Posts by ahorek's team
|
41)
(Message 8769)
Posted 4 Feb 2025 by ahorek's team Post: The FP32 marketing numbers don’t tell the whole story. Actual compute performance depends on how an application utilizes the GPU, each app may behave differently and memory bandwidth can also be a limiting factor. I recall when AMD had a specialized rotate instruction, while NVIDIA had to emulate it with four instructions to achieve the same result. In rare cases, such as SHA-256 calculations, AMD was significantly faster, but that feature was completely irrelevant for gaming. I agree that the situation has changed, and NVIDIA is now far ahead of the competition. While other vendors can still compete in gaming, NVIDIA typically holds a significant advantage when it comes to general computing performance. |
|
42)
(Message 8768)
Posted 4 Feb 2025 by ahorek's team Post: Ok, I have determined the cause of the error. When the system is under load, it first responds by throttling the CPU speed. However, if the load becomes too intense and the device cannot cool itself effectively, the Android system aggressively terminates applications to prevent permanent damage. The new app is more optimized for ARM CPUs, increasing the strain on the device. While mobile devices can handle short bursts of high load, their passive cooling systems are not designed for prolonged heavy workloads. The only solution is to reduce the number of active CPUs in BOINC preferences. Modern mobile CPUs are quite powerful and capable of reaching speeds up to 3GHz, but only for short durations. While utilizing all available cores (typically 8) might seem ideal for maximum performance, it is not a good idea. Temperature is the primary limiting factor. |
|
43)
(Message 8765)
Posted 4 Feb 2025 by ahorek's team Post: Interesting! Where did you obtain the data? FP32 isn't a valid comparison for Asteroids. The app relies on FP64 and the peak performance could vary between generations/vendors. Each new generation usually has a worse FP32/FP64 ratio because FP64 isn't very useful for games and all vendors want to sell (much more expensive) professional cards for computing purposes. https://www.techpowerup.com/gpu-specs/geforce-rtx-4090.c3889 btw INT performance usually isn't in marketing performance charts, but you can measure it with tools like clpeak gaming performance could be very different from computing performance and it also depends on what the app is doing, for instance: primegrid apps typically use INT32 calculations (where NVidia excels) einstein, gpugrid is mainly FP32 with some FP64 (usually favors AMD) asteroids, milkyway (the old separation app) FP64 regarding asteroids, the current app isn't limited by CPU or raw computational power but rather by memory access, so it doesn't scale / is as efficient as it could be, but we're working on optimizations. |
|
44)
(Message 8762)
Posted 31 Jan 2025 by ahorek's team Post: Hi olicheski, sorry, but that card is quite outdated. Even 15 years ago, it was considered a low-end model. This project relies on FP64 support, which isn't supported, not to mention the lack of CUDA support (only 1.2). even if it were technically possible to build an app for such an old card, based on the performance estimate, it would likely take around 5 days to complete a single task. Your C2D can finish 2 tasks in just 2:45 hours, so it's definitely not worth it. there’s a slim possibility that some OpenCL-based projects that don’t depend on FP64 might still run on that card (try Einstein or PrimeGrid), but in any case, the performance will be extremely poor. |
|
45)
(Message 8756)
Posted 19 Jan 2025 by ahorek's team Post: btw, the new app automatically selects the optimal optimization level for your CPU at runtime. This eliminates the need for specific versions to take advantage of features like AVX512. |
|
46)
(Message 8748)
Posted 13 Jan 2025 by ahorek's team Post: yes, your phone should be supported ensure that CPU work is permitted, try to manually update the project, and review the scheduler logs... or perhaps wait a little, sometimes the server won't send new work units immediately or may be temporarily out of work https://asteroidsathome.net/boinc/server_status.php but currently, it should work fine, I'm getting work: https://asteroidsathome.net/boinc/result.php?resultid=542571219 |
|
47)
(Message 8741)
Posted 5 Jan 2025 by ahorek's team Post: OpenCL 1.2 support is enough, the problem is a multigpu setup BOINC GPU type 'ATI', deviceId=1 For some reason, you have three AMD platforms, and each one contains a GPU device. Therefore, the correct matrix (platform:device) should look like this: 1:0 2:0 3:0 but the app iterates cards as devices on the first AMD platform 1:0 1:1 1:2 so 1/3 tasks will fail because the app selects a plaform/device that doesn't exist Error: Failed to create a device group! invalid device (-33) it's a bug in the app that should be fixed. anyway, don't expect good performace, as the card is old and underpowered. It took 10 hours to complete a work unit, even mobile CPUs easily outperform it. I would keep it at Einstein, that can better utilize GCN cards. https://www.techpowerup.com/gpu-specs/radeon-550.c3407 |
|
48)
(Message 8729)
