21) Message boards : News : OneChunk CPU Pre-Announcement (X86_64 & ARM64) (Message 568) Posted 5 May 2021 by Hy Post: Small update, we've decided roughly on the input work and are slowly making pieces of the BOINC template files, and task estimates. Keep in mind, here we are using "BOINC Benchmark Whetstone MIPS = OPS" These break down to: 5,800,000,000 OP/sec per Ryzen 5800X Core Whetstone (what credit is based off of) 12.5 seconds per seed/input line (ie, 10 seeds computed with our current program takes 120-125 seconds on a single 5800X 4.85ghz core) So, 72,500,000,000 "OPS per seed input line per core" on Zen 3 Thus: 21,750,000,000,000 OPS (2.175e^13) (300 lines of input seeds) of computation done per core, per BOINC task. If everything works as expected, then CPU task time estimates should be close to reality, being 60-65mins on a Ryzen 5000 CPU, and say, 90-120mins on a Ryzen 1000 CPU due to IPC and frequency differences. Also for fun stats, a 3.2GHz Broadwell laptop core is 2.2X slower (271 seconds for 10 seeds) versus my 5800X

22) Questions and Answers : Windows : any way to contribute with only cpu? (Message 566) Posted 4 May 2021 by Hy Post: We've listened to your requests and we have something in the works! see: https://minecraftathome.com/minecrafthome/forum_thread.php?id=102 "OneChunk" for more (Bump)

23) Message boards : News : OneChunk CPU Pre-Announcement (X86_64 & ARM64) (Message 565) Posted 4 May 2021 by Hy Post: On Windows 10 (for example), you can use Raspberry Pi Image Writer to write the posted Raspbian 64bit OS to a fresh SD card, and then set up your system as usual (plus connect it to the internet). Make sure you have some sort of cooling on your Pi 4 though, like a heatsink or fan, especially if you are overclocking. Then, you can run " sudo apt-get update && sudo apt-get install boinc boinc-client boinc-manager " to install BOINC, and then you can launch it through either the terminal (boincmgr) or the menu. It is advised though that you have a screen attached to use BOINC via the GUI, but more experienced BOINC/Linux users will be able to use: https://minecraftathome.com/minecrafthome/weak_auth.php boinc --daemon --attach_project "minecraftathome.com/minecrafthome/" "weak_key_here" boinccmd --project minecraftathome.com/minecrafthome/ update boinccmd --get_tasks boinccmd --network_available to connect their screen-free (like SSH) pi to begin crunching when we release the application :) (Also, code updates can be tracked here: https://github.com/minecrafthome/OneChunk but we have like a dozen things on our checklist to do before its running)

24) Message boards : News : OneChunk CPU Pre-Announcement (X86_64 & ARM64) (Message 559) Posted 3 May 2021 by Hy Post: Over on the Minecraft@Home discord, Boysanic ("Boincboi"), Matt (Stronghold simulation coding), other developers and I have been discussing and planning a rollout of our first CPU application. This OneChunk project is "a BOINC project in the works. Our goal is to find all world seeds w/ at least an 11 eye portal in MC Java 1.7.10". So, if this all works out, "you'll finally be able to say you've found an 11 eye portal by yourself!" Essentially, we are combining parts of the "SSG" (Seed-Speedrun) and 11-eye portal seed-filtering programs together to go through a large dataset. This input file is 9GB in size, but we plan to split it up into many small manageable tasks to be distributed through BOINC, which is where you come in! Going through a 320kb sample input takes a 4.8GHz Ryzen 5800 (the fastest single-core CPU) around 2 minutes, simulating and checking over 5,220 worldseeds a second. A 2GHz Raspberry Pi 4, with 64bit Raspbian manages to compute 1,370 worldseeds a second, making it around 3.8x slower but much more efficient. (45W vs 5W) Thanks to work by other developers, I have tested and confirmed that both x86_64 ("64bit Windows and Linux") and ARM64 (64bit Linux ARMv8) configurations will work. The OneChunk 'Filter9000' BOINC application is single-threaded, but will likely have multiple tasks running concurrently on your computer to best utilize its resources and scale to your unique hardware configurations. For 'PC' requirements, we suggest Ubuntu 18.04/20.04 64bit or Windows 7 / 10 64bit, running on an CPU such as an AMD FX or Intel 1st Gen Core i5/i7 CPU or newer. If anyone has a Raspberry Pi 3(B/B+) or 4, we would like to invite you to install the latest 64bit Raspbian on a spare SDcard in preparation. We also expect Nvidia Jetson Nano/TX1/Xavier devices with Ubuntu 18.04 64bit or newer to successfully run our Arm64 tasks, once we configure our BOINC server to provide and schedule tasks for this platform. Please follow instructions on the Raspberry Pi website, but we suggest that more experienced Pi users join us for this experiment! OS image below: https://downloads.raspberrypi.org/raspios_arm64/images/raspios_arm64-2021-04-09/2021-03-04-raspios-buster-arm64.zip Also, check out our Forums for upcoming updates, and to see the first 25-Tall Superflat cactus!

