{ "version": "https://jsonfeed.org/version/1", "title": "CUDA", "description": "\u6765\u81ea NVIDIA \u7684\u5e76\u884c\u8fd0\u7b97\u6846\u67b6", "home_page_url": "go/cuda", "feed_url": "feed/cuda.json", "icon": "https://cdn.v2ex.com/navatar/0614/12e4/896_large.png?m=1597325130", "favicon": "https://cdn.v2ex.com/navatar/0614/12e4/896_normal.png?m=1597325130", "items": [ { "author": { "url": "member/zoe1016aaa", "name": "zoe1016aaa", "avatar": "https://cdn.v2ex.com/avatar/343f/f1da/447597_large.png?m=1769069846" }, "url": "t/1082256", "title": "[\u82f1\u4f1f\u8fbe] [\u5317\u4eac\u6216\u8005\u4e0a\u6d77] [\u6df1\u5ea6\u5b66\u4e60\u6027\u80fd\u4f18\u5316-CUDA]", "id": "t/1082256", "date_published": "2024-10-21T09:05:57+00:00", "content_html": "

\u6211\u4eec\u76ee\u524d\u6b63\u5728\u5bfb\u627e\u4e00\u540d\u6df1\u5ea6\u5b66\u4e60\u6027\u80fd\u8f6f\u4ef6\u5de5\u7a0b\u5e08\uff01\u6211\u4eec\u6b63\u5728\u6269\u5c55\u6211\u4eec\u7684\u63a8\u7406\u7814\u7a76\u4e0e\u5f00\u53d1\u3002\u6211\u4eec\u5bfb\u6c42\u4f18\u79c0\u7684\u8f6f\u4ef6\u5de5\u7a0b\u5e08\u548c\u9ad8\u7ea7\u8f6f\u4ef6\u5de5\u7a0b\u5e08\u52a0\u5165\u6211\u4eec\u7684\u56e2\u961f\u3002\u6211\u4eec\u4e13\u6ce8\u4e8e\u5f00\u53d1 GPU \u52a0\u901f\u7684\u6df1\u5ea6\u5b66\u4e60\u8f6f\u4ef6\u3002\u5168\u7403\u7684\u7814\u7a76\u4eba\u5458\u6b63\u5728\u4f7f\u7528 NVIDIA GPU \u63a8\u52a8\u6df1\u5ea6\u5b66\u4e60\u7684\u9769\u547d\uff0c\u8fd9\u5728\u4f17\u591a\u9886\u57df\u5b9e\u73b0\u4e86\u7a81\u7834\u3002\u52a0\u5165\u6211\u4eec\u7684\u56e2\u961f\uff0c\u6784\u5efa\u4f7f\u65b0\u89e3\u51b3\u65b9\u6848\u6210\u4e3a\u53ef\u80fd\u7684\u8f6f\u4ef6\u3002\u4e0e\u6df1\u5ea6\u5b66\u4e60\u793e\u533a\u5408\u4f5c\uff0c\u5728 Tensor-RT \u4e2d\u5b9e\u73b0\u6700\u65b0\u7b97\u6cd5\u7684\u516c\u5f00\u53d1\u5e03\u3002\u6211\u4eec\u9700\u8981\u4f60\u80fd\u591f\u5728\u5feb\u8282\u594f\u3001\u4ee5\u5ba2\u6237\u4e3a\u4e2d\u5fc3\u7684\u56e2\u961f\u4e2d\u5de5\u4f5c\uff0c\u5e76\u4e14\u5177\u5907\u51fa\u8272\u7684\u6c9f\u901a\u6280\u5de7\u3002

\n

\u4f60\u5c06\u8981\u505a\u7684\u5de5\u4f5c\u5305\u62ec\uff1a

\n\n

\u6211\u4eec\u5e0c\u671b\u770b\u5230\u7684\u8d44\u8d28\uff1a

\n\n

\u5982\u679c\u611f\u5174\u8da3\u8bf7\u8054\u7cfb\uff1a

\n

\u5fae\u4fe1\uff1a18867144803\n\u7b80\u5386\u6295\u9012\uff1a xiaozhao@nvidia.com

\n

\u8bf7\u5907\u6ce8\u6295\u9012\u7684\u5c97\u4f4d\u65b9\u5411\u5982\uff1a\u59d3\u540d+\u6df1\u5ea6\u5b66\u4e60\u6027\u80fd\u4f18\u5316

\n" }, { "author": { "url": "member/cwjwgg", "name": "cwjwgg", "avatar": "https://cdn.v2ex.com/gravatar/5bf0a4af3e7eca9f77152858ac3dfa26?s=73&d=retro" }, "url": "t/1050060", "title": "\u8bad\u7ec3 SVC \u58f0\u97f3\u6a21\u578b 2060 12g \u548c 8G \u7684 3060TI \u54ea\u4e2a\u5feb", "id": "t/1050060", "date_published": "2024-06-17T01:41:01+00:00", "content_html": "\u6211\u770b 2060 \u62e5\u6709 1920 \u4e2a CUDA \u6838\u5fc3 \u4f46 12G
3060TI \u642d\u8f7d\u4e86 4864 \u4e2a CUDA \u6838 8G
\u5e94\u8be5\u662f 3060TI \u66f4\u5feb\u5427 \u80fd\u5757\u591a\u5c11\u5462" }, { "author": { "url": "member/luckyc", "name": "luckyc", "avatar": "https://cdn.v2ex.com/avatar/4105/08ab/73070_large.png?m=1762737894" }, "url": "t/998477", "title": "win11 \u4f7f\u7528 cuda \u8c03\u7528\u4e24\u4e2a gpu \u8ba1\u7b97\u65f6\uff0c\u81ea\u5e26\u4efb\u52a1\u7ba1\u7406\u5668\u770b\u4e0d\u5230 gpu2 \u7684\u4f7f\u7528\u7387\uff1f", "id": "t/998477", "date_published": "2023-12-07T12:53:35+00:00", "content_html": "\u7528 nvdia-smi \u53ef\u770b\u5230\u529f\u8017\u660e\u663e\u589e\u52a0\u4e86\uff0c
\u7528 gpu-z \u4e5f\u53ef\u4ee5\u770b\u5230

\u662f\u4e0d\u662f win11 \u7684 bug \uff1f" }, { "author": { "url": "member/TimeNewRome", "name": "TimeNewRome", "avatar": "https://cdn.v2ex.com/gravatar/06d23ac29662f44881e2f64ea971973d?s=73&d=retro" }, "url": "t/990897", "date_modified": "2023-11-11T03:57:35+00:00", "content_html": "\u201cstatic inline CUresult cuGetProcAddress_v2_ptsz(const char *symbol, void **funcPtr, int driverVersion, cuuint64_t flags, CUdriverProcAddressQueryResult *symbolStatus) {
const int procAddressMask = (CU_GET_PROC_ADDRESS_LEGACY_STREAM|
CU_GET_PROC_ADDRESS_PER_THREAD_DEFAULT_STREAM);
if ((flags & procAddressMask) == 0) {
flags |= CU_GET_PROC_ADDRESS_PER_THREAD_DEFAULT_STREAM;
}
return cuGetPr\u201d

\u8fd9\u4e2a\u5c31\u662f\u65b0\u589e\u7684\u51fd\u6570\u7ed3\u6784\u3002\u4f3c\u4e4e\u8ddf\u4e4b\u524d\u7684 cuGetProcAddress \u51fd\u6570\u5dee\u4e0d\u591a\uff0c\u53ea\u662f\u65b0\u589e\u4e86 CUdriverProcAddressQueryResult \u8fd9\u4e2a\u7ed3\u6784\u4f53\u3002\u8bf7\u95ee\u8fd9\u4e2a\u51fd\u6570\u8be5\u5982\u4f55\u52ab\u6301\u5462\uff1f", "date_published": "2023-11-11T03:41:57+00:00", "title": "[cuda \u51fd\u6570\u52ab\u6301] cuda12.2 \u7248\u672c\u65b0\u589e\u4e86\u4e00\u4e2a\u51fd\u6570 cuGetProcAddress_v2\uff0c\u8bf7\u95ee\u5982\u4f55\u8fdb\u884c\u52ab\u6301\uff1f", "id": "t/990897" }, { "author": { "url": "member/GopherDaily", "name": "GopherDaily", "avatar": "https://cdn.v2ex.com/gravatar/89208b8c3493547fe544b2d5142dc131?s=73&d=retro" }, "url": "t/959842", "title": "Set Max_split_size_mb To Avoid Oom In Pytorch", "id": "t/959842", "date_published": "2023-07-26T04:25:01+00:00", "content_html": "

\u6211\u4eec\u5728 k8s \u4e2d\u90e8\u7f72\u4e86 stable-diffusion-webui\n\u4f9b\u4efb\u4f55\u60f3\u8981\u4f53\u9a8c\u7684 Stable Diffusion Model \u7684\u7528\u6237\u4f7f\u7528.\n\u968f\u7740\u4e00\u4e2a\u53c8\u4e00\u4e2a\u7684\u8bf7\u6c42, \u6211\u4eec\u9891\u7e41\u7684\u9047\u5230 CUDA \u7684 OOM \u9519\u8bef.\n\u5176\u4e2d\u7684\u4e00\u5c0f\u90e8\u5206\u786e\u5b9e\u662f\u56e0\u4e3a\u7528\u6237\u8bf7\u6c42\u9700\u8981\u7684\u8d44\u6e90\u8d85\u8fc7\u4e86\u5bf9\u5e94 GPU \u80fd\u591f\u63d0\u4f9b\u7684\u5185\u5b58.

