Message boards : News : Congratulations to our record finders, and a call for NVIDIA GPU beta testers
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 23 Sep 26 Posts: 242 Credit: 54,980,907 RAC: 1,491,573 |
466/480: the world-record level Congratulations to Jef, KetamiNO [YouTube], vanos0512 and mmonnin, who each found a board at 466/480 with all five official clues, the best known result under the official rules. Well done also to everyone who found one of this week's 111 boards at 465! Call for NVIDIA GPU beta testers We have built a hybrid GPU + CPU version of the search: • the graphics card explores tens of thousands of tickets at the same time; • a few CPU threads compute the deep end-game phases in parallel; • at the end of a task, the last hardest tickets are handed over to ordinary CPU tasks, with the aim of keeping the card supplied with new work. Server-side refill is still being verified. The results are identical to the CPU app, checked ticket by ticket. On our test card (RTX 5070 Ti), a task of 65,552 tickets was completed in 1 h 02. It has only run on one card so far. Before any public release, we need to test it on as many different cards and systems as possible. Who we are looking for: • NVIDIA GTX 750/750 Ti, GTX 900/10/16 and RTX 20/30/40/50 series, plus compatible Quadro/Tesla/TITAN models, subject to exact model and driver checks. A separate 48 KiB variant (2.69) is now available to approved legacy hosts; physical validation on those older cards is still in progress. GTX 760/760 Ti remain outside this beta; • Windows 10/11 or Linux 64-bit (recent distributions); • an up-to-date NVIDIA driver. What to expect: power use varies with the workload; the GPU application also reserves some CPU resources, so expect additional heat and fan noise. It is a beta: tasks may run longer, or fail, while we tune it. How to join: anyone who would like to test can reply below this post or send me a private message. Please include your BOINC username, your card model and memory, your operating system, your driver version, and whether the computer runs most of the day. We will enable the beta on your account and keep you posted. More information : https://boinc.eternityathome.org/eternity/forum_thread.php?id=42 Thank you all! |
|
Send message Joined: 27 Sep 26 Posts: 42 Credit: 141,621,123 RAC: 4,172,786 |
I can support with 7x GPUs 3xxx series and newer in Windows and Linux. What does up-to-date driver mean as typically there has been no benefit for a new driver for compute |
|
Send message Joined: 4 Oct 26 Posts: 7 Credit: 10,374,234 RAC: 113,518 |
I can run it on a Titan V (12 GB), driver 580.99, Ubuntu 24.04. The computer runs 24/7. |
dskagcommunitySend message Joined: 28 Sep 26 Posts: 31 Credit: 8,406,604 RAC: 229,439 |
When i know when, i can for some days temporary enable two "old Hardware" Hosts whitch are waiting for winter. Because my faster 24/7 cards are on productive hosts and im sure there are many other hosts with new cards show up in here beside me ^^ So i can betatest for older hardware i´ve ya wish. RX280s, 760ti(?) and 1060 cards are in there. Two machines, Win10, 32/16GB ram, 4 real cores eatch. I even have a 4core 16GB win10 host with a Nvidia T600 for testings if needed. They are the BOINC1, 2, 3 Computer in my account. ---------- 24/7 Crunching since 2011 ----------- DSKAG Austria: http://www.dskag.at
|
|
Send message Joined: 23 Sep 26 Posts: 242 Credit: 54,980,907 RAC: 1,491,573 |
More technical information : https://boinc.eternityathome.org/eternity/forum_thread.php?id=42 |
|
Send message Joined: 23 Sep 26 Posts: 242 Credit: 54,980,907 RAC: 1,491,573 |
Thanks for offering the older machines! A compatibility check found a limitation in the current application: the GTX 1060 cannot run its kernels yet. The GTX 760 Ti and AMD cards are also outside this beta. Your NVIDIA T600 remains a candidate, but I need a fresh hardware report before confirming it. If you still want to test that machine, please bring it online and update the project in BOINC. There is no need to move your production GPUs away from their usual work |
dskagcommunitySend message Joined: 28 Sep 26 Posts: 31 Credit: 8,406,604 RAC: 229,439 |