25) Message boards : Number crunching : Kaktwoos 2.13 announcement / testing (Message 558) Posted 2 May 2021 by Hy Post: Hello BOINCers! As we've received reports from a few concerned users over the past week regarding issues with kaktwoos-cl, I've put together a new update with some 'finalizations' of some existing features. I'd consider this to be Kaktwoos-cl "2.15" rather than 2.13, but to make things simple we just bumped that version up. First though, if you are on an Nvidia GPU and under Linux, please make sure you have installed your "nvidia-opencl-dev" package, or whichever package includes the Nvidia OpenCL headers, please! Your package manager should supply this on request (apt-get install), plus the latest or most compatible/stable proprietary driver which you likely already have. The majority of the OpenCL errors we have received, and failed tasks are due to users not checking their Tasks once started (missing this error), and not ensuring the OpenCL headers are installed prior to running BOINC. Now, this update shall introduce a fallback (or well, swapped) kernel for Nvidia GPUs. On detection of a RTX or GTX 16XX series GPU, the existing optimizations brought to Kaktwoos 2.11 will be enabled. For any other older Nvidia GPU, these 'optimizations' will be disabled, and the old but stable kernel will be used, bringing 3-5% back those afflicted with reported regressions or on weaker cards. Internally, the AMD GPU detection for optimizing is more generic, and if it detects it is being run on an RDNA 1 or 2 GPU (RX 5000/RX6000) then it will fall back to the generic kernel. The majority of AMD GPUs will use the new AMD kernel as before, but this just ensures that future architectural changes and new product releases are not missed by Kaktwoos-cl The other changes to this code are mostly on the "C" plaintext side, with Neil's reformatting of the code to hopefully make the varying styles of coding over the past year to mesh together better visually. I'm still considering a further safety check to improve the checkpointing system, for users who have an unrecoverable host crash and have a corrupted checkpoint (we say only "0.5%" of users ever experience this, which causes the "infinite kaktwoos" bug). Also, to aid in debugging or user interest, we will now print the seed search range for your current task into your log. And yes, there will be somebody who has a task from 000000000 to XXXXXXXX! Anyways, soon we plan to begin a new (possibly short-term) project involving making another "impossible" possible, which involves finding many "11-eye" dungeons, and any/all in a Single Chunk. Some of you on the Discord server may remember this from December/January, but some recent interest has lead to a boinc-able set of computations and a rather large dataset to work through! This shall be our first official CPU application, if all goes well. Imagine what could happen if you got a village, emerald veins, and an 11-Eye End Portal all in the same chunk...

