Posts by Sergei Chernykh

1) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1894) Posted 5 days ago by Sergei Chernykh Post: There are 2 pages for preferences, please check both: https://sech.me/boinc/Amicable/prefs.php?subset=global https://sech.me/boinc/Amicable/prefs.php?subset=project
2) Message boards : Number crunching : Long running CPU tasks are all timing out. (Message 1891) Posted 6 days ago by Sergei Chernykh Post: A single-core ARM cloud VM is too weak for this. These WUs were tuned to be not too fast on the fastest GPUs (2 minutes per WU on RTX 4090), and not too slow on Raspberry Pi 4. Quad-core cloud VM can finish WUs in less than 5 days: https://sech.me/boinc/Amicable/result.php?resultid=76421032
3) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1889) Posted 7 days ago by Sergei Chernykh Post: macOS OpenCL application works, see for example this task.
4) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1886) Posted 8 days ago by Sergei Chernykh Post: No, it should've been much faster. Your GPU can complete WUs in ~3 minutes, as you can see from the list of completed WUs, and all part 3 WUs are the same for GPUs. Double check that nothing else is using GPU and that you run 1 WU at a time.
5) Message boards : News : Part 3 of the 10^21 search has started (Message 1884) Posted 10 days ago by Sergei Chernykh Post: All old v3.08 WUs need to be re-run (each WU missed ~12.5% of amicable pairs), but earned credit will remain.
6) Message boards : News : Part 3 of the 10^21 search has started (Message 1882) Posted 10 days ago by Sergei Chernykh Post: I found a bug in OpenCL applications, which resulted in many false negatives (around 12.5%). The bug was in the new code that I added for the part 3, so the previous search is unaffected. Version 3.09 has this bug fixed, but I had to restart the part 3. Luckily, we were only a couple of weeks in.
7) Message boards : Number crunching : RAC dropping (Message 1879) Posted 16 days ago by Sergei Chernykh Post: Part 3 WUs are much faster on GPUs compared to CPUs. I might launch the CPU-only application to search for the remaining amicable pairs (m=m1*p where p > 10¹⁴) before the part 3 finishes to give some meaningful work to CPUs. But it's not ready yet because I still need to find a good way to parallelize it.
8) Message boards : Bug tracker : v3.08 (opencl_nvidia) x86_64-pc-linux-gnu Error while computing (Message 1868) Posted 22 days ago by Sergei Chernykh Post: Error code -4 means not enough GPU memory. Try to run only 1 WU at a time. You can also try to reduce kernel size to 21, it will reduce memory usage.
9) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1865) Posted 23 days ago by Sergei Chernykh Post: Edit: I re-read your post and you wrote "4 threads per WU". This is a multi-threaded application, it can use all CPU cores - you don't need to run many WUs in parallel. I tuned part 3 WUs to take ~2 hours on a 16-core Ryzen 7950X (using all 32 threads). Please note that GPUs are much faster with part 3 WUs (less than 2 minutes on RTX 4090), so CPUs are expected to be slow.
10) Message boards : News : Part 3 of the 10^21 search has started (Message 1862) Posted 23 days ago by Sergei Chernykh Post: Part 2 of the 10²¹ search will be finished, for the most part, today. There are still ~2000 unfinished WUs which can take up to a couple of weeks to complete, depending on individual participants processing them. Part 3 of the search will look for amicable pairs where the smaller member of the pair has its largest prime factor between 10¹¹ and 10¹⁴: m=m1p, 10¹¹ < p < 10¹⁴ There are too many primes larger than 10¹⁴ (all primes up to 510¹⁹ need to be checked), so they will require a different approach and a new application when the time comes. Most of the participants have already started getting part 3 WUs. The new work units will require ~2.5-2.7 GB RAM on CPU, and the same amount of RAM on GPU (for GPU applications), so the minimum requirement for GPUs will be 3 GB RAM. Update April 9th, 2024: I found a bug in OpenCL applications, which resulted in many false negatives (around 12.5%). The bug was in the new code that I added for the part 3, so the previous search is unaffected. Version 3.09 has this bug fixed, but I had to restart the part 3. Luckily, we were only a couple of weeks in.
11) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1857) Posted 28 days ago by Sergei Chernykh Post: Which WUs, can you give a link? Some WUs are longer than others, sometimes many times longer. Thankfully, part 3 WUs are much more uniform, and it will start in about a week from now.
12) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1855) Posted 28 days ago by Sergei Chernykh Post: Credit per task will remain the same but tasks will be smaller than we have right now.
13) Message boards : Number crunching : Intel ARC support ? (Message 1843) Posted 3 Mar 2024 by Sergei Chernykh Post: I've updated the GPU application to v3.08 which should have this error -30 bug fixed. But I don't guarantee that there are no other bugs left.
14) Message boards : Number crunching : Intel ARC support ? (Message 1841) Posted 3 Mar 2024 by Sergei Chernykh Post: System restarted and it processed a couple of tasks, but in error This is a bug in the code, I'll fix it soon. If it doesn't work after this, I'll get back to it tomorrow.
15) Message boards : Number crunching : Intel ARC support ? (Message 1839) Posted 3 Mar 2024 by Sergei Chernykh Post: I've checked in the BOINC's code how "GPUs not usable" error can happen, and it happens when you use remote desktop to login to a computer, or if you switch between different logged in users on the same computer.
16) Message boards : Number crunching : Intel ARC support ? (Message 1836) Posted 3 Mar 2024 by Sergei Chernykh Post: I've changed the plan class to: <plan_class> <name>opencl_intel_gpu</name> <opencl/> <gpu_type>intel</gpu_type> <host_summary_regex>Arc</host_summary_regex> <min_gpu_ram_mb>800</min_gpu_ram_mb> <gpu_ram_used_mb>800</gpu_ram_used_mb> <cpu_frac>0.0001</cpu_frac> </plan_class> This regex is very simple and it should match your host.
17) Message boards : Number crunching : Intel ARC support ? (Message 1834) Posted 3 Mar 2024 by Sergei Chernykh Post: I entered the wrong name for Intel GPUs in the plan class, it should be fixed now.
18) Message boards : Number crunching : Intel ARC support ? (Message 1832) Posted 3 Mar 2024 by Sergei Chernykh Post: I have enabled Intel ARC GPUs with the new GPU application version (3.07). If you have GPU tasks enabled on your end, you should be getting tasks now.
19) Message boards : Number crunching : Intel ARC support ? (Message 1831) Posted 3 Mar 2024 by Sergei Chernykh Post: Thanks! I will try to remove the "Intel bad" check from my code tomorrow, and use the similar config.
20) Message boards : Number crunching : Intel ARC support ? (Message 1827) Posted 3 Mar 2024 by Sergei Chernykh Post: Maybe I could find a set of parameters for plan class that would filter out integrated GPUs: https://boinc.berkeley.edu/trac/wiki/AppPlanSpec#GPUApps For example, I could limit min_gpu_ram_mb to 5 GB for Intel GPUs, min_gpu_peak_speed to some value that Intel ARC GPUs pass and integrated GPUs don't pass. I'd like to know how exactly PrimeGrid did this.
21) Message boards : Number crunching : RAC dropping (Message 1825) Posted 3 Mar 2024 by Sergei Chernykh Post: It depends on which number ranges the project is checking. Some are slower, some are faster. Thankfully, part 3 WUs are much more uniform, and I tweaked them to be a bit faster than what part 2 has.
22) Message boards : Number crunching : Intel ARC support ? (Message 1824) Posted 3 Mar 2024 by Sergei Chernykh Post: https://github.com/SChernykh/Amicable/blob/boinc-opencl-version-128-bit/Amicable/Amicable_OpenCL.cpp#L138 Now I remember why I disabled it. Intel has many CPUs with integrated graphics (which is not Intel ARC), and their driver quality is very bad, they never worked. Maybe they improved today.