\n

\u5269\u4e0b\u7684, \u5360\u5927\u90e8\u5206\u7684, \u662f\u7c7b\u4f3c\u5982\u4e0b\u7684\u4ee4\u4eba\u56f0\u60d1\u7684\u573a\u666f.

\n
{\"error\": \"OutOfMemoryError\", \"detail\": \"\", \"body\": \"\", \"errors\": \"CUDA out of memory. Tried to allocate 1024.00 MiB (GPU 0; 11.76 GiB total capacity; 7.92 GiB already allocated; 784.31 MiB free; 10.63 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF\"}\n
\n

\u6839\u636e\u5bf9 memory_stats \u7684\u7406\u89e3:

\n\n

\u8fd9\u90e8\u5206\u5185\u5b58\u53bb\u54ea\u513f\u4e86\u5462? \u4e3a\u4ec0\u4e48\u5728\u7528\u6237\u7533\u8bf7\u7684\u65f6\u5019\u4f9d\u7136\u6ca1\u6709\u88ab\u56de\u6536\u5462?

\n

pytorch \u662f\u5982\u4f55\u5206\u914d\u5185\u5b58\u7684?.

\n

\u5f53\u7528\u6237\u8bf7\u6c42\u5185\u5b58\u65f6, pytorch \u7684\u5904\u7406\u6d41\u7a0b\u53ef\u4ee5\u7b80\u5316\u4e3a:

\n
    \n
  1. \u5c1d\u8bd5\u901a\u8fc7 get_free_block \u53bb\u5bfb\u627e\u6ee1\u8db3\u8981\u6c42\u7684\u7a7a\u95f2 Block
    2. \n
  2. \u5982\u679c\u5931\u8d25, \u5219\u901a\u8fc7 trigger_free_memory_callbacks \u53bb\u56de\u6536\u5df2\u5206\u914d\u4f46\u4e0d\u518d\u4f7f\u7528\u7684 Block \u540e, \u518d\u6b21\u5c1d\u8bd5 get_free_block
    3. \n
  3. \u5982\u679c\u5931\u8d25, \u5219\u901a\u8fc7 alloc_block \u53bb\u5411 GPU \u7533\u8bf7\u65b0\u7684 Block
    4. \n
  4. \u5982\u679c\u5931\u8d25, \u5219\u901a\u8fc7 release_available_cached_blocks \u5c06\u5df2\u7533\u8bf7\u4f46\u672a\u5206\u914d\u7684 Block \u91ca\u653e\u540e\u518d\u6b21\u5c1d\u8bd5 alloc_block
    5. \n
  5. \u5982\u679c\u5931\u8d25, \u5219\u901a\u8fc7 release_cached_blocks \u5c06\u6240\u6709\u5df2\u7533\u8bf7\u4f46\u672a\u5206\u914d\u7684 Block \u91ca\u653e, \u518d\u6b21\u5c1d\u8bd5 alloc_block
    6. \n
\n

\u6211\u4eec\u6ce8\u610f\u5230 pytorch \u5411 GPU \u7533\u8bf7\u548c\u5206\u914d\u7ed9\u7528\u6237\u7684\u5185\u5b58\u90fd\u4ee5 Block \u4e3a\u5355\u4f4d.\npytorch \u5411 GPU \u7533\u8bf7\u7684 Block \u5927\u5c0f\u5e76\u4e0d\u56fa\u5b9a, \u53d7\u5f53\u65f6\u7528\u6237\u8bf7\u6c42\u5185\u5b58\u5927\u5c0f\u7684\u5f71\u54cd.\n\u7528\u6237\u91ca\u653e\u5185\u5b58\u540e, Block \u8fd4\u56de\u7ed9 pytorch \u5e76\u6210\u4e3a\u7a7a\u95f2\u72b6\u6001.\n\u7528\u6237\u4e0b\u6b21\u7533\u8bf7\u65f6\u4f18\u5148\u4f1a\u590d\u7528\u7a7a\u95f2 Block, \u800c\u4e0d\u662f\u76f4\u63a5\u5411 GPU \u7533\u8bf7.

\n

\u5982\u679c\u7528\u6237\u7533\u8bf7\u7684\u5185\u5b58\u5927\u5c0f\u5c0f\u4e8e\u6ee1\u8db3\u8981\u6c42\u7684\u7a7a\u95f2 Block, pytorch \u4f1a\u8fdb\u884c\u4e00\u6b21 split \u64cd\u4f5c.\n\u5c06 Block \u5206\u5272\u6210\u4e24\u4e2a Block, \u9664\u53bb\u7528\u6237\u8bf7\u6c42\u5927\u5c0f\u7684\u5185\u5b58\u4f1a\u88ab\u5206\u5272\u6210\u4e00\u4e2a\u72ec\u7acb\u7684 Block,\n\u7559\u5f85\u540e\u7528\u5e76\u901a\u8fc7\u53cc\u5411\u94fe\u8868\u548c\u5206\u914d\u7ed9\u7528\u6237\u7684 Block \u76f8\u5173\u8054.

\n

trigger_free_memory_callbacks \u7684\u56de\u6536\u8fc7\u7a0b\u4f1a\u5c06\u76f8\u90bb\u7684\u7a7a\u95f2 Block \u5408\u5e76, \u63d0\u9ad8\u540e\u7eed\u5206\u914d\u7684\u7075\u6d3b\u6027.

\n

\u76f8\u8f83\u4e8e\u5176\u4ed6\u5185\u5b58\u7ba1\u7406\u673a\u5236, pytorch \u7684\u5185\u5b58\u7ba1\u7406\u76f8\u5bf9\u7b80\u7565:

\n\n

\u4e0a\u8ff0\u7684\u4e24\u70b9, \u9020\u6210\u4e86 pytorch \u53ef\u80fd\u56e0\u4e3a Block \u788e\u7247\u5316, \u5bfc\u81f4\u5927\u91cf\u5185\u5b58\u65e0\u6cd5\u88ab\u4f7f\u7528.

\n

\u5047\u8bbe\u5728\u67d0\u6b21\u5206\u914d\u5185\u5b58\u65f6, pytorch \u6839\u636e\u7528\u6237\u8bf7\u6c42\u5411 GPU \u7533\u8bf7\u4e86\u4e00\u4e2a 256M \u7684 Block.
\n<-------------------------- 256M ----------------------------->

\n

\u7ecf\u8fc7\u591a\u6b21\u5206\u914d\u548c\u56de\u6536, \u5176\u4f7f\u7528\u60c5\u51b5\u53ef\u80fd\u53d8\u6210\u5982\u4e0b.
\n<-- 28M(allocated) --><-- 100M(free) --><-- 28M(allocated) --><-- 100M(free) -->

\n

\u6b64\u65f6\u5982\u679c\u7528\u6237\u7533\u8bf7 160M \u5185\u5b58:

\n\n

max_split_size_mb \u7684\u4f5c\u7528

\n

max_split_size_mb \u7684\u4f5c\u7528\u5728\u4e8e\u7981\u6b62 pytorch \u5bf9\u4efb\u4f55\u5927\u4e8e\u8be5\u5927\u5c0f\u7684 Block \u8fdb\u884c\u5206\u5272\u64cd\u4f5c, \u4ece\u800c\u63a7\u5236\u788e\u7247\u5316\u7684\u7a0b\u5ea6.\n\u6211\u4eec\u4e0a\u6587\u8bb2\u8bc9\u7684\u90fd\u662f\u5728\u672a\u4e3b\u52a8\u8bbe\u7f6e max_split_size_mb \u7684\u60c5\u51b5\u4e0b\u7684\u903b\u8f91, \u6b64\u65f6 max_split_size_mb \u53d6\u9ed8\u8ba4\u503c MAX_INT.

