy x87 instructions. Is there some other way the model could do exponentiation? Ditto for log10. (Both are at FP32 precision afaict). Modern CPUs would prefer (in energy per FLOP) different instructions.