23) Message boards : Number crunching : Intel ARC support ? (Message 1818) Posted 2 Mar 2024 by Sergei Chernykh Post: You can just download https://sech.me/boinc/Amicable/download/Amicable_OpenCL_v_3_06.exe and then run it with command line /from 25719259777759 /to 25719259875541 /task_size 500185564895 . If it runs and finishes without errors, then it should work fine on Intel ARC.
24) Message boards : Number crunching : Intel ARC support ? (Message 1816) Posted 2 Mar 2024 by Sergei Chernykh Post: Both AMD and NVIDIA use the same OpenCL application here, so if Intel drivers are stable enough to run it, it's possible. You can run the GPU app in offline mode and pass it the command line from any workunit to see if it works.
25) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1794) Posted 11 Feb 2024 by Sergei Chernykh Post: The test finished, new applications were confirmed to work.
26) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1787) Posted 24 Jan 2024 by Sergei Chernykh Post: It's good that macOS application works because I don't have any Mac around to test it. If you want to get 100% consistent GPU load, you'll need to run 2 tasks in parallel - you have enough GPU memory for this.
27) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1785) Posted 24 Jan 2024 by Sergei Chernykh Post: I've added a few more work units to the test. You should get some tasks soon.
28) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1783) Posted 24 Jan 2024 by Sergei Chernykh Post: Yes, no one with macOS had test applications enabled.
29) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1781) Posted 24 Jan 2024 by Sergei Chernykh Post: I will reduce GPU memory requirements back to 1 GB as soon as this test is over, so you can keep crunching until the end of part 2.
30) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1779) Posted 24 Jan 2024 by Sergei Chernykh Post: It requires ~2550 MB with kernel size 21 and ~2300 MB with kernel size 18, so it's definitely more than 2047 MB. GPUs with <= 2GB RAM can't run it, they will most likely fail with an error. P.S. Maybe your GPU drivers used system memory?
31) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1775) Posted 24 Jan 2024 by Sergei Chernykh Post: I hope this error is fixed in v3.06 as well.
32) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1774) Posted 24 Jan 2024 by Sergei Chernykh Post: The self-test should be fixed in v3.06 of the test application, please check it.
33) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1772) Posted 24 Jan 2024 by Sergei Chernykh Post: Hmm, I'm actually getting useful data now, for example in https://sech.me/boinc/Amicable/result.php?resultid=72251208 there is a list of numbers which are supposed to be amicable, but they're not. It means there is some bug in the self test (in combination with GCC compiler), I'll look into it.
34) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1771) Posted 24 Jan 2024 by Sergei Chernykh Post: This test is to find out how many computers actually fail the self-test. If it prints that your GPU couldn't find some known amicable numbers, then it's useless for the project (with the current application). If it turns out that all Linux hosts fail the test, then I will need to figure out how to fix it, even if it's not a bug in my code (Windows hosts work fine).
35) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1770) Posted 24 Jan 2024 by Sergei Chernykh Post: The older version doesn't have this self-test, so even if it ran smoothly it could skip some numbers. Self-test makes sure that the OpenCL code is compiled correctly and works as it was supposed to work. I've updated the test application to a new version, it has a few bugfixes and it also prints more detailed information if self-test fails. You can test how your computers work with the new version.
36) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1767) Posted 23 Jan 2024 by Sergei Chernykh Post: "Self-test failed" means that the compiled OpenCL code couldn't process a known amicable number properly and returned false negative for at least one of them. The new GPU application has a built-in list of ~2000 numbers to check all the essential code paths. If you get this error, the only way to fix it is to try different driver versions. The driver/OS combination you use can't compile the OpenCL code correctly.
37) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1765) Posted 23 Jan 2024 by Sergei Chernykh Post: The beta test is running now. I've added 2000 work units to find different ways of how new GPU applications can fail. You can expect many failed tasks there, so if you don't want it, you can disable "Run test applications" in https://sech.me/boinc/Amicable/prefs.php?subset=project
38) Message boards : News : Beta test of part 3 of the 10^21 search (Message 1764) Posted 22 Jan 2024 by Sergei Chernykh Post: Part 2 of the 10^21 search will be over some time in March this year, so it's time to start getting ready for part 3! I found a few bugs in the current GPU applications that needed to be fixed (CPU applications are fine). To make sure everything works as intended, I will launch a small beta-test of the part 3 search by the end of this week. The new work units will require ~2.5-2.7 GB RAM on CPU, and the same amount of RAM on GPU (for GPU applications), so the minimum requirement for GPUs will be 3 GB RAM. The purpose of the test is to confirm that all OS/CPU/GPU combinations work properly and confirm that 3 GB GPUs can process the new work units.