\n

\u6211\u4eec\u5e76\u6ca1\u6709\u627e\u5230\u5b98\u65b9\u63a8\u8350\u7684 max_split_size_mb, \u6211\u4eec\u4e5f\u4e0d\u719f\u6089 pytorch \u548c nvida, \u5f88\u96be\u7ed9\u51fa\u4e00\u4e2a\u5f88\u597d\u7684\u63a8\u8350\u503c.\n\u4ece\u5b9e\u9645\u4f7f\u7528\u6765\u548c\u76f4\u89c2\u903b\u8f91\u6765\u8bf4, 128/256/512 \u4e4b\u7c7b\u7684\u503c\u90fd\u662f\u53ef\u9009\u7684, \u5207\u5b9e\u7684\u907f\u514d\u4e86 OOM, \u4e5f\u6ca1\u6709\u5bfc\u81f4\u660e\u663e\u7684\u6027\u80fd\u8d1f\u62c5.

\n

garbage_collection_threshold

\n

pytorch \u9ed8\u8ba4\u4ec5\u5728\u65e0\u6cd5\u83b7\u53d6\u5230\u5408\u9002\u7684\u7a7a\u95f2 Block \u65f6\u89e6\u53d1\u56de\u6536,\n\u8fd9\u4e2a\u503c\u53ef\u4ee5\u63a7\u5236\u5f53 allocated/capacity \u8d85\u8fc7\u6b64\u503c\u65f6\u89e6\u53d1\u4e3b\u52a8\u7684\u56de\u6536.

\n

Expandable Segments

\n

pytorch \u6700\u65b0(>v2.0.1)\u7684 master \u5206\u652f\u4e2d\u6dfb\u52a0\u4e86 Expandable Segments,\n\u53ef\u80fd\u4e5f\u53ef\u4ee5\u7f13\u89e3\u788e\u7247\u5316\u7684\u95ee\u9898.

\n

References

\n\n" }, { "author": { "url": "member/zoe1016aaa", "name": "zoe1016aaa", "avatar": "https://cdn.v2ex.com/avatar/343f/f1da/447597_large.png?m=1769069846" }, "url": "t/947399", "title": "[\u82f1\u4f1f\u8fbe NVIDIA] [\u4e0a\u6d77/\u5317\u4eac/\u6df1\u5733] [CUDA \u76f8\u5173\u5c97\u4f4d]", "id": "t/947399", "date_published": "2023-06-09T10:09:29+00:00", "content_html": "

[ \u5730\u70b9 ] \uff1aShanghai/Beijing/Shenzhen

\n

[ \u53d1\u9001\u7b80\u5386\u5230 ] :xiaozhao@nvidia.com

\n

[ WeChat \u53ef\u52a0\u5fae\u4fe1 ] \uff1a18867144803

\n

\u4ee3\u7801\u80fd\u529b\u3009\u5de5\u4f5c\u5e74\u9650

\n

Deep Learning Performance Architect-Compiler/LLM-TensorRT

\n

\u4e3b\u8981\u505a\u7684\u662f\u56f4\u7ed5\u6df1\u5ea6\u5b66\u4e60\u7aef\u5230\u7aef\u7684 AI \u8f6f\u4ef6\u5168\u6808\uff0c\u5305\u62ec\u4f46\u4e0d\u9650\u4e8e\u8bad\u7ec3\u6846\u67b6\u3001\u6838\u5fc3\u8ba1\u7b97\u5e93\u3001\u63a8\u7406\u4f18\u5316\u5de5\u5177\uff08\u6bd4\u5982 TensorRT \uff09\uff0cAI \u7f16\u8bd1\u5668\uff0c\u6a21\u578b\u538b\u7f29\u7b49\u5168\u6808\u8f6f\u4ef6\u6808\u3002\u4ee5\u53ca\u53ef\u4ee5\u5728 AI \u8f6f\u4ef6\u5168\u6808\u57fa\u7840\u4e0a\u5f71\u54cd\u5230\u4e0b\u4e00\u4ee3\u751a\u81f3\u4e0b\u4e24\u4ee3\u786c\u4ef6\u67b6\u6784\u7684\u7279\u6027\u8bbe\u8ba1\u3002

\n

Required skills: \u826f\u597d C++\u7f16\u7a0b\uff0c\u719f\u6089 AI \u8f6f\u4ef6\u6808\u5e95\u5c42\u6216\u8005\u8ba1\u7b97\u673a\u4f53\u7cfb\u7ed3\u6784\uff0c\u719f\u6089\u4e0a\u5c42\u7b97\u6cd5\u4e0e Python \u662f\u52a0\u5206\u9879\u3002

\n

\u5730\u70b9\uff1a\u5317\u4eac\u4e0e\u4e0a\u6d77

\n

Deep Learning Performance Architect-TensorRT

\n

\u8d1f\u8d23 NVIDIA \u6df1\u5ea6\u5b66\u4e60\u63a8\u7406\u5f15\u64ce TensorRT \u7684\u8bbe\u8ba1\u3001\u5f00\u53d1\u548c\u7ef4\u62a4\u5de5\u4f5c(e.g. TensorRT \u6a21\u578b\u5bfc\u5165\u7684\u6d41\u7a0b\u548c\u76f8\u5173\u5de5\u5177\uff0c\u56fe\u4f18\u5316\uff0c\u7b97\u5b50\u7684 CUDA \u5b9e\u73b0\u53ca\u4ee3\u7801\u751f\u6210\uff0c\u7b97\u5b50\u6027\u80fd\u4f18\u5316\u7b49)\uff0c\u4ee5\u53ca\u5bf9\u5f53\u524d\u4e3b\u6d41\u7684\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u4f7f\u7528 TensorRT \u8fdb\u884c\u63a8\u7406\u7684\u6027\u80fd\u8fdb\u884c\u5206\u6790\u548c\u4f18\u5316\u3002\u540c\u65f6\uff0c\u8fd8\u5c06\u4e0e NVIDIA GPU \u4f53\u7cfb\u7ed3\u6784\u8bbe\u8ba1\u56e2\u961f\u5408\u4f5c\u6765\u63a8\u52a8 NVIDIA \u6df1\u5ea6\u5b66\u4e60\u89e3\u51b3\u65b9\u6848\u7684\u8f6f\u786c\u4ef6\u534f\u540c\u8bbe\u8ba1\u548c\u7814\u53d1\u3002

\n

\u5c97\u4f4d\u57fa\u672c\u8981\u6c42: \u719f\u7ec3\u638c\u63e1 C++\u7f16\u7a0b

\n

\u5176\u5b83\u5bc6\u5207\u76f8\u5173\u7684\u6280\u80fd /\u7ecf\u9a8c: \u6df1\u5ea6\u5b66\u4e60\u6846\u67b6 /\u6df1\u5ea6\u5b66\u4e60\u7f16\u8bd1\u5668\u5f00\u53d1\uff0c\u6027\u80fd\u5206\u6790 /\u5efa\u6a21 /\u4f18\u5316\u76f8\u5173\u7684\u65b9\u6cd5\u8bba /\u5de5\u5177\uff0c\u8ba1\u7b97\u673a\u4f53\u7cfb\u7ed3\u6784\u76f8\u5173\u77e5\u8bc6\uff0cCUDA kernel \u5f00\u53d1 /\u4f18\u5316

\n

\u5730\u70b9\uff1a\u5317\u4eac\u4e0e\u4e0a\u6d77

\n

Deep Learning Performance Architect-Operator

\n

\u4e3b\u8981\u505a\u7684\u662f\u9488\u5bf9\u4e0d\u540c GPU \u67b6\u6784\u4e3a TensorRT, cuDNN, cuBLAS, cuSPARSE \u7b49\u6df1\u5ea6\u5b66\u4e60\u7b97\u5b50\u5e93\u63d0\u4f9b\u9ad8\u6027\u80fd\u57fa\u7840\u7b97\u5b50\u4ee5\u53ca\u7b97\u5b50\u878d\u5408\u5b9e\u73b0\uff0c\u5305\u542b\u5728\u7ebf\u4ee3\u7801\u751f\u6210\uff0c\u4ee3\u7801\u878d\u5408\u7b49\u76f8\u5173\u5f00\u53d1\u5de5\u4f5c\uff0c\u4ee5\u53ca\u6839\u636e\u5f53\u4ee3 GPU \u4f18\u5316\u74f6\u9888\u5f71\u54cd\u540e\u7eed\u786c\u4ef6\u67b6\u6784\u7279\u5f81\u8bbe\u8ba1\u548c\u9a8c\u8bc1\u5de5\u4f5c\u3002

\n

Required skills: \u826f\u597d C++\u7f16\u7a0b\uff0c\u719f\u6089\u8ba1\u7b97\u673a\u4f53\u7cfb\u7ed3\u6784\uff0c \u6709 TVM, MLIR \u76f8\u5173\u5f00\u53d1\u7ecf\u9a8c\u662f\u52a0\u5206\u9879\u3002

