Anurag1734/cuda-error-resolution-analysis
07
1[2 {3 "post_stream": {4 "posts": [5 {6 "id": 475176,7 "name": "Ανδρεας Κατσινουλας",8 "username": "adrew25",9 "avatar_template": "/user_avatar/discuss.pytorch.org/adrew25/{size}/43719_2.png",10 "created_at": "2025-09-08T17:25:32.751Z",11 "cooked": "<p>Hi new to the forum and i am a bit confused with the HW right now in the market.<br>\nI’m planning a new workstation build with a Ryzen 9 9950X3D, 96GB RAM, and dual GPUs. The issue is that most X870E motherboards only provide one PCIe 5.0 x16 slot, while the other GPU would have to run on a PCIe 4.0/5.0 x4 slot (through the chipset).<br>\nMy main workloads are:<br>\n1.Training and fine-tuning LLMs (PyTorch, DDP/HF Accelerate).<br>\n2. Inference for LLMs.<br>\n3. Computer vision training and inference.<br>\n4! And of course causal pc usage( Playing a game etc.)<br>\nMy questions are:<br>\n1.How much of an impact does having one GPU on PCIe x16 and the other on PCIe x4 have on training (gradient synchronization, data loading, etc.)?<br>\n2. Is inference also affected significantly by the reduced bandwidth, or is it mostly negligible?<br>\n3. Would it make sense to prioritize an AM5 board with x8/x8 support, or is the bottleneck only noticeable for very large models?<br>\nThanks in advance for any insights i am struggling to figure out before i give a tone of money. ( Miss the days where the hw was simpler… now the consumer hw is bloated with wifi and rgb….)</p>",12 "post_number": 1,13 "post_type": 1,14 "posts_count": 2,15 "updated_at": "2025-09-08T17:25:32.751Z",16 "reply_count": 0,17 "reply_to_post_number": null,18 "quote_count": 0,19 "incoming_link_count": 114,20 "reads": 14,21 "readers_count": 13,22 "score": 497.8,23 "yours": false,24 "topic_id": 222946,25 "topic_slug": "impact-of-pcie-lane-configuration-on-multi-gpu-training-and-inference",26 "display_username": "Ανδρεας Κατσινουλας",27 "primary_group_name": null,28 "flair_name": null,29 "flair_url": null,30 "flair_bg_color": null,31 "flair_color": null,32 "flair_group_id": null,33 "badges_granted": [],34 "version": 1,35 "can_edit": false,36 "can_delete": false,37 "can_recover": false,38 "can_see_hidden_post": false,39 "can_wiki": false,40 "read": true,41 "user_title": null,42 "bookmarked": false,43 "actions_summary": [44 {45 "id": 2,46 "count": 247 }48 ],49 "moderator": false,50 "admin": false,51 "staff": false,52 "user_id": 50461,53 "hidden": false,54 "trust_level": 1,55 "deleted_at": null,56 "user_deleted": false,57 "edit_reason": null,58 "can_view_edit_history": true,59 "wiki": false,60 "post_url": "/t/impact-of-pcie-lane-configuration-on-multi-gpu-training-and-inference/222946/1",61 "can_accept_answer": false,62 "can_unaccept_answer": false,63 "accepted_answer": false,64 "topic_accepted_answer": null,65 "can_vote": false66 },67 {68 "id": 475384,69 "name": "Tony Robinson",70 "username": "tonyr",71 "avatar_template": "/user_avatar/discuss.pytorch.org/tonyr/{size}/74253_2.png",72 "created_at": "2025-09-14T14:15:18.933Z",73 "cooked": "<p>I’m also speccing a dual (5090) box right now and was wondering the same. At the moment I’m doing LLM research, so just passing in a batch of token IDs, very low bandwidth. I have a local hardware builder (<a href=\"https://woc.co.uk/\" rel=\"noopener nofollow ugc\">https://woc.co.uk/</a>) whom I really trust as I’ve been buying computers for machine learning from them for the last 25 years. They have convinced me to get a Xeon W5-3525 CPU with ASUS PRO WS W790E-SAGE motherboard because nothing less is going to support two 16x PCIe lanes. The dual 5090 setup is common for them, they supply to Cambridge University and local startups, they make a business out of getting ti right. Now, I’m no expert, but I do know that the 5090s will cost me £4,000 here in the UK, the supporting hardware is £4,200, so that’s as much for the computer as the GPUs. This doesn’t seem unreasonable to me, it’s easy to swap out GPUs in a few years when a 6090 or 7090 comes out, it’s hard to upgrade a motherboard (I’ve done it once - never again - too many things to unplug and forget where they go). Your game plan will be different, but I struggled with the single x16 slot issue for a while and then just gave up and went on trusting my supplier.