39) Message boards : Bug tracker : Amicable Numbers using ~8 gigabytes of Ram (Message 1763) Posted 22 Jan 2024 by Sergei Chernykh Post: This is memory allocated by the program. Not all of it is in the physical computer memory, some part of it will be swapped out if you exceed 8 GB (in your case). It's not recommended because swapping slows down computations significantly.
40) Message boards : Bug tracker : 2) Another workunit that is stuck but both computers finished the task (Message 1756) Posted 13 Jan 2024 by Sergei Chernykh Post: There were more than 1000 such workunits, I don't know why exactly. I've manually unstuck them.
41) Message boards : Random stuff : Support for Intel GPUs? (Message 1748) Posted 2 Jan 2024 by Sergei Chernykh Post: If they support OpenCL (they should), then the existing GPU app should already work there.
42) Message boards : Getting started : process exited with code 255 (Message 1723) Posted 20 Sep 2023 by Sergei Chernykh Post: If you google for "Error getting function data from server", it looks like some problem with GPU driver. Newer (or sometimes older) driver versions might work without this error.
43) Message boards : Number crunching : 10^22 (Message 1717) Posted 17 Aug 2023 by Sergei Chernykh Post: 10^21 search will also have part 3 and it will take more time than part 2, we're talking years and years more. So it's a bit early to talk about 10^22
44) Message boards : Bug tracker : WU waiting for validation but both tasks are finished since the day before yesterday! (Message 1713) Posted 3 Aug 2023 by Sergei Chernykh Post: Fixed
45) Message boards : Number crunching : Long running tasks (Message 1706) Posted 30 Jul 2023 by Sergei Chernykh Post: I commented regarding this issue before, it's not the first time the project goes through slow search ranges. The positive side of this is we're finding a lot of new amicable pairs now (more than 4000 new pairs today).
46) Message boards : Bug tracker : Missing French translations for project actions in Boinc Manager (Message 1696) Posted 2 Jul 2023 by Sergei Chernykh Post: gui_urls.xml is a single file in project root folder, it doesn't have any translations. As far as I can see, it's not possible to add translations for it.
47) Message boards : Bug tracker : Infinite loop WU ? (Message 1688) Posted 17 Jun 2023 by Sergei Chernykh Post: Factorization '0' is incorrect: 0 is not a prime This error means that your checkpoint file is corrupt. You should manually delete it from working folder - the file name is "amicable_checkpoint".
48) Message boards : Bug tracker : Infinite loop WU ? (Message 1686) Posted 17 Jun 2023 by Sergei Chernykh Post: I tested this WU on both Windows and Linux CPU versions, and it didn't get into an infinite loop. Windows: 11:00:11 (7588): Can't open init data file - running in standalone mode 11:35:25 (7588): called boinc_finish(0) Linux: 12:05:18 (2396): Can't open init data file - running in standalone mode 12:40:55 (2396): called boinc_finish(0) It took around 35-36 minutes on 7950X, so it should take no more than 5-6 hours even on weaker CPU. It's one of those WUs that are better suited for GPU application.
49) Message boards : Bug tracker : Your ReCaptcha response was not correct. Please try again. (Message 1671) Posted 7 May 2023 by Sergei Chernykh Post: Thanks for the hint, I've added https://www.google.com/recaptcha/api.js to the page template.
50) Message boards : Number crunching : No new amicable pairs (Message 1666) Posted 27 Apr 2023 by Sergei Chernykh Post: There are not many new pairs right now, but they are being found: https://sech.me/ap/log/2023/2023-04-27/VOLUNTEERS%40BOINC.txt - in case someone was wondering if everything works correctly.

Next 50