\n

\u5730\u70b9\uff1a\u4e0a\u6d77\u4e0e\u5317\u4eac

\n

Deep Learning Performance Architect

\n

\u4e3b\u8981\u505a\u7684\u662f\u56f4\u7ed5\u8fd0\u7b97\u67b6\u6784\u7684\u5168\u6808\u4f18\u5316\uff0c\u5305\u62ec\u4f46\u4e0d\u9650\u4e8e\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u5206\u6790\u4e0e\u9884\u6d4b\uff0c\u67b6\u6784\u7684\u6027\u80fd\u5206\u6790\uff0c\u7f16\u8bd1\u5668\u6027\u80fd\u5206\u6790\u4ee5\u53ca\u5bf9\u4e3b\u6d41\u8fd0\u7b97\u67b6\u6784\uff0c\u8f6f\u4ef6\u751f\u6001\u7684\u5206\u6790\u3002\u4f7f NVIDIA \u8f6f\u4ef6\u751f\u6001\u4e0e\u8ba1\u7b97\u67b6\u6784\u66f4\u597d\u7684\u652f\u6301\u4e3b\u6d41\u5e94\u7528\u3002

\n

Required skills: \u826f\u597d C++/Python \uff0c\u719f\u6089 AI \u8f6f\u4ef6\u6216\u8005\u8ba1\u7b97\u673a\u4f53\u7cfb\u7ed3\u6784\u3002

\n

\u5730\u70b9\uff1a\u5317\u4eac\u4e0e\u4e0a\u6d77

\n

Developer Technology Engineer-AI

\n

\u5ba2\u6237\u7684\u6df1\u5ea6\u5b66\u4e60\u548c\u9ad8\u80fd\u6027\u8ba1\u7b97\u5e94\u7528\u5728 NVIDIA \u751f\u6001\u4e0a\u7684\u79fb\u690d\u548c\u4f18\u5316\u3002\u8fd9\u4e9b\u5e94\u7528\u5305\u62ec\u5927\u8bed\u8a00\u6a21\u578b\uff0cCV \uff0cSpeech,\u63a8\u8350\u7cfb\u7edf\u548c\u5206\u5b50\u52a8\u529b\u5b66\uff0c\u8ba1\u7b97\u529b\u5b66\uff0c\u8ba1\u7b97\u91cf\u5b50\u5316\u5b66\u7b49\u3002\u901a\u8fc7\u7b97\u6cd5\u548c\u5de5\u7a0b\u4f18\u5316\uff0c\u63d0\u4f9b\u7cfb\u7edf\u7ea7\u7684\u4f18\u5316\u65b9\u6848\u3002\u6df1\u5ea6\u4e0e\u5185\u90e8\u67b6\u6784\u548c\u4ea7\u54c1\u56e2\u961f\u5408\u4f5c\uff0c\u6784\u5efa\u548c\u5b8c\u5584 NVIDIA \u8f6f\u786c\u4ef6\u52a0\u901f\u751f\u6001\u3002

\n

Required skills: Required Skills: \u826f\u597d C/C++\u7f16\u7a0b\u80fd\u529b\uff0c\u5206\u6790\u80fd\u529b\u548c\u6c9f\u901a\u80fd\u529b\uff0c\u719f\u6089\u6df1\u5ea6\u5b66\u4e60\u6216 GPU \u52a0\u901f\u8ba1\u7b97\u8f6f\u4ef6\u6808\uff0c\u624e\u5b9e\u7684\u6df1\u5ea6\u5b66\u4e60\u7406\u8bba\u57fa\u7840\u6216\u7cbe\u901a GPU \u67b6\u6784\u548c\u4f18\u5316\u3002

\n

\u5730\u70b9\uff1a\u5317\u4eac\uff0c\u4e0a\u6d77\u4e0e\u6df1\u5733

\n" }, { "author": { "url": "member/leven87", "name": "leven87", "avatar": "https://cdn.v2ex.com/gravatar/b4497986025202f2280dc8497ab80cb7?s=73&d=retro" }, "url": "t/837601", "title": "\u5982\u679c\u5b9e\u73b0 openmpi \u548c cuda \u7f16\u7a0b\u7684\u7ed3\u5408", "id": "t/837601", "date_published": "2022-03-03T02:04:27+00:00", "content_html": "

\u5404\u4f4d V \u53cb\u597d\uff0c\u6211\u521a\u63a5\u89e6 cuda \u7f16\u7a0b\u3002\u73b0\u5728\u53ef\u4ee5\u5b9e\u73b0\u5229\u7528\u5355 cpu \u548c gpu \u6765\u52a0\u901f\u8fd0\u7b97\u3002 \u73b0\u5728\u9700\u8981\u5b9e\u73b0\u591a cpu \u548c gpu \u6765\u8fdb\u4e00\u6b65\u52a0\u901f\u8fd0\u7b97\uff0c \u770b\u7f51\u4e0a\u4f8b\u5b50\uff0c\u9700\u8981\u7528\u5230 openmpi, \u8fd8\u8981\u5f00\u542f\u5b83\u7684 cuda \u652f\u6301\u3002 \u8bf7\u95ee\uff1a\n\u8fd9\u6761\u9053\u8def\u662f\u5426\u6b63\u786e\uff1f\n\u8fd8\u6709\u54ea\u4e9b\u9700\u8981\u6ce8\u610f\u7684\u5730\u65b9\uff0ccuda \u4ee3\u7801\u7684\u4fee\u6539\uff0c\u6216\u8005\u914d\u7f6e\u5565\u7684\uff1f

\n" }, { "author": { "url": "member/gouchaoer", "name": "gouchaoer", "avatar": "https://cdn.v2ex.com/avatar/58d5/5587/189082_large.png?m=1480987620" }, "url": "t/831540", "date_modified": "2022-01-31T12:29:28+00:00", "content_html": "

\u95ee\u4e2a\u6280\u672f\u95ee\u9898\uff0c\u5f00\u59cb\u89c9\u5f97\u5f88\u7b80\u5355\uff0c\u4f46\u662f\u641c\u4e86\u5f88\u4e45\u6ca1\u7ed3\u679c\u3002\u3002\u3002\u6211\u7528 NVIDIA \u7684 nvdec \u628a 4 \u4e2a h264 \u7684\u89c6\u9891\u89e3\u7801\u51fa\u6765\u6210\u4e86 rgba \u7684 4 \u5f20 raw \u56fe\u50cf\uff0c\u8bf7\u95ee\u6211\u600e\u4e48\u628a\u5b83\u8f93\u51fa\u5230\u663e\u5361\u7684 4 \u4e2a dp \u53e3\uff1f\u6700\u597d\u662f NVIDIA \u7684 api \uff0cOpenGL \u554a drm \u4e4b\u7c7b\u7684\u5305\u88c5\u8fc7\u7684\u4e5f\u884c

\n", "date_published": "2022-01-31T12:27:55+00:00", "title": "\u600e\u4e48\u628a\u663e\u5361\u663e\u5b58\u4e2d\u7684 rgba \u56fe\u50cf\u6e32\u67d3\u8f93\u51fa\uff1f", "id": "t/831540" }, { "author": { "url": "member/wangx0102", "name": "wangx0102", "avatar": "https://cdn.v2ex.com/avatar/2964/0b5a/432993_large.png?m=1695724205" }, "url": "t/794158", "date_modified": "2021-08-06T11:44:19+00:00", "content_html": "

\u5bfc\u5e08\u7ed9\u4e86\u4e00\u4e2a\u7a0b\u5e8f\uff0c\u5b9e\u73b0\u4e86\u4e00\u4e2a\u4e2d\u95f4\u4ef6\u53ef\u4ee5\u5b9e\u73b0 CPU \u548c GPU \u8fd0\u7b97\u7684\u8d1f\u8f7d\u5747\u8861\u3002

\n

\u6211\u7684\u521d\u6b65\u60f3\u6cd5\u662f\u628a CUDA \u7a0b\u5e8f\u6253\u5305\u6210 exe \u6216\u8005.so \u5565\u7684\uff0c\u7136\u540e\u7528 Python \u8c03\u7528\uff0c\u4f7f\u7528 Celery \u5b9e\u73b0\u5206\u5e03\u5f0f\u96c6\u7fa4\u3002