</p>",74 "post_number": 2,75 "post_type": 1,76 "posts_count": 2,77 "updated_at": "2025-09-17T00:40:50.487Z",78 "reply_count": 0,79 "reply_to_post_number": null,80 "quote_count": 0,81 "incoming_link_count": 0,82 "reads": 9,83 "readers_count": 8,84 "score": 1.8,85 "yours": false,86 "topic_id": 222946,87 "topic_slug": "impact-of-pcie-lane-configuration-on-multi-gpu-training-and-inference",88 "display_username": "Tony Robinson",89 "primary_group_name": null,90 "flair_name": null,91 "flair_url": null,92 "flair_bg_color": null,93 "flair_color": null,94 "flair_group_id": null,95 "badges_granted": [],96 "version": 2,97 "can_edit": false,98 "can_delete": false,99 "can_recover": false,100 "can_see_hidden_post": false,101 "can_wiki": false,102 "link_counts": [103 {104 "url": "https://woc.co.uk/",105 "internal": false,106 "reflection": false,107 "title": "World of Computers Ltd",108 "clicks": 1109 }110 ],111 "read": true,112 "user_title": null,113 "bookmarked": false,114 "actions_summary": [],115 "moderator": false,116 "admin": false,117 "staff": false,118 "user_id": 30062,119 "hidden": false,120 "trust_level": 2,121 "deleted_at": null,122 "user_deleted": false,123 "edit_reason": null,124 "can_view_edit_history": true,125 "wiki": false,126 "post_url": "/t/impact-of-pcie-lane-configuration-on-multi-gpu-training-and-inference/222946/2",127 "can_accept_answer": false,128 "can_unaccept_answer": false,129 "accepted_answer": false,130 "topic_accepted_answer": null131 }132 ],133 "stream": [134 475176,135 475384136 ]137 },138 "timeline_lookup": [139 [140 1,141 47142 ],143 [144 2,145 41146 ]147 ],148 "suggested_topics": [149 {150 "fancy_title": "NVIDIA L40S-48Q and “RuntimeError: CUDA error: operation not supported”",151 "id": 212716,152 "title": "NVIDIA L40S-48Q and \"RuntimeError: CUDA error: operation not supported\"",153 "slug": "nvidia-l40s-48q-and-runtimeerror-cuda-error-operation-not-supported",154 "posts_count": 11,155 "reply_count": 9,156 "highest_post_number": 11,157 "image_url": null,158 "created_at": "2024-11-08T19:29:41.292Z",159 "last_posted_at": "2025-02-11T14:25:17.459Z",160 "bumped": true,161 "bumped_at": "2025-02-11T14:26:35.274Z",162 "archetype": "regular",163 "unseen": false,164 "pinned": false,165 "unpinned": null,166 "visible": true,167 "closed": false,168 "archived": false,169 "bookmarked": null,170 "liked": null,171 "tags_descriptions": {},172 "like_count": 1,173 "views": 1360,174 "category_id": 1,175 "featured_link": null,176 "has_accepted_answer": false,177 "posters": [178 {179 "extras": null,180 "description": "Original Poster",181 "user": {182 "id": 7291,183 "username": "Chris_Palmer",184 "name": "Chris Palmer",185 "avatar_template": "/user_avatar/discuss.pytorch.org/chris_palmer/{size}/12322_2.png",186 "trust_level": 1187 }188 },189 {190 "extras": null,191 "description": "Frequent Poster",192 "user": {193 "id": 3534,194 "username": "ptrblck",195 "name": "",196 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",197 "admin": true,198 "moderator": true,199 "trust_level": 2200 }201 },202 {203 "extras": "latest",204 "description": "Most Recent Poster",205 "user": {206 "id": 82625,207 "username": "briskajanis1",208 "name": "Briskajanis1",209 "avatar_template": "/user_avatar/discuss.pytorch.org/briskajanis1/{size}/75599_2.png",210 "trust_level": 0211 }212 }213 ]214 },215 {216 "fancy_title": "Why is my generator model M_gen not training when optimizing based on the classifier model M_cls?",217 "id": 216263,218 "title": "Why is my generator model M_gen not training when optimizing based on the classifier model M_cls?",219 "slug": "why-is-my-generator-model-m-gen-not-training-when-optimizing-based-on-the-classifier-model-m-cls",220 "posts_count": 7,221 "reply_count": 4,222 "highest_post_number": 7,223 "image_url": null,224 "created_at": "2025-02-05T10:43:44.579Z",225 "last_posted_at": "2025-02-09T22:42:53.438Z",226 "bumped": true,227 "bumped_at": "2025-02-09T22:42:53.438Z",228 "archetype": "regular",229 "unseen": false,230 "pinned": false,231 "unpinned": null,232 "visible": true,233 "closed": false,234 "archived": false,235 "bookmarked": null,236 "liked": null,237 "tags_descriptions": {},238 "like_count": 0,239 "views": 119,240 "category_id": 1,241 "featured_link": null,242 "has_accepted_answer": false,243 "posters": [244 {245 "extras": "latest",246 "description": "Original Poster, Most Recent Poster",247 "user": {248 "id": 82500,249 "username": "Brayn_O_Conner",250 "name": "Brayn O'Conner",251 "avatar_template": "/user_avatar/discuss.pytorch.org/brayn_o_conner/{size}/75477_2.png",252 "trust_level": 0253 }254 },255 {256 "extras": null,257 "description": "Frequent Poster",258 "user": {259 "id": 18088,260 "username": "KFrank",261 "name": "K. Frank",262 "avatar_template": "/letter_avatar_proxy/v4/letter/k/ecb155/{size}.png",263 "trust_level": 2264 }265 }266 ]267 },268 {269 "fancy_title": "C++ Libtorch MHA issues",270 "id": 217590,271 "title": "C++ Libtorch MHA issues",272 "slug": "c-libtorch-mha-issues",273 "posts_count": 1,274 "reply_count": 0,275 "highest_post_number": 1,276 "image_url": null,277 "created_at": "2025-03-08T12:05:10.374Z",278 "last_posted_at": "2025-03-08T12:05:10.414Z",279 "bumped": true,280 "bumped_at": "2025-03-08T14:24:18.722Z",281 "archetype": "regular",282 "unseen": false,283 "pinned": false,284 "unpinned": null,285 "visible": true,286 "closed": false,287 "archived": false,288 "bookmarked": null,289 "liked": null,290 "tags_descriptions": {},291 "like_count": 0,292 "views": 27,293 "category_id": 1,294 "featured_link": null,295 "has_accepted_answer": false,296 "posters": [297 {298 "extras": "latest single",299 "description": "Original Poster, Most Recent Poster",300 "user": {301 "id": 78419,302 "username": "HenryZNNUE",303 "name": "Henry Z",304 "avatar_template": "/user_avatar/discuss.pytorch.org/henryznnue/{size}/76039_2.png",305 "trust_level": 1306 }307 }308 ]309 },310 {311 "fancy_title": "Hyperparameter Sweeping by Only Measuring Convergence",312 "id": 217603,313 "title": "Hyperparameter Sweeping by Only Measuring Convergence",314 "slug": "hyperparameter-sweeping-by-only-measuring-convergence",315 "posts_count": 1,316 "reply_count": 0,317 "highest_post_number": 1,318 "image_url": null,319 "created_at": "2025-03-08T19:04:40.171Z",320 "last_posted_at": "2025-03-08T19:04:40.209Z",321 "bumped": true,322 "bumped_at": "2025-03-08T19:04:40.209Z",323 "archetype": "regular",324 "unseen": false,325 "pinned": false,326 "unpinned": null,327 "visible": true,328 "closed": false,329 "archived": false,330 "bookmarked": null,331 "liked": null,332 "tags_descriptions": {},333 "like_count": 0,334 "views": 64,335 "category_id": 1,336 "featured_link": null,337 "has_accepted_answer": false,338 "posters": [339 {340 "extras": "latest single",341 "description": "Original Poster, Most Recent Poster",342 "user": {343 "id": 81605,344 "username": "benjamin-perry-duke",345 "name": "Ben Perry",346 "avatar_template": "/user_avatar/discuss.pytorch.org/benjamin-perry-duke/{size}/74629_2.png",347 "trust_level": 1348 }349 }350 ]351 },352 {353 "fancy_title": "Pytorch data loading best practices, any good resources to explore?",354 "id": 220778,355 "title": "Pytorch data loading best practices, any good resources to explore?",356 "slug": "pytorch-data-loading-best-practices-any-good-resources-to-explore",357 "posts_count": 2,358 "reply_count": 0,359 "highest_post_number": 2,360 "image_url": null,361 "created_at": "2025-06-12T21:53:47.819Z",362 "last_posted_at": "2025-06-13T00:35:24.341Z",363 "bumped": true,364 "bumped_at": "2025-06-13T00:35:24.341Z",365 "archetype": "regular",366 "unseen": false,367 "pinned": false,368 "unpinned": null,369 "visible": true,370 "closed": false,371 "archived": false,372 "bookmarked": null,373 "liked": null,374 "tags_descriptions": {},375 "like_count": 0,376 "views": 40,377 "category_id": 1,378 "featured_link": null,379 "has_accepted_answer": false,380 "posters": [381 {382 "extras": null,383 "description": "Original Poster",384 "user": {385 "id": 84681,386 "username": "Ptuser400",387 "name": "",388 "avatar_template": "/letter_avatar_proxy/v4/letter/p/f19dbf/{size}.png",389 "trust_level": 0390 }391 },392 {393 "extras": "latest",394 "description": "Most Recent Poster",395 "user": {396 "id": 3534,397 "username": "ptrblck",398 "name": "",399 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",400 "admin": true,401 "moderator": true,402 "trust_level": 2403 }404 }405 ]406 }407 ],408 "tags_descriptions": {},409 "fancy_title": "Impact of PCIe lane configuration on multi GPU training and inference",410 "id": 222946,411 "title": "Impact of PCIe lane configuration on multi GPU training and inference",412 "posts_count": 2,413 "created_at": "2025-09-08T17:25:32.561Z",414 "views": 116,415 "reply_count": 0,416 "like_count": 2,417 "last_posted_at": "2025-09-14T14:15:18.933Z",418 "visible": true,419 "closed": false,420 "archived": false,421 "has_summary": false,422 "archetype": "regular",423 "slug": "impact-of-pcie-lane-configuration-on-multi-gpu-training-and-inference",424 "category_id": 1,425 "word_count": 447,426 "deleted_at": null,427 "user_id": 50461,428 "featured_link": null,429 "pinned_globally": false,430 "pinned_at": null,431 "pinned_until": null,432 "image_url": null,433 "slow_mode_seconds": 0,434 "draft": null,435 "draft_key": "topic_222946",436 "draft_sequence": null,437 "unpinned": null,438 "pinned": false,439 "current_post_number": 1,440 "highest_post_number": 2,441 "deleted_by": null,442 "actions_summary": [443 {444 "id": 4,445 "count": 0,446 "hidden": false,447 "can_act": false448 },449 {450 "id": 8,451 "count": 0,452 "hidden": false,453 "can_act": false454 },455 {456 "id": 10,457 "count": 0,458 "hidden": false,459 "can_act": false460 },461 {462 "id": 7,463 "count": 0,464 "hidden": false,465 "can_act": false466 }467 ],468 "chunk_size": 20,469 "bookmarked": false,470 "topic_timer": null,471 "message_bus_last_id": 0,472 "participant_count": 2,473 "show_read_indicator": false,474 "thumbnails": null,475 "slow_mode_enabled_until": null,476 "can_vote": false,477 "vote_count": 0,478 "user_voted": false,479 "discourse_zendesk_plugin_zendesk_id": null,480 "discourse_zendesk_plugin_zendesk_url": "https://your-url.zendesk.com/agent/tickets/",481 "details": {482 "can_edit": false,483 "notification_level": 1,484 "participants": [485 {486 "id": 30062,487 "username": "tonyr",488 "name": "Tony Robinson",489 "avatar_template": "/user_avatar/discuss.pytorch.org/tonyr/{size}/74253_2.png",490 "post_count": 1,491 "primary_group_name": null,492 "flair_name": null,493 "flair_url": null,494 "flair_color": null,495 "flair_bg_color": null,496 "flair_group_id": null,497 "trust_level": 2498 },499 {500 "id": 50461,501 "username": "adrew25",502 "name": "Ανδρεας Κατσινουλας",503 "avatar_template": "/user_avatar/discuss.pytorch.org/adrew25/{size}/43719_2.png",504 "post_count": 1,505 "primary_group_name": null,506 "flair_name": null,507 "flair_url": null,508 "flair_color": null,509 "flair_bg_color": null,510 "flair_group_id": null,511 "trust_level": 1512 }513 ],514 "created_by": {515 "id": 50461,516 "username": "adrew25",517 "name": "Ανδρεας Κατσινουλας",518 "avatar_template": "/user_avatar/discuss.pytorch.org/adrew25/{size}/43719_2.png"519 },520 "last_poster": {521 "id": 30062,522 "username": "tonyr",523 "name": "Tony Robinson",524 "avatar_template": "/user_avatar/discuss.pytorch.org/tonyr/{size}/74253_2.png"525 },526 "links": [527 {528 "url": "https://woc.co.uk/",529 "title": "World of Computers Ltd",530 "internal": false,531 "attachment": false,532 "reflection": false,533 "clicks": 1,534 "user_id": 30062,535 "domain": "woc.co.uk",536 "root_domain": "woc.co.uk"537 }538 ]539 },540 "bookmarks": []541 },542 {543 "post_stream": {544 "posts": [545 {546 "id": 475348,547 "name": "",548 "username": "codeguru",549 "avatar_template": "/letter_avatar_proxy/v4/letter/c/45deac/{size}.png",550 "created_at": "2025-09-12T22:49:57.932Z",551 "cooked": "<p>I have two sets of implementations in C++CUDA. Both implementations require include files with clashing definitions.</p>\n<p>In CMake, I can do the following (please ignore minor mistake – making a case):</p>\n<pre data-code-wrap=\"makefile\"><code class=\"lang-makefile\"># first object\nadd_library(FirstObject OBJECT src/file1.cpp src/file2.cu)\ntarget_include_directories(FirstObject PRIVATE include_dir1)\ntarget_set_properties(FirstObject PROPERTY POSITION_INDEPENDENT_CODE ON)\ntarget_compile_options(FirstObject PRIVATE custom_options1)\n\n# second object\nadd_library(SecondObject OBJECT src/file3.cpp src/file4.cu)\ntarget_include_directories(SecondObject PRIVATE include_dir2)\ntarget_set_properties(SecondObject PROPERTY POSITION_INDEPENDENT_CODE ON)\ntarget_compile_options(SecondObject PRIVATE custom_options2)\n\n# final shared lib\nadd_library(FinalSharedLib SHARED $<TARGET_OBJECTS:FirstObject> $<TARGET_OBJECTS:SecondObject> shared_library.cpp)\n...\n</code></pre>\n<p>I want to do the same in Setup tools. The best solution I came across was writing custom build class, based on <code>build_ext</code>, but could not extend it to use CudaExtension.</p>\n<p>An existing example will be helpful.