Posted 4 Jan 2025 by ahorek's team Post: Radim & Josef Ďurech are still maintaining the project, they just aren't active on forums anymore I and Georgi are working on the app workunits require some preprocessing and are manually filled in, enabling the prioritization of specific tasks. However, when the queue runs dry, there are days when no work is available. Take a look at history charts: https://grafana.kiska.pw/d/boinc/boinc?orgId=1&var-project=Asteroids@Home&from=now-30d&to=now While not everything is perfect and yes, there are issues, the project is alive... |
|
49)
(Message 8720)
Posted 29 Dec 2024 by ahorek's team Post: einstein@home offers a beta (CPU) app, that should work on your Android btw I recently tested my Ardeno 750 GPU: 1/ Primegrid performance (GFN) was terrible https://github.com/galloty/genefer22/issues/12#issuecomment-2558303608 2/ AP27 even worse. Unusable. 3/ Einstein (BRP4) was ok (like Intel iGPUs). However, there were server/platform challenges, so it may take some time for them to release it if ever. 4/ The Asteroids app cannot utilize the GPU because Adreno lacks FP64 support. |
|
50)
(Message 8718)
Posted 29 Dec 2024 by ahorek's team Post: your phone uses a mixed CPU architecture 1x fast core 3x performance cores 4x efficiency cores Using all cores isn't ideal due to thermal concerns. In your case, I recommend limiting usage to 4 cores. The app already utilizes optimal ASIMD optimizations, so no further architectural improvements are possible. It's still a mobile device, so don't expect performance comparable to laptop/desktop/server PCs. |
|
51)
(Message 8698)
Posted 22 Dec 2024 by ahorek's team Post: ok, a correction, exit code 194 means EXIT_ABORTED_BY_CLIENT which is raised by the boinc. It's hard to say why the WU has been aborted on your side. You may run out of memory, free disk space, there was an unexpected shutdown or something like that... |
|
52)
(Message 8693)
Posted 21 Dec 2024 by ahorek's team Post: after some digging, WIndows 11 ARM already does support most existing projects via x64 emulation. for optimal performance, we have to build a native ARM app which I did and tested in a VM. We can release it in the next version if all goes well on the server side. note that Boinc does not officially support Windows on ARM yet, but the new installer will be released very soon. |
|
53)
(Message 8692)
Posted 21 Dec 2024 by ahorek's team Post: your computers are hidden, it may help if you unhide them. anyway, Mac M4 mini should work fine. What does the event log say? |
|
54)
(Message 8691)
Posted 20 Dec 2024 by ahorek's team Post: 1/ HSA_STATUS_ERROR_OUT_OF_RESOURCES is likely due to outdated drivers https://askubuntu.com/questions/1435708/hsa-status-error-out-of-resources-rocm5-1-1-ubuntu-22-04 could you try more recent drivers if there are any for Vegas? 2/ "out of host memory" means the selected intensity is too high your driver reports CL_DEVICE_MAX_SAMPLERS = 5686, but we expect 16, 32 etc. Grid dim: 79604 = 2 * 7 * 5686 in comparison with 7900 xtx Grid dim: 1536 = 2 * 48 * 16 The driver refuses to compile the application since the required intensity is insanely high. This issue can be resolved in the app by modifying how we calculate the optimal intensity for the target GPU. We should likely avoid using CL_DEVICE_MAX_SAMPLERS in the formula. |
|
55)
(Message 8681)
Posted 13 Dec 2024 by ahorek's team Post: exit code 194 = ERROR_ITERATED_DATA_EXCEEDS_64k I can't verify the work unit because it has already been successfully processed by other hosts, but the issue might stem from the code (AVX512?) or a random system failure on your host. all of your other tasks (~2200) were valid, so it looks like a random glitch... I have no idea, sorry. |
|
56)
(Message 8679)
Posted 9 Dec 2024 by ahorek's team Post: The current app requires a CPU with ARM v8a support, but your device uses v7l. In a future version, we may restore support for older Android devices. However, this could take some time as we work on completing other planned features. |
|
57)
(Message 8660)
Posted 5 Dec 2024 by ahorek's team Post: There were two issues, a faulty network cable and insufficient free disk space. Unfortunately, I haven’t received any additional information from admins since Monday. Perhaps something else went wrong, I don't know... I’m unable to take any action as I don’t have access to the server. |
|
58)
(Message 8636)
Posted 2 Dec 2024 by ahorek's team Post: There was a partial fault in the network cable, along with some disk space issues. The admins are currently investigating these problems. Hopefully, the server will be back to normal soon. |
|
59)
(Message 8624)
Posted 15 Nov 2024 by ahorek's team Post: As far as I know, no BOINC project currently offers native support for Windows 11 on ARM. Building an asteroids app for it shouldn't be too difficult, since we already support ARM. I can give it a shot. However, I don’t have access to such a device, perhaps you could help with testing? |
|
60)
(Message 8620)
Posted 6 Nov 2024 by ahorek's team Post: yes, Apple Silicon ARM-based chips (M1+) have a native application older x86 Intel-based MACs (with Mac OS 10.14+) are also supported |
Previous 20 · Next 20