\n

\u5e0c\u671b\u5927\u5bb6\u80fd\u6709\u66f4\u597d\u7684\u60f3\u6cd5

\n", "date_published": "2021-08-06T11:38:23+00:00", "title": "\u5982\u4f55\u5b9e\u73b0 CUDA \u7684\u5206\u5e03\u5f0f\u5e76\u884c\u8fd0\u7b97\uff1f", "id": "t/794158" }, { "author": { "url": "member/huzhikuizainali", "name": "huzhikuizainali", "avatar": "https://cdn.v2ex.com/avatar/1869/a390/522912_large.png?m=1752498684" }, "url": "t/775344", "title": "\u6e38\u620f\u672c\u4e0a\u7528 cuda \u662f\u4ec0\u4e48\u4f53\u9a8c\uff1f", "id": "t/775344", "date_published": "2021-05-07T01:48:50+00:00", "content_html": "1 \u3001\u6709\u4eba\u5728\u6e38\u620f\u672c\u4e0a\u7528\u8fc7 cuda \u4e48\uff1f\u4f53\u9a8c\u5982\u4f55\uff1f\u8003\u8651\u5230\u91cd\u91cf\u589e\u5927\uff0c\u5f85\u673a\u53d8\u77ed\uff0c\u53d1\u70ed\u3002\u540c\u7b97\u529b\u7684\u589e\u5f3a\u76f8\u6bd4\u3002\u7efc\u5408\u5229\u5f0a\u5f97\u5931\uff0c\u5e26\u4e2a\u6e38\u620f\u672c\u8dd1 cuda \u662f\u5426\u503c\u5f97\uff1f

2 \u3001\u8003\u8651\u5230\u79fb\u52a8\u529e\u516c\uff0c\u6570\u636e\u5b89\u5168\uff08\u6570\u636e\u548c\u4ee3\u7801\u4e0a\u4f20\u5230\u4e91\u7aef\u4e0d\u592a\u653e\u5fc3\uff09\uff0c\u5982\u679c\u7528\u6e38\u620f\u672c+cuda\uff0c\u662f\u5426\u8fd8\u6709\u5176\u4ed6\u4e1d\u6ed1\u4f53\u9a8c\u7684\u65b9\u6848\uff1f

3 \u3001\u5404\u4f4d\u5927\u795e\u73b0\u5728\u662f\u600e\u4e48\u89e3\u51b3\u79fb\u52a8\u5f00\u53d1\u7684\u5462\uff1f

4 \u3001\u5de5\u4f5c\u5f53\u4e2d\u7528 cuda \u548c\u4e0d\u7528 cuda \u5bf9\u6bd4\uff0c\u51fa\u7ed3\u679c\u7684\u65f6\u95f4\u80fd\u8282\u7701\u591a\u5c11\uff1f\u53ef\u5426\u5206\u4eab\u4e00\u4e0b\u201c\u771f\u5b9e\u201d\u5bf9\u6bd4\u6848\u4f8b\uff1f" }, { "author": { "url": "member/BillHuang", "name": "BillHuang", "avatar": "https://cdn.v2ex.com/avatar/343f/5feb/460327_large.png?m=1610344655" }, "url": "t/743810", "title": "Tesla k20m \u4f7f\u7528\u95ee\u9898", "id": "t/743810", "date_published": "2021-01-11T05:59:26+00:00", "content_html": "\u6709\u4eba\u7528\u8fc7\u8fd9\u663e\u5361\u5417\uff1f\u7f51\u4e0a\u82b1 300 \u6dd8\u5230\u7684\uff0c\u63d2\u4e0a\u540e\uff0c\u5f00\u673a\u4e3b\u677f\u62a5\u663e\u5361\u9519\u4e86\uff0c\u662f\u6211\u64cd\u4f5c\u4e0d\u5bf9\uff0c\u8fd8\u662f\u8fd9\u5361\u6709\u95ee\u9898\uff1f" }, { "author": { "url": "member/Kai", "name": "Kai", "avatar": "https://cdn.v2ex.com/avatar/021b/bc7e/1024_large.png?m=1657391813" }, "url": "t/682893", "title": "CUDA on WSL", "id": "t/682893", "date_published": "2020-06-18T19:40:48+00:00", "content_html": "

https://devblogs.nvidia.com/announcing-cuda-on-windows-subsystem-for-linux-2/

\n" }, { "author": { "url": "member/different", "name": "different", "avatar": "https://cdn.v2ex.com/avatar/94f6/7670/374456_large.png?m=1661142546" }, "url": "t/591013", "title": "\u5173\u4e8e\u4f7f\u7528 GPU \u751f\u6210\u968f\u673a\u6570\uff08cuda/opencl\uff09", "id": "t/591013", "date_published": "2019-08-11T14:16:02+00:00", "content_html": "

\u7531\u4e8e\u7279\u6b8a\u539f\u56e0\uff08\u539f\u56e0\u5f88\u7279\u6b8a\uff08\u624b\u52a8\u72d7\u5934\uff09\uff09\u5e76\u4e0d\u80fd\u4f7f\u7528 cuda \u81ea\u5e26\u7684\u968f\u673a\u51fd\u6570\u3002

\n

\u56e0\u6b64\uff0c\u7ffb\u8f66\u4e86....\u3002

\n

\u76ee\u7684\uff1a\u5728\u4e0d\u4f7f\u7528 cuda \u81ea\u5e26\u7684\u968f\u673a\u51fd\u6570\u524d\u63d0\u4e0b\uff0c\u4f7f\u7528 cuda/opencl \u7684\u4e00\u4e2a\u5185\u6838\u51fd\u6570\u751f\u6210 10000 \u4e2a\u9ad8\u65af\u5206\u5e03\u7684\u968f\u673a\u6570\u3002

\n

\u672c\u4eba\u5df2\u5c1d\u8bd5\u4e00\u4e0b\u6b65\u9aa4\uff1a

\n

1.\u5728 cpu \u751f\u6210 10000 \u7684\u968f\u673a\u51fd\u6570\uff08\u5e94\u8be5\u662f\u7ebf\u6027\u540c\u4f59\u7b97\u6cd5\uff09

\n

2.\u5728 cpu \u4f7f\u7528 The Box \u2013 Muller transform \uff08\u542c\u8bf4\u548c\u7ebf\u6027\u540c\u4f59\u7b97\u6cd5\u4f7f\u7528\u8d77\u6765\u4f1a\u7ffb\u8f66..\uff09\u7b97\u6cd5\u5c06\u6b65\u9aa4 1 \u7684\u968f\u673a\u6570\u8f6c\u6210\u6b63\u6001\u5206\u5e03

\n

3.\u7136\u540e\u68c0\u9a8c\u662f\u5426\u4e3a\u6b63\u6001\u5206\u5e03\uff0c\u7ed3\u679c\u662f\u5bf9\u7684.

\n

4.\u81f3\u6b64\uff0c\u5df2\u7ecf\u751f\u6210\u4e86\u4e00\u4e2a 10000 \u4e2a\u670d\u4ece\u9ad8\u65af\u5206\u5e03\u7684\u968f\u673a\u6570\u5566\uff0c\u5c06\u5176\u4fdd\u5b58\u5230\u6570\u7ec4 a\u3002

\n

\u4e8b\u5b9e\u4e0a\u9700\u8981\u4e0d\u65ad\u751f\u6210\u5e76\u4f7f\u7528\u6570\u7ec4 a\u3002

\n

\u56e0\u6b64\u8003\u8651 GPU

\n

\u5206\u6790\uff1a\u4e0a\u8ff0\u7684 cpu \u4ee3\u7801\u662f\u5e8f\u5217\u8fdb\u884c\u7684\uff0c\u4e5f\u5c31\u662f\u53ea\u6709\u4e00\u4e2a\u968f\u673a\u79cd\u5b50\uff0c\u7136\u540e\u5728\u4e00\u4e2a\u7ebf\u7a0b\u5185\u5b8c\u6210\u4e86 10000 \u4e2a\u968f\u673a\u6570\u7684\u751f\u6210\u3002

\n

\u7136\u540e\u5c06\u4ee3\u7801\u6539\u6539\u653e\u5230 GPU \u4e0a\u9762\u6765\u751f\u6210\u3002(\u76ee\u6807\u662f\u5b9e\u73b0\u4e0e cuda \u7684\u51fd\u6570 curandGenerateNormal(cuda::generator, cudaRand, number, 0.0, 1.0); \u4e00\u6478\u4e00\u6837\u7684\u529f\u80fd)\u3002

\n

\u4e3a\u4e86\u5f97\u5230\u4e0e curandGenerateNormal \u51fd\u6570\u76f8\u540c\u7684\u7ed3\u679c\uff0c\u6211\u5c1d\u8bd5\u6bcf\u4e2a\u5185\u6838\u7ebf\u7a0b\u7ef4\u62a4\u4e00\u4e2a\u79cd\u5b50\uff0c\u4e5f\u5c31\u662f\u6709 10000 \u4e2a\u968f\u673a\u6570\u79cd\u5b50\u3002(\u8c03\u7528\u4e00\u6b21\u5185\u6838\uff0c\u7136\u540e\u6267\u884c\u4e00\u4e07\u4e2a\u7ebf\u7a0b\uff0c\u6bcf\u9694\u7ebf\u7a0b\u4f7f\u7528\u81ea\u5df1\u7684\u79cd\u5b50\u751f\u6210\u4e00\u4e2a\u968f\u673a\u6570\uff0c\u7136\u540e\u7ec4\u5408\u5230\u6570\u7ec4 a \u4e2d)\n\u4f46\u662f\u76ee\u524d\uff0c\u6211\u505a\u4e86\u8bd5\u9a8c\u4e2d\uff0c\u5982\u679c\u6bcf\u4e2a\u5185\u6838\u7ebf\u7a0b\u7ef4\u62a4\u4e00\u4e2a\u79cd\u5b50\uff0c\u6bcf\u4e2a\u7ebf\u7a0b\u7ef4\u62a4 a[i](i \u4e3a\u7ebf\u7a0b id),\u6700\u540e\u7684\u51fa\u6765\u7684\u5e76\u4e0d\u670d\u4ece\u9ad8\u65af\u5206\u5e03\u3002