</p>",552 "post_number": 1,553 "post_type": 1,554 "posts_count": 2,555 "updated_at": "2025-09-12T23:00:30.114Z",556 "reply_count": 0,557 "reply_to_post_number": null,558 "quote_count": 0,559 "incoming_link_count": 1,560 "reads": 6,561 "readers_count": 5,562 "score": 6.2,563 "yours": false,564 "topic_id": 223066,565 "topic_slug": "c-cuda-extension-build-object-files-first-then-shared-lib-extension",566 "display_username": "",567 "primary_group_name": null,568 "flair_name": null,569 "flair_url": null,570 "flair_bg_color": null,571 "flair_color": null,572 "flair_group_id": null,573 "badges_granted": [],574 "version": 2,575 "can_edit": false,576 "can_delete": false,577 "can_recover": false,578 "can_see_hidden_post": false,579 "can_wiki": false,580 "read": true,581 "user_title": null,582 "bookmarked": false,583 "actions_summary": [],584 "moderator": false,585 "admin": false,586 "staff": false,587 "user_id": 78888,588 "hidden": false,589 "trust_level": 1,590 "deleted_at": null,591 "user_deleted": false,592 "edit_reason": null,593 "can_view_edit_history": true,594 "wiki": false,595 "post_url": "/t/c-cuda-extension-build-object-files-first-then-shared-lib-extension/223066/1",596 "can_accept_answer": false,597 "can_unaccept_answer": false,598 "accepted_answer": false,599 "topic_accepted_answer": null,600 "can_vote": false601 },602 {603 "id": 475462,604 "name": "",605 "username": "codeguru",606 "avatar_template": "/letter_avatar_proxy/v4/letter/c/45deac/{size}.png",607 "created_at": "2025-09-16T20:17:39.194Z",608 "cooked": "<p>I solved this by subclassing BuildExtension. This works very well for C++ files. For CUDA files, I simply invoke NVCC compile commands via Python Subprocess lib.</p>\n<p>Better solution would be to set compiler to NVCC and then use the same interface as C++ files. But this would need additional settings.</p>",609 "post_number": 2,610 "post_type": 1,611 "posts_count": 2,612 "updated_at": "2025-09-16T20:17:39.194Z",613 "reply_count": 0,614 "reply_to_post_number": null,615 "quote_count": 0,616 "incoming_link_count": 0,617 "reads": 2,618 "readers_count": 1,619 "score": 0.4,620 "yours": false,621 "topic_id": 223066,622 "topic_slug": "c-cuda-extension-build-object-files-first-then-shared-lib-extension",623 "display_username": "",624 "primary_group_name": null,625 "flair_name": null,626 "flair_url": null,627 "flair_bg_color": null,628 "flair_color": null,629 "flair_group_id": null,630 "badges_granted": [],631 "version": 1,632 "can_edit": false,633 "can_delete": false,634 "can_recover": false,635 "can_see_hidden_post": false,636 "can_wiki": false,637 "read": true,638 "user_title": null,639 "bookmarked": false,640 "actions_summary": [],641 "moderator": false,642 "admin": false,643 "staff": false,644 "user_id": 78888,645 "hidden": false,646 "trust_level": 1,647 "deleted_at": null,648 "user_deleted": false,649 "edit_reason": null,650 "can_view_edit_history": true,651 "wiki": false,652 "post_url": "/t/c-cuda-extension-build-object-files-first-then-shared-lib-extension/223066/2",653 "can_accept_answer": false,654 "can_unaccept_answer": false,655 "accepted_answer": false,656 "topic_accepted_answer": null657 }658 ],659 "stream": [660 475348,661 475462662 ]663 },664 "timeline_lookup": [665 [666 1,667 43668 ],669 [670 2,671 39672 ]673 ],674 "suggested_topics": [675 {676 "fancy_title": "Extracting Individual Tensors from a PyTorch .pt File for Custom CUDA/C++ Inference",677 "id": 213219,678 "title": "Extracting Individual Tensors from a PyTorch .pt File for Custom CUDA/C++ Inference",679 "slug": "extracting-individual-tensors-from-a-pytorch-pt-file-for-custom-cuda-c-inference",680 "posts_count": 1,681 "reply_count": 0,682 "highest_post_number": 1,683 "image_url": null,684 "created_at": "2024-11-20T17:13:34.875Z",685 "last_posted_at": "2024-11-20T17:13:34.940Z",686 "bumped": true,687 "bumped_at": "2024-11-20T17:13:34.940Z",688 "archetype": "regular",689 "unseen": false,690 "pinned": false,691 "unpinned": null,692 "visible": true,693 "closed": false,694 "archived": false,695 "bookmarked": null,696 "liked": null,697 "tags_descriptions": {},698 "like_count": 0,699 "views": 40,700 "category_id": 11,701 "featured_link": null,702 "has_accepted_answer": false,703 "posters": [704 {705 "extras": "latest single",706 "description": "Original Poster, Most Recent Poster",707 "user": {708 "id": 81026,709 "username": "Saad_Amin",710 "name": "Saad Amin",711 "avatar_template": "/user_avatar/discuss.pytorch.org/saad_amin/{size}/73451_2.png",712 "trust_level": 1713 }714 }715 ]716 },717 {718 "fancy_title": "Resnet18 behaving