Im not sure if it is a 760 or 750 card (read the hardware limitationsthread), i need to check that in the afternoon. Cos the 1060 seems to have a problem in the app and its in the same PC like the 7xx card, is it ok when it remains inside JUST IN CASE its NOT a 760x? or does i have to wait for something (not producing too much errors) or i have to exclude the other GP in cc_config? And yes i will turn on the PC with the T600 too then :) I have a 980GTX too in one of my bigger 24/7 crunchers. Because..i see no Beta offer in this GPU range..i could pause Einstein for a moment to let a Eternity GPU go through ^^ Its the Host BOINCWIN4 (for technical infos) ---------- 24/7 Crunching since 2011 ----------- DSKAG Austria: http://www.dskag.at
|
|
Send message Joined: 23 Sep 26 Posts: 242 Credit: 54,980,907 RAC: 1,491,573 |
Hi, Good news: we have developed and released version 2.69, with a separate engine designed for the shared-memory limits of older NVIDIA GPUs. Your registered GTX 980 Ti, GTX 1060 and T600 hosts are now enabled. Please reconnect them, allow new tasks and update the project. We still need their first successful results to confirm compatibility. GTX 750/750 Ti are also candidates; GTX 760/760 Ti remain unsupported. For machines mixing supported and unsupported GPUs, please send the host IDs so we can check the configuration first. Thanks for offering these cards! |
|
Send message Joined: 28 Sep 26 Posts: 12 Credit: 24,121,189 RAC: 491,531 |
Ok, you can include me in the test group. Drago75, Linux Mint 22.3 with RTX 3060m, 3070ti, 4080 and 5070. They run for a great deal of the day. |
dskagcommunitySend message Joined: 28 Sep 26 Posts: 31 Credit: 8,406,604 RAC: 229,439 |
soooo...good news its a plain nvidia 750 :) so its capable too! its in the same host where the 1060 is into. id: 569 reconnecting? like leave project and rejoin again? got work on the t600 machine! the other one with 1060 and 750 is still waiting for approve i think. the 980gtx i can enable in my nightshift ^^ ---------- 24/7 Crunching since 2011 ----------- DSKAG Austria: http://www.dskag.at
|
|
Send message Joined: 23 Sep 26 Posts: 242 Credit: 54,980,907 RAC: 1,491,573 |
Hi dskagcommunity, Thanks for confirming it is a GTX 750! That model is a candidate for our legacy GPU application. By “reconnect”, I meant bringing the computer back online — please do not detach/rejoin or reset the project. Simply allow new tasks and click Update. Host 569 has already received GPU tasks and reported computing activity. However, the server currently identifies both cards as GTX 1060s, so we still need to clarify the device identification before confirming the GTX 750 separately. Good to hear the T600 has received work too! Please leave the current tasks running. You can bring the GTX 980 Ti host online whenever convenient. Thanks again! |
dskagcommunitySend message Joined: 28 Sep 26 Posts: 31 Credit: 8,406,604 RAC: 229,439 |
thats always a problem with boinc. it looks at gpu0, and set the number of gpus in front of it nethertheless witch gpus are really in as gpu1 2 3 and so on. thats why you habe to use compatible driverversions in that systems witch i do always. my 3060 rig has not 3 times a 3060 in it too. it has 3060 gpu0 1070ti gpu1 and 1060 as gpu2 . so it sees 3060 in gpu0.and.set the number.of.gpus.in front [3]. when a project not cabable of running on one of those you disable them.in cc config..as expample due low memory gpu2 runs asteroid and the others memoryhungry einstein o4abAS. so thats how you run mixed systems. its kinda resposibility of the volunteer. but yes i can confirm it runs on both gpus at the moment. even when its showing 3c+1nv on both running while its just a 4 core + (in boinc disabled) 4ht coremaschine ^^ the faster one was running on the 1060 clearly. ---------- 24/7 Crunching since 2011 ----------- DSKAG Austria: http://www.dskag.at
|
dskagcommunitySend message Joined: 28 Sep 26 Posts: 31 Credit: 8,406,604 RAC: 229,439 |
So the 980 is now ongoing too. Can it be you assign a % on CPU Cores from a host automaticly to the Tasks? Cos on the real 16 core machine it shows now 6C+1NV. Is it planned to make that changeable for the user itself? And no problem, im glad i could help ^^ Looked like a very good working Betaphase so far xD ---------- 24/7 Crunching since 2011 ----------- DSKAG Austria: http://www.dskag.at