\n

\u4e5f\u5c31\u662f\u8bf4\uff0c\u7eb5\u5411\u53bb\u770b\u7684\u8bdd\uff08 cpu \u4e32\u884c\uff09\u662f\u53ef\u4ee5\u5f97\u5230\u9ad8\u65af\u5206\u5e03\u7684\u968f\u673a\u6570\uff0c\u6a2a\u5411\u5e76\u4e0d\u884c\u3002

\n

\u4e5f\u5c31\u662f\u8bf4\uff0c\u5047\u5982\u6709 a \u6570\u7ec4\uff0cb \u6570\u7ec4....z \u6570\u7ec4\u4e2d\uff0c\u6bcf\u4e2a\u6570\u7ec4\u81ea\u4e2a\u662f\u9ad8\u65af\u5206\u5e03\uff0c\u4f46\u662f a...z \u4e2d\uff0c\u5404\u53d6\u4e00\u4e2a\u51fa\u6765\uff0c\u7ec4\u5408\u5728\u4e00\u8d77\uff0c\u5e76\u4e0d\u670d\u4ece\u9ad8\u65af\u5206\u5e03\u3002

\n

\u800c\u5982\u679c\u4ece\u76f4\u89c2\u4e0a\u51fa\u53d1\uff0c\u4e0a\u8ff0\u5e94\u8be5\u4e5f\u670d\u4ece\u9ad8\u65af\u5206\u5e03\uff0c\u4f46\u662f\u7531\u4e8e\u968f\u673a\u79cd\u5b50\u7684\u95ee\u9898\uff0c\u53ef\u80fd\u5bfc\u81f4\u5176 a....z \u53ef\u80fd\u6709\u76f8\u5173\u6027\u3002\u5177\u4f53\u539f\u56e0\u6211\u4e5f\u4e0d\u662f\u5f88\u6e05\u695a\u3002

\n

\u4e0d\u77e5\u9053\u8868\u8fbe\u6e05\u695a\u6ca1\uff0c\u5404\u4f4d\u5144\u53f0\u6709\u6ca1\u6709\u4e86\u89e3\u8fc7\u76f8\u5173\u7684\u4fe1\u606f\uff1f

\n

\u4e00\u53e5\u8bdd\u6982\u62ec\u5c31\u662f\uff1acurandGenerateNormal \u51fd\u6570\u76f8\u540c\u7684\u529f\u80fd...

\n

\u6240\u4ee5\u60f3\u95ee\u95ee\u5927\u4f19\u6709\u505a\u8fc7\u76f8\u5173\u7684\u7814\u7a76\u5417\uff1f

\n" }, { "author": { "url": "member/different", "name": "different", "avatar": "https://cdn.v2ex.com/avatar/94f6/7670/374456_large.png?m=1661142546" }, "url": "t/580600", "date_modified": "2019-07-06T09:05:15+00:00", "content_html": "

\u6307\u7684\u662f\u53cc\u7cbe\u5ea6\u3002

\n

\u4e0d\u77e5\u9053\u662f\u4e0d\u662f\u7f16\u8bd1\u7684\u65f6\u5019\u53cc\u7cbe\u5ea6\u9700\u8981\u6dfb\u52a0\u4e00\u4e9b\u5176\u4ed6\u6307\u4ee4\uff1f

\n

\u4e0b\u9762\u662f kernel\u3002

\n

void CSR(int i,unsigned int N,\nunsigned int *xadj,unsigned int *adjncy,\ndouble *dataxx,double *datayy,double *datazz,\ndouble *Cspin,\ndouble *CHDemag,double *CH)

\n

{

\n
if(i < N)\n{\n\tdouble dot[3]={0,0,0};\n\tfor(int n = xadj[i] ; n < xadj[i+1]; n++)\n\t{\n\t\tunsigned int neigh=adjncy[n];\n\t\tprintf(\"%d\\n\",n);\n\t\tprintf(\"%f,%f,%f\\n\",dataxx[n],datayy[n],datazz[n]);\n\t\tdouble val[3] = {dataxx[n],datayy[n],datazz[n]};\n\t\tfor(unsigned int co = 0 ; co < 3 ; co++)\n\t\t{\n\t\t\tdot[co]+=(val[co]*Cspin[3*neigh+co]);\n\t\t}\n\t}\n\tdouble a=CHDemag[3*i];\n\tdouble b=CHDemag[3*i+1];\n\tdouble c=CHDemag[3*i+2];\n\tCH[3*i]=a+dot[0];\n\tCH[3*i+1]=b+dot[1];\n\tCH[3*i+2]=c+dot[2];\n}\n
\n

}

\n

\u901a\u8fc7\u663e\u5361\u53c2\u6570\u6765\u770b\uff0crtx \u5e94\u8be5\u662f\u6ca1\u6709\u53cc\u7cbe\u5ea6\u8ba1\u7b97\u5355\u5143\u7684\u3002\u800c titan v \u7684\u53cc\u7cbe\u5ea6\u5e94\u8be5\u8fd8\u884c\u3002

\n

\u800c\u6211\u8dd1\u7684\u65f6\u5019\uff0ctitan v \u6bd4 rtx \u6162\u4e86\u4e09\u5206\u4e4b\u4e00\u3002\u3002

\n

\u6c42\u89e3

\n", "date_published": "2019-07-06T08:59:28+00:00", "title": "cuda \u8ba1\u7b97 titan v \u4e3a\u4f55\u6bd4 rtx2080ti \u66f4\u6162\uff1f", "id": "t/580600" }, { "author": { "url": "member/Livid", "name": "Livid", "avatar": "https://cdn.v2ex.com/avatar/c4ca/4238/1_large.png?m=1776858751" }, "url": "t/528949", "title": "DeOldify", "id": "t/528949", "date_published": "2019-01-21T00:27:22+00:00", "content_html": "https://github.com/jantic/DeOldify/blob/master/README.md

\u4e00\u4e2a\u6709\u8da3\u7684 CUDA \u9879\u76ee\uff0c\u53ef\u4ee5\u5c06\u9ed1\u767d\u8001\u7167\u7247\u53d8\u6210\u5f69\u8272\u7684\u3002" }, { "author": { "url": "member/Antidictator", "name": "Antidictator", "avatar": "https://cdn.v2ex.com/avatar/33e8/6eb3/201659_large.png?m=1485621704" }, "url": "t/369282", "date_modified": "2018-06-14T05:50:48+00:00", "content_html": "

http://i.imgur.com/4s1hBfN.png

\n

\u4e0d\u6b7b\u5fc3\u95ee\u4e00\u4e0b\u7b14\u8bb0\u672c 1050 \u652f\u6301 cudnn \u5417\uff1f

\n

\u65e2\u7136\u652f\u6301 cuda\uff0c\u600e\u4e48\u4f1a\u4e0d\u6210\u529f\u5462\uff1f

\n

http://i.imgur.com/mT99ID0.jpg

\n", "date_published": "2017-06-18T04:02:43+00:00", "title": "\u4e0d\u6b7b\u5fc3\u95ee\u4e00\u4e0b\u7b14\u8bb0\u672c 1050 \u652f\u6301 cudnn \u5417\uff1f", "id": "t/369282" }, { "author": { "url": "member/OldFinder", "name": "OldFinder", "avatar": "https://cdn.v2ex.com/avatar/3136/7816/191399_large.png?m=1473652706" }, "url": "t/312305", "date_modified": "2016-10-12T11:51:58+00:00", "content_html": "\u76ee\u524d\u9700\u6c42\u662f\u7528\u5230\u56fe\u5f62\u8bc6\u522b\u548c\u6570\u636e\u7684\u6574\u7406\u548c\u7edf\u8ba1\uff0c\u6570\u91cf\u7ea7\u4e5f\u5c31\u662f\u51e0\u5341\u4e07\u6761\u7684\uff0c\u4e0d\u7b97\u5f88\u5927\u3002", "date_published": "2016-10-12T11:07:25+00:00", "title": "Python+CUDA\uff0c\u5927\u5bb6\u6709\u4ec0\u4e48\u63a8\u8350\u7684\u503c\u5f97\u6df1\u5165\u5b66\u4e60\u4e86\u89e3\u7684\u9879\u76ee\u6216\u8005\u8457\u4f5c\u4e48\uff1f", "id": "t/312305" }, { "author": { "url": "member/xiangtianxiao", "name": "xiangtianxiao", "avatar": "https://cdn.v2ex.com/avatar/0220/9919/93403_large.png?m=1422028256" }, "url": "t/297613", "date_modified": "2018-06-14T05:50:45+00:00", "content_html": "