differently in libtorch than pytorch",719 "id": 218041,720 "title": "Resnet18 behaving differently in libtorch than pytorch",721 "slug": "resnet18-behaving-differently-in-libtorch-than-pytorch",722 "posts_count": 2,723 "reply_count": 0,724 "highest_post_number": 2,725 "image_url": null,726 "created_at": "2025-03-19T21:49:56.247Z",727 "last_posted_at": "2025-04-09T11:40:25.261Z",728 "bumped": true,729 "bumped_at": "2025-04-09T11:40:25.261Z",730 "archetype": "regular",731 "unseen": false,732 "pinned": false,733 "unpinned": null,734 "visible": true,735 "closed": false,736 "archived": false,737 "bookmarked": null,738 "liked": null,739 "tags_descriptions": {},740 "like_count": 0,741 "views": 59,742 "category_id": 11,743 "featured_link": null,744 "has_accepted_answer": false,745 "posters": [746 {747 "extras": null,748 "description": "Original Poster",749 "user": {750 "id": 83366,751 "username": "Youssef_Mohamed",752 "name": "Youssef Mohamed",753 "avatar_template": "/user_avatar/discuss.pytorch.org/youssef_mohamed/{size}/76251_2.png",754 "trust_level": 0755 }756 },757 {758 "extras": "latest",759 "description": "Most Recent Poster",760 "user": {761 "id": 55864,762 "username": "azhanmohammed",763 "name": "Azhan Mohammed",764 "avatar_template": "/user_avatar/discuss.pytorch.org/azhanmohammed/{size}/49537_2.png",765 "trust_level": 2766 }767 }768 ]769 },770 {771 "fancy_title": "Seeking Future Support for C++ Logging in torch_logging.set_logs: Any Plans?",772 "id": 215737,773 "title": "Seeking Future Support for C++ Logging in torch_logging.set_logs: Any Plans?",774 "slug": "seeking-future-support-for-c-logging-in-torch-logging-set-logs-any-plans",775 "posts_count": 1,776 "reply_count": 0,777 "highest_post_number": 1,778 "image_url": null,779 "created_at": "2025-01-23T01:31:25.612Z",780 "last_posted_at": "2025-01-23T01:31:25.647Z",781 "bumped": true,782 "bumped_at": "2025-01-23T07:04:11.689Z",783 "archetype": "regular",784 "unseen": false,785 "pinned": false,786 "unpinned": null,787 "visible": true,788 "closed": false,789 "archived": false,790 "bookmarked": null,791 "liked": null,792 "tags_descriptions": {},793 "like_count": 0,794 "views": 32,795 "category_id": 11,796 "featured_link": null,797 "has_accepted_answer": false,798 "posters": [799 {800 "extras": "latest single",801 "description": "Original Poster, Most Recent Poster",802 "user": {803 "id": 82250,804 "username": "jokercw147",805 "name": "Jokercw147",806 "avatar_template": "/user_avatar/discuss.pytorch.org/jokercw147/{size}/75245_2.png",807 "trust_level": 0808 }809 }810 ]811 },812 {813 "fancy_title": "Tensor subtraction",814 "id": 216053,815 "title": "Tensor subtraction",816 "slug": "tensor-subtraction",817 "posts_count": 4,818 "reply_count": 2,819 "highest_post_number": 4,820 "image_url": null,821 "created_at": "2025-01-30T14:25:47.879Z",822 "last_posted_at": "2025-01-31T07:24:41.863Z",823 "bumped": true,824 "bumped_at": "2025-01-31T07:24:41.863Z",825 "archetype": "regular",826 "unseen": false,827 "pinned": false,828 "unpinned": null,829 "visible": true,830 "closed": false,831 "archived": false,832 "bookmarked": null,833 "liked": null,834 "tags_descriptions": {},835 "like_count": 1,836 "views": 183,837 "category_id": 11,838 "featured_link": null,839 "has_accepted_answer": false,840 "posters": [841 {842 "extras": "latest",843 "description": "Original Poster, Most Recent Poster",844 "user": {845 "id": 73501,846 "username": "Dirk10001",847 "name": null,848 "avatar_template": "/letter_avatar_proxy/v4/letter/d/c37758/{size}.png",849 "trust_level": 1850 }851 },852 {853 "extras": null,854 "description": "Frequent Poster",855 "user": {856 "id": 41458,857 "username": "J_Johnson",858 "name": "J Johnson",859 "avatar_template": "/user_avatar/discuss.pytorch.org/j_johnson/{size}/55494_2.png",860 "trust_level": 2861 }862 }863 ]864 },865 {866 "fancy_title": "Reducing PyTorch C++ API Library Size for Inference-Only Use",867 "id": 220282,868 "title": "Reducing PyTorch C++ API Library Size for Inference-Only Use",869 "slug": "reducing-pytorch-c-api-library-size-for-inference-only-use",870 "posts_count": 2,871 "reply_count": 0,872 "highest_post_number": 2,873 "image_url": null,874 "created_at": "2025-05-23T14:50:16.708Z",875 "last_posted_at": "2025-05-26T14:32:56.153Z",876 "bumped": true,877 "bumped_at": "2025-05-26T14:32:56.153Z",878 "archetype": "regular",879 "unseen": false,880 "pinned": false,881 "unpinned": null,882 "visible": true,883 "closed": false,884 "archived": false,885 "bookmarked": null,886 "liked": null,887 "tags_descriptions": {},888 "like_count": 1,889 "views": 114,890 "category_id": 11,891 "featured_link": null,892 "has_accepted_answer": false,893 "posters": [894 {895 "extras": null,896 "description": "Original Poster",897 "user": {898 "id": 32936,899 "username": "fgauthier",900 "name": "",901 "avatar_template": "/user_avatar/discuss.pytorch.org/fgauthier/{size}/25742_2.png",902 "trust_level": 2903 }904 },905 {906 "extras": "latest",907 "description": "Most Recent Poster",908 "user": {909 "id": 82975,910 "username": "zhxchen17",911 "name": "Zhengxu Chen",912 "avatar_template": "/user_avatar/discuss.pytorch.org/zhxchen17/{size}/75908_2.png",913 "trust_level": 1914 }915 }916 ]917 }918 ],919 "tags_descriptions": {},920 "fancy_title": "C++/CUDA extension: build object files first, then shared lib extension",921 "id": 223066,922 "title": "C++/CUDA extension: build object files first, then shared lib extension",923 "posts_count": 2,924 "created_at": "2025-09-12T22:49:57.871Z",925 "views": 30,926 "reply_count": 0,927 "like_count": 0,928 "last_posted_at": "2025-09-16T20:17:39.194Z",929 "visible": true,930 "closed": false,931 "archived": false,932 "has_summary": false,933 "archetype": "regular",934 "slug": "c-cuda-extension-build-object-files-first-then-shared-lib-extension",935 "category_id": 11,936 "word_count": 179,937 "deleted_at": null,938 "user_id": 78888,939 "featured_link": null,940 "pinned_globally": false,941 "pinned_at": null,942 "pinned_until": null,943 "image_url": null,944 "slow_mode_seconds": 0,945 "draft": null,946 "draft_key": "topic_223066",947 "draft_sequence": null,948 "unpinned": null,949 "pinned": false,950 "current_post_number": 1,951 "highest_post_number": 2,952 "deleted_by": null,953 "actions_summary": [954 {955 "id": 4,956 "count": 0,957 "hidden": false,958 "can_act": false959 },960 {961 "id": 8,962 "count": 0,963 "hidden": false,964 "can_act": false965 },966 {967 "id": 10,968 "count": 0,969 "hidden": false,970 "can_act": false971 },972 {973 "id": 7,974 "count": 0,975 "hidden": false,976 "can_act": false977 }978 ],979 "chunk_size": 20,980 "bookmarked": false,981 "topic_timer": null,982 "message_bus_last_id": 0,983 "participant_count": 1,984 "show_read_indicator": false,985 "thumbnails": null,986 "slow_mode_enabled_until": null,987 "can_vote": false,988 "vote_count": 0,989 "user_voted": false,990 "discourse_zendesk_plugin_zendesk_id": null,991 "discourse_zendesk_plugin_zendesk_url": "https://your-url.zendesk.com/agent/tickets/",992 "details": {993 "can_edit": false,994 "notification_level": 1,995 "participants": [996 {997 "id": 78888,998 "username": "codeguru",999 "name": "",1000 "avatar_template": "/letter_avatar_proxy/v4/letter/c/45deac/{size}.png",1001 "post_count": 2,1002 "primary_group_name": null,1003 "flair_name": null,1004 "flair_url": null,1005 "flair_color": null,1006 "flair_bg_color": null,1007 "flair_group_id": null,1008 "trust_level": 11009 }1010 ],1011 "created_by": {1012 "id": 78888,1013 "username": "codeguru",1014 "name": "",1015 "avatar_template": "/letter_avatar_proxy/v4/letter/c/45deac/{size}.png"1016 },1017 "last_poster": {1018 "id": 78888,1019 "username": "codeguru",1020 "name": "",1021 "avatar_template": "/letter_avatar_proxy/v4/letter/c/45deac/{size}.png"1022 }1023 },1024 "bookmarks": []1025 },1026 {1027 "post_stream": {1028 "posts": [1029 {1030 "id": 475418,1031 "name": "Nick Meyne",1032 "username": "mutineering",1033 "avatar_template": "/user_avatar/discuss.pytorch.org/mutineering/{size}/75699_2.png",1034 "created_at": "2025-09-15T13:37:38.026Z",1035 "cooked": "<p>Hello.. I am on ubuntu noble 24.04 with python 3.12 on a desktop machine which has a working installation of CUDA13 and a couple of graphics cards. I can’t use pip to install torch without the ‘externally-managed-environment’ error….. and sudo python3-pytorch gives the ‘no installation candidate’ error. Am I doing something wrong? Or do I just have to wait until an installation candidate appears… or is there something else I can safely do in the meantime?