26) Message boards : Number crunching : New kaktwoos 2.12 beta application has issues (Message 550) Posted 29 Apr 2021 by Hy Post: From what I saw, your computers were working on Kaktwoos-cl 2.12 tasks yesterday without errors. Our admin, Neil saw and mentioned to us some sort of error while making uploads/updates to the BOINC backend, hence moving 2.12 to beta and reverting to 2.11 to continue work. We have made him aware of the issue, and when he is available should ensure work is sent out to clients who had swapped between versions and/or are having issues. I haven't personally seen this new error myself, and if you can post an example of a task or computer that failed with 2.12 but worked well previously, I'd appreciate it. Sometimes tasks are cancelled by the server which may appear as an error, but resolve themselves quickly. Other errors persist until a reboot, despite configuration changes being submitted to our servers

27) Message boards : Number crunching : Kaktwoos tasks always stall out and never finish if interrupted (Message 549) Posted 29 Apr 2021 by Hy Post: In terms of 'overriding' the default checkpoint interval, that is set to 30 seconds (BOINC-side), but requesting a checkpoint by suspension and automatic checkpoints are a part of Kaktwoos-cl. Basically, whenever you see your progress bar tick forwards, a checkpoint has been made. I've honestly never seen any request or feedback, or usage of the custom *timer* for minimum checkpointing mind you Coding wise, this is because I have set it to checkpoint (regardless of if it is run by BOINC or not) every 200 million seeds checked. Most GPUs are 25-50 mseed/s, meaning most GPUs checkpoint every 8-4 seconds per task, though this supposedly can be given an override by the BOINC checkpoint timer function and the manual checkpointing request sent to Kaktwoos-cl. The BOINC documentation didn't appear to provide a standardized checkpoint format, just the API to know when to checkpoint which we use as another reference for when to checkpoint, say if your GPU is very slow (like Intel graphics or AMD APUs) and takes more than 30 seconds to finish 200 million seeds.

28) Message boards : Number crunching : Kaktwoos tasks always stall out and never finish if interrupted (Message 543) Posted 26 Apr 2021 by Hy Post: Additionally, we've found one of the situations which triggers this issue, potentially related to the checkpointing system (or how BOINC resumes from a major fault). Basically, one of our developers shut down his system hard while a task was nearly complete, and it appears something happened with our checkpoint file which caused it to become corrupted or incorrectly saved. The issue is, because checkpointing is so fast it's almost impossible to have this occur with the timing, or situation required to cause the infinite-kaktwoos problem. For reference, our normal checkpoint files are around 12kb in size, while this one was only 1kb. But, even if you kill Kaktwoos in various ways (as we tested before) this issue isn't reliably caused. We might make an alternative fix such as a basic verifier for the checkpoints that are being reloaded, but generally this issue is only due to a computer being either shut down before the task is properly suspended, or a random GPU crash causing kaktwoos-cl to freeze while it is writing a checkpoint *early on* and thus corrupting the most important data and leading to it checking unknown parameters.

29) Message boards : Number crunching : Raspberry Pi VC4CL (Message 541) Posted 25 Apr 2021 by Hy Post: Sadly, Raspberry Pis are a bit too slow on the GPU side even with OpenCL exposed. Another issue is well, a BOINC client for the Raspberry Pi, and having an armv8 or aarch64 build for these Pis. I did some work making the first Super Mario 64 Port run on the Raspberry Pi, and I've got some experience with Pis over the years that make me say this isn't too possible. I have also run Kaktwoos-cl on the Nintendo Switch with a CUDA-> OpenCL translation layer and let me just say it is extremely slow, and the Switch (overclocked even) is miles ahead of any raspberry Pi in performance. That being said, for any other projects, even CPU based, there might be a chance to put the newer Pis to some use. Though, I given a 15watt power draw, even a weak 2 or 4 core laptop within the past 3 years will be faster.