\u6bd4\u5982\u8bf4\u4e09\u5343\u5757\u5de6\u53f3\u7684\u663e\u5361\uff0c GTX 1070 \u62e5\u6709 1920 \u4e2a\u6d41\u5904\u7406\u5355\u5143\uff0c 8G \u663e\u5b58\u3002\nQuadro M2000 \u53ea\u6709 768 \u4e2a\u5355\u5143\uff0c 4G \u663e\u5b58\u3002

\n

\u6e38\u620f\u5361\u7684\u8bf1\u60d1\u592a\u5927\u4e86\u554a\uff0c\u663e\u5b58\u5927\uff0c\u5355\u5143\u591a...\u6211\u77e5\u9053\u4e13\u4e1a\u5361\u5728 CAD \u65b9\u9762\u53ef\u80fd\u6709\u52a0\u6210\uff0c\u4f46\u662f\u4e0d\u77e5\u9053\u5bf9\u4e8e CUDA \u8fd9\u79cd\u5e76\u884c\u8ba1\u7b97\u6709\u6ca1\u6709\u4f18\u5316\uff0c\u6216\u8005\u8bf4\u53ef\u4ee5\u66f4\u52a0\u7a33\u5b9a\uff1f

\n", "date_published": "2016-08-06T13:17:19+00:00", "title": "\u5199 CUDA\uff0c\u4f7f\u7528\u4e13\u4e1a\u5361\u4e0e\u6e38\u620f\u5361\u6709\u4ec0\u4e48\u533a\u522b\uff1f", "id": "t/297613" }, { "author": { "url": "member/Jolly23", "name": "Jolly23", "avatar": "https://cdn.v2ex.com/avatar/8f94/502f/172360_large.png?m=1603016148" }, "url": "t/294997", "date_modified": "2016-07-26T04:59:07+00:00", "content_html": "

\u4e0d\u662f\u7528\u6765\u6253\u6e38\u620f\uff0c\n\u662f\u641e\u6df1\u5ea6\u5b66\u4e60\u7528\u7684\uff08 CUDA \uff0f caffe \uff0f DIGITS \uff09

\n

\u4e4b\u524d\u82b1 30k \u4e00\u5757\u7684\u4ef7\u683c\uff0c\u4e70\u4e86\u51e0\u5757 tesla K40C \uff0c\u88c5 ubuntu \u8dd1\u6df1\u5ea6\u5b66\u4e60\u4e86\uff0c\u8fd0\u7b97\u80fd\u529b\u771f\u662f\u5f3a\u608d\uff0c\u4f46\u662f\u53e6\u4e00\u4f4d\u5bfc\u5e08\u63a5\u53d7\u4e0d\u4e86\u8fd9\u4e2a\u91c7\u8d2d\u4ef7\u683c\uff0c\u53ea\u80fd\u4e70 5k \u5de6\u53f3\u7684\u5361\uff0c\u6c42\u63a8\u8350\uff01\u80fd\u88c5 ubuntu \u5c31\u884c\uff0c\uff08\u4ed6\u4e4b\u524d\u6709\u5757\u6cf0\u5766\uff0c\u88c5 ubuntu \u663e\u793a unknown chipset maxwell \uff09\uff0c\u5e94\u8be5\u662f\u9ea6\u514b\u65af\u97e6\u67b6\u6784\u7684\u5361\u88c5\u4e0d\u4e86 ubuntu \uff0c\u6211\u4e5f\u4e0d\u592a\u6e05\u695a\u5177\u4f53\u60c5\u51b5\u3002

\n

\u8981\u6c42\uff1a\u5728\u8fd9\u5f20\u8868\u91cc\u7684\u5361\nhttps://developer.nvidia.com/cuda-gpus

\n

\u4ef7\u683c 5k \u5de6\u53f3\u5c31\u884c\uff0c\u7ed9\u63a8\u8350\u70b9\uff0c\u8c22\u8c22\u5404\u4f4d

\n", "date_published": "2016-07-26T04:37:59+00:00", "title": "\u6025\u6c42\u63a8\u8350\u4e2a 5k \u4eba\u6c11\u5e01\u5de6\u53f3\u7684\u8fd0\u7b97 GPU\uff0c\u80fd\u88c5 ubuntu \u5c31\u884c\uff0c\u8dd1\u6df1\u5ea6\u5b66\u4e60\u7528\u7684\uff0c\u5fc5\u987b\u5728 nvidia \u8fd0\u7b97\u80fd\u529b\u8868\u91cc\u9762\u7684\u5361", "id": "t/294997" }, { "author": { "url": "member/andrewzhou", "name": "andrewzhou", "avatar": "https://cdn.v2ex.com/avatar/57e7/710f/42926_large.png?m=1668704021" }, "url": "t/286358", "date_modified": "2016-10-12T15:40:34+00:00", "content_html": "\u6700\u8fd1\u8003\u8651\u6362\u5de5\u4f5c\uff08\u6362\u884c\u4e1a\uff09\uff0c\u5e0c\u671b\u8f6c\u5230\u4eba\u5de5\u667a\u80fd\u7b49\u65b0\u9886\u57df\uff0c\u4f46\u662f\u5bf9\u8fd9\u4e9b\u9886\u57df\u5546\u4e1a\u5e94\u7528\u4e0a\u63a5\u89e6\u5f88\u5c11,\u76ee\u524d\u6bd4\u8f83\u4e2d\u610f CUDA \u548c EMC \u5b58\u50a8\u76f8\u5173\u7684\u5de5\u4f5c\u3002\u6c42\u6307\u5bfc\uff1f\r
1 \uff0c\u7b80\u5355\u4e86\u89e3\u4e86\u4e00\u4e0b CUDA/openCL \uff0c\u611f\u89c9\u786e\u5b9e\u6740\u4f24\u529b\u8fc5\u731b\uff0c\u5bf9 ML \u4e4b\u7c7b\u7684\u5e94\u7528\u786e\u5b9e\u6709\u50ac\u5316\u5242\u7684\u4f5c\u7528\uff0c\u4f46\u662f\u76ee\u524d\u771f\u6b63\u4f7f\u7528 OpenCL/CUDA \u4f5c\u4e3a\u5e95\u5c42\u57fa\u7840\u7684\u6210\u529f\u6848\u4f8b\u591a\u5417\uff1f\u672a\u6765\u5e02\u573a\u662f\u5426\u4f1a\u6301\u7eed\u6269\u5f20\u3002\r
2 \uff0c\u5b58\u50a8\u90e8\u5206\u5728\u4e91\u8ba1\u7b97 /\u4eba\u5de5\u667a\u80fd\u672a\u6765\u662f\u4ec0\u4e48\u89d2\u8272\uff1f\u76ee\u524d\u4e3b\u6d41\u5546\u4e1a\u5e02\u573a\u4e0a\u6e38\u6709\u54ea\u4e9b\u4ea7\u4e1a /\u516c\u53f8\uff1f\r
3 \uff0c\u524d\u4e9b\u65e5\u5b50\u611f\u89c9 FPGA/GPU/TPU \u6495 B \u4e86\u4e00\u987f\uff0c\u8fd9\u4e9b\u9ad8\u5927\u6df1\u7684\u9886\u57df\u6211\u4eec\u4e5f\u53ea\u662f\u770b\u5ba2\uff1f\r
\u4e2a\u4eba\u5bf9\u8fd9\u4e9b\u8ba1\u7b97\u7ed3\u6784\u4e0d\u662f\u5f88\u4e86\u89e3\uff0c\u7b80\u5355\u8ba4\u4e3a FPGA \u4f5c\u4e3a\u5e95\u5c42\u4e2d\u95f4\u5b9e\u73b0\u4e86\u6807\u51c6\u7684 OpenCL \u4e0d\u77e5\u9053\u6709\u6ca1\u6709\u5927\u89c4\u6a21\u7684\u5546\u7528\u6848\u4f8b\u3002\u770b\u8d77\u6765\u786e\u5b9e\u4f1a\u6bd4 GPU \u7b49\u6548\u7387\u9ad8\u5f88\u591a\u3002\r
GPU \u5e94\u8be5\u662f\u76ee\u524d\u6210\u529f\u7684\u5546\u4e1a\u6848\u4f8b\uff0c\u4f46\u662f CUDA \u81ea\u5df1\u5b98\u7f51\u4e0a\u8bf4\u5168\u7403\u6709 700+\u8ba1\u7b97\u96c6\u7fa4\uff0c\u8fd9\u4e2a 700 \u4e0d\u662f\u5230\u662f\u4ec0\u4e48\u6982\u5ff5\u3002\r
TPU \u4e0d\u8bf4\u4e86\u6ca1\u6709\u516c\u5f00\u8d44\u6599\uff0c\u4e2a\u4eba\u8ba4\u4e3a\u5e94\u8be5\u5c40\u9650\u6027\u6bd4\u8f83\u5927\u3002\r
\r
\u5c0f\u767d\u6c42\u55b7. :)", "date_published": "2016-06-17T01:29:17+00:00", "title": "OpenCL/CUDA/\u4e91\u5b58\u50a8\u6c42\u725b\u4eba\u6307\u5bfc\uff0c\u5c0f\u4f19\u4f34\u4eec\u6765\u56f4\u89c2 \uff1a\uff09", "id": "t/286358" }, { "author": { "url": "member/dqh3000", "name": "dqh3000", "avatar": "https://cdn.v2ex.com/avatar/1b72/837a/141112_large.png?m=1521240581" }, "url": "t/225518", "date_modified": "2016-03-09T10:45:37+00:00", "content_html": "