</p>",1036 "post_number": 1,1037 "post_type": 1,1038 "posts_count": 3,1039 "updated_at": "2025-09-15T13:37:38.026Z",1040 "reply_count": 0,1041 "reply_to_post_number": null,1042 "quote_count": 0,1043 "incoming_link_count": 9,1044 "reads": 10,1045 "readers_count": 9,1046 "score": 42.0,1047 "yours": false,1048 "topic_id": 223104,1049 "topic_slug": "installation-candidates",1050 "display_username": "Nick Meyne",1051 "primary_group_name": null,1052 "flair_name": null,1053 "flair_url": null,1054 "flair_bg_color": null,1055 "flair_color": null,1056 "flair_group_id": null,1057 "badges_granted": [],1058 "version": 1,1059 "can_edit": false,1060 "can_delete": false,1061 "can_recover": false,1062 "can_see_hidden_post": false,1063 "can_wiki": false,1064 "read": true,1065 "user_title": null,1066 "bookmarked": false,1067 "actions_summary": [],1068 "moderator": false,1069 "admin": false,1070 "staff": false,1071 "user_id": 85871,1072 "hidden": false,1073 "trust_level": 0,1074 "deleted_at": null,1075 "user_deleted": false,1076 "edit_reason": null,1077 "can_view_edit_history": true,1078 "wiki": false,1079 "post_url": "/t/installation-candidates/223104/1",1080 "can_accept_answer": false,1081 "can_unaccept_answer": false,1082 "accepted_answer": false,1083 "topic_accepted_answer": true,1084 "can_vote": false1085 },1086 {1087 "id": 475423,1088 "name": "",1089 "username": "ptrblck",1090 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png",1091 "created_at": "2025-09-15T20:29:37.240Z",1092 "cooked": "<p>Install a virtual Python environment and the first error should be gone.</p>",1093 "post_number": 2,1094 "post_type": 1,1095 "posts_count": 3,1096 "updated_at": "2025-09-15T20:29:37.240Z",1097 "reply_count": 1,1098 "reply_to_post_number": null,1099 "quote_count": 0,1100 "incoming_link_count": 0,1101 "reads": 7,1102 "readers_count": 6,1103 "score": 21.4,1104 "yours": false,1105 "topic_id": 223104,1106 "topic_slug": "installation-candidates",1107 "display_username": "",1108 "primary_group_name": null,1109 "flair_name": null,1110 "flair_url": null,1111 "flair_bg_color": null,1112 "flair_color": null,1113 "flair_group_id": null,1114 "badges_granted": [],1115 "version": 1,1116 "can_edit": false,1117 "can_delete": false,1118 "can_recover": false,1119 "can_see_hidden_post": false,1120 "can_wiki": false,1121 "read": true,1122 "user_title": "",1123 "bookmarked": false,1124 "actions_summary": [1125 {1126 "id": 2,1127 "count": 11128 }1129 ],1130 "moderator": true,1131 "admin": true,1132 "staff": true,1133 "user_id": 3534,1134 "hidden": false,1135 "trust_level": 2,1136 "deleted_at": null,1137 "user_deleted": false,1138 "edit_reason": null,1139 "can_view_edit_history": true,1140 "wiki": false,1141 "post_url": "/t/installation-candidates/223104/2",1142 "can_accept_answer": false,1143 "can_unaccept_answer": false,1144 "accepted_answer": true,1145 "topic_accepted_answer": true1146 },1147 {1148 "id": 475461,1149 "name": "Nick Meyne",1150 "username": "mutineering",1151 "avatar_template": "/user_avatar/discuss.pytorch.org/mutineering/{size}/75699_2.png",1152 "created_at": "2025-09-16T18:33:28.516Z",1153 "cooked": "<p>All good and up and running. Thanks.</p>\n<pre><code class=\"lang-auto\">import torch\n>>> torch.cuda.is_available()\nTrue\n</code></pre>\n<p><img src=\"https://discuss.pytorch.org/images/emoji/apple/slightly_smiling_face.png?v=14\" title=\":slightly_smiling_face:\" class=\"emoji only-emoji\" alt=\":slightly_smiling_face:\" loading=\"lazy\" width=\"20\" height=\"20\"></p>",1154 "post_number": 3,1155 "post_type": 1,1156 "posts_count": 3,1157 "updated_at": "2025-09-16T18:34:47.874Z",1158 "reply_count": 0,1159 "reply_to_post_number": 2,1160 "quote_count": 0,1161 "incoming_link_count": 0,1162 "reads": 6,1163 "readers_count": 5,1164 "score": 46.2,1165 "yours": false,1166 "topic_id": 223104,1167 "topic_slug": "installation-candidates",1168 "display_username": "Nick Meyne",1169 "primary_group_name": null,1170 "flair_name": null,1171 "flair_url": null,1172 "flair_bg_color": null,1173 "flair_color": null,1174 "flair_group_id": null,1175 "badges_granted": [],1176 "version": 1,1177 "can_edit": false,1178 "can_delete": false,1179 "can_recover": false,1180 "can_see_hidden_post": false,1181 "can_wiki": false,1182 "read": true,1183 "user_title": null,1184 "reply_to_user": {1185 "id": 3534,1186 "username": "ptrblck",1187 "name": "",1188 "avatar_template": "/user_avatar/discuss.pytorch.org/ptrblck/{size}/1823_2.png"1189 },1190 "bookmarked": false,1191 "actions_summary": [1192 {1193 "id": 2,1194 "count": 11195 }1196 ],1197 "moderator": false,1198 "admin": false,1199 "staff": false,1200 "user_id": 85871,