30) Message boards : Number crunching : Kaktwoos tasks always stall out and never finish if interrupted (Message 540) Posted 25 Apr 2021 by Hy Post: Hello, we've pushed the code to our Github and are verifying it. We've had this issue for some time, but now that we're seeing it more often due to more users, we've decided to take some steps to solve it. This should be pushed out to BOINC for all clients to update to in a day or two, I trust Neil's work and the changes don't seem to be large

31) Message boards : News : 23 tall blocks cactus found by PDW and edjotron (Message 539) Posted 25 Apr 2021 by Hy Post: Update! We've found a 25 tall Superflat Cactus, which I believe means we broke our (Superflat) record after over 6 months of waiting! To generate, please open Minecraft Java 1.14.4, choose Superflat preset -> Desert (14*sand instead of 8*sand) and input seeds: 233775207740551 at -27413824 69 -7724480 155193555785287 at -5704928 69 -24000160 Thanks to Vanos0512 (UID:869) and Greger (UID:12272) for calculating this record-breaker.

32) Message boards : News : 23 tall blocks cactus found by PDW and edjotron (Message 527) Posted 4 Mar 2021 by Hy Post: Just a quick announcement: By this weekend, we shall have all Y=65 tasks finished, and Y=66 tasks (100-150K) launched. Sadly, there are *no new tall cacti* to report. No 24s, not even for Superflat or preliminary verification

33) Questions and Answers : Windows : Work units continue running long after 100% (Message 526) Posted 4 Mar 2021 by Hy Post: Yes, please do! For some unknown reason, some workunits will randomly not listen to their internal "seedspace" limits, meaning they'll search for cacti in a much wider area then they should. If you have any tasks running 30-50% longer, and/or stuck at "100%" for a long time please abort, and then allow other tasks. We suspect it may be due to either memory errors, various sync (CPU/GPU) issues, or just bad luck. Out of thousands, or even tens of thousands of workunits/tasks, only a few encounter this issue making it incredibly hard to resolve.

34) Message boards : Number crunching : Kaktwoos Optimizations are LIVE! (and Y=65 tasks too) (Message 510) Posted 14 Feb 2021 by Hy Post: Thanks for that check. Will def need to see if the server needs to reverify its files for that. I don't understand why all the other updates went through, except for this one. Do you mind trying to find your BOINC data & project folder (Maybe /usr/ or /var/boinc) and copying over those files in my previous post to a path like "boincdata/projects/minecraftathome.com_minecrafthome/(files)"? Then chmod +x kaktwoos.bin and that should get BOINC closer to running

35) Message boards : Number crunching : Kaktwoos Optimizations are LIVE! (and Y=65 tasks too) (Message 508) Posted 14 Feb 2021 by Hy Post: Thanks for that. It seems like on weaker and/or older Nvidia cards that they lose around 5% performance, while others don't lose any. Meanwhile, the RTX 2000 series has a lot to gain from these, so I can look into doing what we did on AMD and switching back to the original kernel for the 750ti/1060. I might have to manually check (sadly) through a bunch of GPU results to see how good the changes are as well, but yeah. v2.10 -> 2.11 RTX 3070 (160 -> 176m/s!) (RX 6800 is doing well at 153m/s, without optimizations) RTX 2080S (103 -> 107!) RTX 2080 (90 -> 108.9!) RTX 2070S (103.4 -> 108) Titan X Pascal (64.8 -> 61) X RX Vega? (45.5 -> 52!) GTX 1650 (45.3 -> 45.7) GTX 1070 (37.5 -> 35.5) X GTX 1050ti (13.2 -> 14.7?) GTX 970 (32 -> 30m/s) X GTX 1060 (27.3 -> 25.8) X RX 580 (21 -> 26.4!) GTX 950 (16.5 -> 15.5) X QDR4000M (14.5 -> 13.3) X (Note that RX 580 is finally as fast as a GTX 1060 in kaktwoos, when both are at their 'best' performance)