|
rebirtherSend message Joined: 24 Sep 26 Posts: 29 Credit: 39,233 RAC: 755 |
you can use a setting in app_config.xml <app_config>
<app>
<name>app name</name>
<gpu_versions>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>change cpu_usage with your value |
|
Send message Joined: 28 Sep 26 Posts: 12 Credit: 24,121,189 RAC: 491,531 |
So far my 3060m, 5060ti, 5070 and 4080 all produced valid results. Regardless of the GPU and CPU type that I used all tasks were using 1,05 cpu threads. One curious observation is that my 4080 is a good deal slower than my 5070. The 5070 runs at 100% close to it`s TDP while the 4080 runs close 100% at 120 of 300W. So I tried to run two tasks concurrently using MPS which worked seemingly well at first, now at a consumption of 176W One task finished after 45 minutes, the second crashed right at the end at the very same second the first one reported success. I had ChatGPT run a quick check of the STDERR and here is the result: Bug report: RTX 4080 — CUDA illegal memory access near task completion While running two Eternity GPU tasks concurrently on my RTX 4080, one completed successfully and the other failed at approximately 99.49%. Task details - Result ID: 3362594 - Workunit ID: 3307352 - Host ID: 133 - Engine: e2_bw_nvidia_avx2_0.4 - Runner: bw_runner_nvidia_fill_0.13_linux - BOINC client: 8.3.0 - Platform: Linux x86_64 - BOINC exit status: 195 — EXIT_CHILD_FAILED Last recorded progress bw_runner: progress 10697/10752 (99%), restored runtime 2489 s Error NVIDIA_NATIVE_FAILURE 134 terminate called after throwing an instance of 'std::runtime_error' what(): CUDA: an illegal memory access was encountered (cudaMemcpyAsync(&made, m.nodes, sizeof made, cudaMemcpyDeviceToHost, m.stream)) The wrapper subsequently reported: app exit status: 0x86 called boinc_finish(195) Observations The initial GPU resource check passed (NVIDIA_RESOURCE_PROBE_OK). At the last recorded progress update, 55 of 10,752 work items remained, placing the task just above the configured 0.5% tail handoff threshold (--runner-tail-fraction 0.005). Successful tasks logged pipeline: handoff near this threshold. The failed task did not log a handoff before terminating. This may indicate an issue while processing the remaining GPU work or approaching the handoff, but the logs do not establish the underlying cause. Concurrent execution may also be relevant; reproducibility has not yet been established. Workunit 3307352 |
dskagcommunitySend message Joined: 28 Sep 26 Posts: 31 Credit: 8,406,604 RAC: 229,439 |
i dont change potential mandatory things in betas an sabotage them ;) ---------- 24/7 Crunching since 2011 ----------- DSKAG Austria: http://www.dskag.at
|
|
Send message Joined: 20 Oct 26 Posts: 2 Credit: 53,250,998 RAC: 849,739 |
RFGuy_KCCO NVIDIA GeForce RTX 2080 SUPER (driver version 580.99, CUDA version 13.0, compute capability 7.5, 7.60GB, 7.60GB available, 11244 GFLOPS peak) OS: Linux Linuxmint: Linux Mint 22.3 [6.17.0-44-generic|libc 2.39] Computer runs 24x7 |
|
Send message Joined: 23 Sep 26 Posts: 242 Credit: 54,980,907 RAC: 1,491,573 |
Thank you Dargo75, this is a very useful report, and great to see valid results from all four cards! About 1 CPU thread: that is normal. Extra CPU threads are only used when deep endgames show up, so they are not busy all the time. RTX 4080 at 120 W of 300 W: we see the same under-loading on some other cards. The GPU is waiting on the CPU-side endgames. We are working on it; it is not a fault of your machine. Two tasks at once with MPS: please run only one GPU task per card during the beta. Your crash (illegal memory access near the end, result 3362594) has been passed on to the developers with your log. Running several tasks per card is not supported yet. |
|
Send message Joined: 23 Sep 26 Posts: 242 Credit: 54,980,907 RAC: 1,491,573 |
Thanks rebirther, that's a handy tip in general! For the beta, though, we'd prefer everyone to keep the default CPU setting: the app occasionally needs those CPU threads for deep endgames, and keeping the defaults makes our measurements comparable across machines. |
|
Send message Joined: 29 Sep 26 Posts: 4 Credit: 2,503,434 RAC: 77,792 |
I run 2 GTX 970 and can help with this graphic cards ! |