\u53d1\u73b0\u662f\u53ef\u4ee5\u7684\uff0c\u7528 canvas \u7684\u6bcf\u4e2a\u50cf\u7d20\u6a21\u62df\u4e00\u4e2a\u77e9\u9635\u7684\u503c\uff0c\u7136\u540e\u5728 fragment shader \u91cc\u9762\u8ba1\u7b97\u5c31\u53ef\u4ee5\u5b9e\u73b0\u77e9\u9635\u4e58\u6cd5\u4e86

\n\n

\u628a float16 \u7684\u8fd1\u4f3c\u7269\u653e\u5728 rgba \u4e2d\uff08 a \u51e0\u4e4e\u4e0d\u80fd\u7528\uff09\uff0c\u7136\u540e\u5230 fragment \u91cc\u9762\u8fd8\u539f\uff0c\u6700\u540e\u8f93\u51fa\u5230\u5c4f\u5e55\u4e0a\u7684\u56fe\u50cf\u5c31\u662f\u8ba1\u7b97\u7ed3\u679c

\n\n

\u5927\u6982\u5728 650m \u4e0b\u9762\u53ef\u4ee5\u6709 150Gflops+\u7684\u6210\u7ee9\uff08 float16 \uff09

\n\n

\u7136\u540e\u6211\u5728\u60f3\u8fd9\u4e1c\u897f\u6709\u4ec0\u4e48\u5375\u7528\u2026\u2026\uff08\u6211\u77e5\u9053\u8fd9\u4e2a\u4e16\u754c\u4e0a\u6709\u4e2a\u65e0\u4eba\u9e1f\u7684 webcl \uff09

\n", "date_published": "2015-10-04T02:45:09+00:00", "title": "\u60f3\u4e86\u60f3\u7528 WebGL \u80fd\u4e0d\u80fd\u505a\u79d1\u5b66\u8ba1\u7b97", "id": "t/225518" }, { "author": { "url": "member/hardware", "name": "hardware", "avatar": "https://cdn.v2ex.com/gravatar/cc4fb30c125148f251f7345709336b55?s=73&d=retro" }, "url": "t/203204", "date_modified": "2016-03-09T10:43:36+00:00", "content_html": "

\u81ea\u5df1\u53ea\u6709\u4e24\u4e2amacbook pro retina\uff0c\u60f3\u5b66\u5b66caffe\u4ec0\u4e48\u7684\uff0c\u73b0\u5728\u5728\u7528parallel desktop\u8dd1linux\uff0c\u611f\u89c9\u633a\u6162\u7684\u3002\u6709\u6ca1\u6709\u4eba\u7528\u5916\u63a5\u663e\u5361\u505aGPU\u7f16\u7a0b\u7684\uff1f\u80fd\u5de5\u4f5c\u4e48\uff1f

\n", "date_published": "2015-07-03T15:59:31+00:00", "title": "\u6709\u4eba\u7528\u96f7\u7535\u8f6c PCI-e \u8bbe\u5907\u5916\u63a5\u663e\u5361\u8dd1\u8fc7 CUDA \u7684\u4e48\uff1f", "id": "t/203204" }, { "author": { "url": "member/haoji", "name": "haoji", "avatar": "https://cdn.v2ex.com/avatar/ec8b/57b0/2879_large.png?m=1697947304" }, "url": "t/87690", "date_modified": "2016-03-09T10:42:06+00:00", "content_html": "\u6700\u8fd1\u5728\u7814\u7a76 GPU CUDA \u7f16\u7a0b\uff0c\u4e0d\u77e5\u9053\u6709\u6ca1\u6709 V2EXer \u5bf9\u8fd9\u65b9\u9762\u6bd4\u8f83\u4e86\u89e3\u7684\uff1f

\u5728\u628a\u57fa\u7840\u7684 Hello\u3001vecAdd \u90a3\u4e9b\u7a0b\u5e8f\u770b\u4e86\u4e4b\u540e\uff0c\u7740\u624b\u5b9e\u73b0\u4e00\u4e2a\u7c7b grep \u547d\u4ee4\uff1b\u7136\u540e\u5229\u7528\u4e00\u4e2a\u5927\u6587\u4ef6\u6d4b\u8bd5\u6027\u80fd\u3002

\u5bf9\u6bd4\u6d4b\u8bd5\u540e\u53d1\u73b0\uff0c\u6211\u7528 CUDA \u5b9e\u73b0\u7684 mygrep \u6bd4 linux \u81ea\u5e26\u7684 grep \u6162\u5f88\u591a\uff1a1m55s v.s. 0m3s

\u6c57\u989c\u2026\u2026

\u6709\u6ca1\u6709\u719f\u6089\u8fd9\u65b9\u9762\u7684\u670b\u53cb\u53ef\u4ee5\u7559\u4e2a\u8054\u7cfb\u65b9\u5f0f\u6307\u70b9+\u63a2\u8ba8\u4e00\u4e0b\uff1f", "date_published": "2013-11-01T06:30:41+00:00", "title": "\u5173\u4e8e GPU CUDA \u7f16\u7a0b\u7684\u4f18\u5316\u95ee\u9898", "id": "t/87690" }, { "author": { "url": "member/fanzeyi", "name": "fanzeyi", "avatar": "https://cdn.v2ex.com/avatar/a9a1/d531/585_large.png?m=1491194258" }, "url": "t/13337", "date_modified": "2016-03-09T10:43:24+00:00", "content_html": "RT..\r\n
\u9171\r\n
\u6211\u7528 OpenCL \u5c45\u7136\u6ca1\u901f\u5ea6..\r\n
\u7528 CUDA \u6700\u5feb \u7528 CPU \u5f88\u6162..", "date_published": "2011-05-22T04:48:43+00:00", "title": "CPU or GPU? CUDA or OpenCL ?", "id": "t/13337" } ] } ubao msn snddm index pchome yahoo rakuten mypaper meadowduck bidyahoo youbao zxmzxm asda bnvcg cvbfg dfscv mmhjk xxddc yybgb zznbn ccubao uaitu acv GXCV ET GDG YH FG BCVB FJFH CBRE CBC GDG ET54 WRWR RWER WREW WRWER RWER SDG EW SF DSFSF fbbs ubao fhd dfg ewr dg df ewwr ewwr et ruyut utut dfg fgd gdfgt etg dfgt dfgd ert4 gd fgg wr 235 wer3 we vsdf sdf gdf ert xcv sdf rwer hfd dfg cvb rwf afb dfh jgh bmn lgh rty gfds cxv xcv xcs vdas fdf fgd cv sdf tert sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf shasha9178 shasha9178 shasha9178 shasha9178 shasha9178 liflif2 liflif2 liflif2 liflif2 liflif2 liblib3 liblib3 liblib3 liblib3 liblib3 zhazha444 zhazha444 zhazha444 zhazha444 zhazha444 dende5 dende denden denden2 denden21 fenfen9 fenf619 fen619 fenfe9 fe619 sdf sdf sdf sdf sdf zhazh90 zhazh0 zhaa50 zha90 zh590 zho zhoz zhozh zhozho zhozho2 lislis lls95 lili95 lils5 liss9 sdf0ty987 sdft876 sdft9876 sdf09876 sd0t9876 sdf0ty98 sdf0976 sdf0ty986 sdf0ty96 sdf0t76 sdf0876 df0ty98 sf0t876 sd0ty76 sdy76 sdf76 sdf0t76 sdf0ty9 sdf0ty98 sdf0ty987 sdf0ty98 sdf6676 sdf876 sd876 sd876 sdf6 sdf6 sdf9876 sdf0t sdf06 sdf0ty9776 sdf0ty9776 sdf0ty76 sdf8876 sdf0t sd6 sdf06 s688876 sd688 sdf86