36) Message boards : Number crunching : Kaktwoos Optimizations are LIVE! (and Y=65 tasks too) (Message 506) Posted 13 Feb 2021 by Hy Post: Hi, that's... weird. The files seem to match fine on our server, and are downloadable (manually) Can you try to 'reset' MC@Home project in your BOINC client and/or re-add us, then press update and see if the files download? I'm looking at our server, and the files themselves are there, the signatures seem to match up and so do their sizes. Yet, atm, Linux-Intel Kaktwoos isn't downloading properly...despite us not modifying it unlike everything else for this update and optimization <file_info> <name>kaktwoos_2.11_opencl_intel_gpu.bin</name> <url>https://minecraftathome.com/minecrafthome/download/kaktwoos_2.11_opencl_intel_gpu.bin</url> <nbytes>1511656</nbytes> <name>kaktwoos_2.11_opencl_intel_gpu.cl</name> <url>https://minecraftathome.com/minecrafthome/download/kaktwoos_2.11_opencl_intel_gpu.cl</url> <nbytes>4120</nbytes>

37) Message boards : Number crunching : Kaktwoos Optimizations are LIVE! (and Y=65 tasks too) (Message 504) Posted 12 Feb 2021 by Hy Post: Hello, just a quick post to announce optimizations are live for Kaktwoos and so are a new set of tasks, following our news post of the first naturally generated 23-block tall cactus being found! If you see the executable "kaktwoos_2.11" being run on your computer, then the optimized BOINC application is now on your system. It may take a few hours for your systems to be synced, but going forward we hope to collect performance improvements (or regressions) for as many GPU models as possible. These optimizations primarily are for Nvidia and AMD GPUs, aside from the RX 5000/6000 series due to major performance losses from to their architecture being different. Any Kaktwoos-cl compatible AMD card, or APU may see up to 30% performance improvement, and up to 50% in the case of Vega architecture GPUs due to their large caches which were previously underutilized. For some, your computers may use a bit more power than before, but this does not mean they are less efficient at running our new code. Optimizations summary: "kaktwoos-amd.cl" RX 5600/5700/6800/6900 - Optimizations disabled as new kernel is much slower it seems Vega 56 went from 42 -> 63m/s (million seeds a second) (+50%!) RX 480 went from 17 -> 25m/s (+30%!) r5 540M went from 3.9 -> 5.35m/s (+30%!) "kaktwoos-nv.cl" RTX 3090 went from 225 -> 232m/s (+5%?) RTX 2060 went from 52 -> 73/80m/s (+60%) GTX 1650M went from 50 -> 52m/s (+5%) GTX 1080ti went from 61 -> 58.4m/s (-3%) (may just be test kernel and environment) GTX 1070 had no change in perf, 53.3m/s GTX 960 had a slight reduction, 18.2 -> 17.5 m/s (-3%) If you would like to help us out on this front of testing uncommon GPUs, please join our discord and check the #questions_and_help channel after going through the welcoming process and reading our rules! Or, just post in this thread after going through "Your Tasks", either online or using the BOINC Client's advanced view and comparing the completed tasks before Feb. 11/12 to the latest validated ones. In the case of any errors, or worse performance consistently from these changes, please post here, message us so we can either diagnose the issue on our end, or attempt to revert the kernel for that specific card (if performance losses are greater than 5-10%, and it is on a common model!)

38) Message boards : News : 23 tall blocks cactus found by PDW and edjotron (Message 503) Posted 11 Feb 2021 by Hy Post: If anyone's interested in the changes we're making to kaktwoos (the optimizations), here's the github and code changes that you can hopefully try to understand https://github.com/minecrafthome/kaktwoos-cl/pull/23 (Application code) https://github.com/minecrafthome/minecrafthome/pull/69 (BOINC configuration code) https://minecraftathome.com/minecrafthome/forum_thread.php?id=88 Thread on these changes, and how you can help us (if you'd like!) Update: Have received various performance improvement reports, and some older graphics cards receiving penalties. We have also fixed a backend issue regarding Intel-Linux v2.11 application, causing some clients to not be able to work on tasks on that platform. In another thread I compiled a report on which GPUs may need to use the older kernel, and so I must ensure I match those up with their specific chip/OpenCL ID for Nvidia now. For reference, my Vega 56 is "GFX900", which can be input or used by the kaktwoos-cl-amd to load a different kernel

39) Message boards : Number crunching : WU that won't stop running, What do I do 🤷‍♂️ (Message 500) Posted 6 Feb 2021 by Hy Post: Hi, if you ever see a task/WU like that, please terminate it or close BOINC and reload as this often occurs when a GPU is unstable, a driver bug is encountered or there are resource clashes leading to kaktwoos just stalling and/or BOINC not getting the message it's done. Normally if a progress update hasn't been submitted in the expected time, BOINC will report a compute error, but this isn't reliable and we've had to make it rather long so real tasks don't get killed off (as when we tried to adjust the initial run-time estimates and bounds).

40) Message boards : News : The Grand BOINCale for Kaktwoos? (Message 498) Posted 1 Feb 2021 by Hy Post: Hi again to everyone who is reading this and/or working with us (Minecraft@Home) to find the tallest cacti, and document as many 22+ height cacti in Minecraft. We're now about 35% through our first workload (y=64) and have collected over 90 unique Superflat-category Cacti seeds and coordinates, as well as around two natural-gen failed tall cacti. While this sounds like bad news, we have been working on and using some software that is making it much easier to verify which cacti can generate naturally or not (once your tasks post cactus seeds to validate). Additionally, we have been testing some new optimizations from the talented 'Balint', who originally worked with me and our administrator Chip to port Kaktwoos to OpenCL and to as many GPUs as possible. Without promising anything, if our testing goes in the direction we hope (faster, not slower) we may be able to speed up the majority of AMD GPUs by between 30-50%. I have verified this myself with my Vega 56, where it now runs at an impressive '60m/s'; 60 million seeds per second from 40. An RX 480 has increased to near GTX 1070 levels of compute for kaktwoos, being 25m/s vs GTX's 30m/s or so. A RTX 3090 though stands between 220-300 million seeds per second, depending on if it is dedicated for compute or as a gaming/display card while it is running kaktwoos. Sadly, these OpenCL kernel changes do not play as consistently on Nvidia hardware, where a RTX 2060 gained from 50 to 60m/s speed, but yet an RTX 3090 lost 5%. While this didn't seem terrible, my laptop GTX 1650 (50m/s) was crippled to 33m/s while using the attempted optimizations. Oh yeah things are gonna be good. Due to the nature of BOINC, we cannot (currently) assign sets of GPUs (outside of vendor) to use a certain kernel, and thus these optimization attempts cannot be made universal (and hope is now placed on good RX 5700/6800 results). AMD makes up 35-40% of our compute base, and so that would make these users much more competitive, in both searching for cacti and completing tasks for credits faster. Otherwise, here's some statistics on how awfully improbable it is to find a *potential* tall-cacti now: We've found 8 23-tall cacti so far in our current run and 69 22-tall (so far, these aren't validated either) out of 15,317~ unique tasks/workunits validated. in other words, finding a 23-tall cacti is a 0.0005-0.0006 likelihood even at the 'better but not best' y=64 layer. Until the next update (in a few days), have some misc. cacti: The 2nd Natural-Gen cactus found by our 'new' validator, but sadly in pieces due to caves in the chunk causing chaos! I tried out the python 'deep_daze' image gen neural net with the phrase "a tall green cactus in a desert" as an input, and let it run on Google Colab. Funnily enough, the longer it ran, the taller this cacti got. Seems like kaktwoos isn't the only GPU Compute application that can stacc the cact!

Previous 20 · Next 20