{"id":2338,"date":"2018-08-19T17:02:33","date_gmt":"2018-08-19T09:02:33","guid":{"rendered":"http:\/\/www.animator.idv.tw\/?p=2338"},"modified":"2018-08-19T17:02:33","modified_gmt":"2018-08-19T09:02:33","slug":"nvidia%e7%99%bc%e8%a1%a8%e6%96%b0quadro-rtx%e7%b3%bb%e5%88%97%e9%a1%af%e7%a4%ba%e5%8d%a1","status":"publish","type":"post","link":"https:\/\/www.animator.idv.tw\/?p=2338","title":{"rendered":"Nvidia\u767c\u8868\u65b0Quadro RTX\u7cfb\u5217\u986f\u793a\u5361"},"content":{"rendered":"<p><img decoding=\"async\" loading=\"lazy\" src=\"https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/NVidia_RTX.jpg\" alt=\"\" width=\"1024\" height=\"576\" class=\"aligncenter size-full wp-image-2418\" srcset=\"https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/NVidia_RTX.jpg 1024w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/NVidia_RTX-300x169.jpg 300w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/NVidia_RTX-768x432.jpg 768w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/NVidia_RTX-150x84.jpg 150w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><br \/>\nNvidia\u5728\u4eca\u5e74Siggraph\u5ba3\u5e03\u63a8\u51fa\u65b0\u67b6\u69cb\u7684Quadro\u7cfb\u5217\u5c08\u696d\u5de5\u4f5c\u7ad9\u986f\u5361\u7684\u4e09\u6b3e\u65b0\u7522\u54c1\uff1aQuadro RTX 5000\uff0cQuadro RTX 6000\u548cQuadro RTX 8000\u3002<br \/>\n<iframe loading=\"lazy\" width=\"800\" height=\"405\" src=\"https:\/\/www.youtube.com\/embed\/tjf-1BxpR9c?rel=0\" frameborder=\"0\" allow=\"autoplay; encrypted-media\" allowfullscreen><\/iframe><br \/>\nQuadro RTX \u7cfb\u5217\u65b0\u54c1\u7684\u7279\u5fb4\u5305\u62ec\uff1a<\/p>\n<ul>\n<li>RT Core \u53ef\u505a\u5230\u7269\u4ef6\u548c\u74b0\u5883\u7684\u5373\u6642\u5149\u7dda\u8ffd\u8e64\uff0c\u4e26\u7269\u7406\u4e0a\u7cbe\u78ba\u7684\u9670\u5f71\u3001\u53cd\u5c04\u548c\u6298\u5c04\u53ca\u5168\u57df\u5149\u7167\u3002\n<li>Tensor Core \u53ef\u52a0\u901f\u6df1\u5ea6\u795e\u7d93\u7db2\u8def\u8a13\u7df4\u548c\u63a8\u7406\u3002\n<li>\u5716\u9748\u6d41\u5f0f\u591a\u8655\u7406\u5668\uff08Turing Streaming Multiprocessor\uff09\u67b6\u69cb\u6709 4,608 \u500b CUDA core\uff0c\u53ef\u63d0\u4f9b\u9ad8\u9054 16teraflops \u7684\u8a08\u7b97\u6548\u80fd\uff0c\u4e26\u5217\u904b\u7b97\u6bcf\u79d2 16 \u842c\u5104\u6b21\u6574\u6578\u904b\u7b97\uff0c\u4ee5\u52a0\u901f\u6a21\u64ec\u771f\u5be6\u4e16\u754c\u7684\u7269\u7406\u6a21\u64ec\u3002\n<li>\u5148\u9032\u7684\u7a0b\u5f0f\u5316\u8457\u8272\u6280\u8853\u53ef\u63d0\u9ad8\u8907\u96dc\u8996\u6548\u548c\u5716\u5f62\u5bc6\u96c6\u578b\u5de5\u4f5c\u9ad4\u9a57\u3002\n<li>\u9996\u6b21\u63a1\u7528\u8d85\u5feb\u901f\u7684\u4e09\u661f 16Gb GDDR6 \u8a18\u61b6\u9ad4\uff0c\u652f\u63f4\u66f4\u8907\u96dc\u7684\u8a2d\u8a08\u3001\u5927\u91cf\u5efa\u7bc9\u8cc7\u6599\u96c6\u30018K \u96fb\u5f71\u5167\u5bb9\u7b49\u3002\n<li>NVIDIA NVLink \u53ef\u900f\u904e\u9ad8\u901f\u93c8\u8def\u806f\u901a\u5169\u500b GPU\uff0c\u5c07\u8a18\u61b6\u9ad4\u5bb9\u91cf\u5ef6\u4f38\u81f3 96GB\uff0c\u4e26\u53ef\u900f\u904e\u9ad8\u9054 100GB\/s \u7684\u8cc7\u6599\u50b3\u9001\uff0c\u63d0\u4f9b\u66f4\u9ad8\u6548\u80fd\u3002\n<li>\u63d0\u4f9b\u5c0d USB Type-C \u548c VirtualLink \u786c\u9ad4\u652f\u63f4\u3002\n<li>\u589e\u5f37\u578b\u6280\u8853\u53ef\u63d0\u9ad8VR\u61c9\u7528\u6548\u80fd\uff0c\u5305\u62ec\u53ef\u8b8a\u901f\u7387\u8457\u8272(Variable Rate Shading)\u3001\u591a\u8996\u89d2\u6e32\u67d3(Multi-View Rendering)\u548c VRWorks Audio\u3002\n<\/ul>\n<h5>\u5168\u65b0\u5716\u9748(Turing)\u67b6\u69cb<\/h5>\n<p>RTX\u7cfb\u5217\u662f\u57fa\u65bc\u65b0\u7684\u5716\u9748(Turing)\u67b6\u69cb\uff0c\u88ab\u7a31\u70ba\u4e16\u4e0a\u7b2c\u4e00\u500b\u786c\u9ad4\u652f\u63f4\u5373\u6642\u5149\u8de1\u8ffd\u8e2a(Real Time Ray Tracing)\u7684GPU\u3002\u5167\u5efa\u65b0\u7684RT\u6838\u5fc3(RT Core)\uff0c\u5c08\u70baRaytrace\u52a0\u901f\u4f7f\u7528\uff0c\u7528\u65bc\u52a0\u901fDirectX 12\u4e2d\u65b0\u7684DirectX Raytrace\u7684 Nvidia\u786c\u9ad4\u52a0\u901f\u6280\u8853\u3002<\/p>\n<p>\u6bcf\u500b\u90fd\u6709\u4e00\u7d44RT\u6838\u5fc3(RT core) &#8211; \u65b0\u7684\u5c08\u7528\u5149\u8de1\u8ffd\u8e2a(Raytrace)\u8655\u7406\u5668 &#8211; \u4ee5\u53caNvidia\u7576\u524d\u7684Volta\u67b6\u69cb\u4e2d\u7684\u5169\u7a2e\u6838\u5fc3\u985e\u578b\uff1a\u7528\u65bc\u901a\u7528GPU\u8a08\u7b97\u7684CUDA\u548c\u7528\u65bcAI\u8a08\u7b97\u7684Tensor Core\u3002<\/p>\n<p><img decoding=\"async\" loading=\"lazy\" src=\"https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/Pascal_vs_Turing.jpg\" alt=\"\" width=\"1916\" height=\"815\" class=\"aligncenter size-full wp-image-2408\" srcset=\"https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/Pascal_vs_Turing.jpg 1916w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/Pascal_vs_Turing-300x128.jpg 300w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/Pascal_vs_Turing-768x327.jpg 768w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/Pascal_vs_Turing-1024x436.jpg 1024w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/Pascal_vs_Turing-150x64.jpg 150w\" sizes=\"(max-width: 1916px) 100vw, 1916px\" \/><\/p>\n<p>Nvidia\u8868\u793aTuring\u67b6\u69cb\u6bd4\u65e9\u671f\u4f7f\u7528\u7684\u5927\u591a\u6578\u73fe\u6709Quadro\u5361\u4e2d\u7684Pascal\u67b6\u69cb\u52a0\u901f\u4e86\u5149\u8de1\u8ffd\u8e2a\u64cd\u4f5c25\u500d\u3002<\/p>\n<p>\u65b0\u67b6\u69cbTuring\u652f\u6301\u6df7\u5408\u6e32\u67d3 &#8211; \u5149\u8de1\u8ffd\u8e2a(Raytrace)\u548c\u5149\u67f5\u5316(Rasterize)\u7684\u6df7\u5408\uff0c\u65e8\u5728\u9032\u4e00\u6b65\u52a0\u901f\u8907\u96dc\u7684\u5de5\u4f5c &#8211; \u8207Nvidia\u8072\u7a31\u6bd4Pascal\u63d0\u9ad86\u500d\u5149\u67f5\u5316(Rasterize)\u901f\u5ea6\u3002<\/p>\n<h5>RT\u6838\u5fc3 + Tensor\u6838\u5fc3<\/h5>\n<p>\u65b0Turing\u7684RT\u6838\u5fc3\u63d0\u4f9b\u4e8610Giga Rays\/sec\u7684\u901f\u5ea6\uff0c\u8207\u524d\u4e00\u4ee3Pascal\u7684\u67b6\u69cb\u76f8\u6bd4\uff0c\u5728Raytrace\u7684\u6548\u7387\u4e0a\u63d0\u5347\u4e8625\u500d\uff0c\u4e26\u53ef\u8207\u5149\u67f5\u5316\u6df7\u5408\u6e32\u67d3\u3002\u5728\u4e4b\u524dNvidia GTC 2018\u8207Microsoft\u5c55\u793aDirectX 12 Raytrace\u7684DXR\u6280\u8853\u6642\uff0c\u540c\u6642\u767c\u8868RTX\u7684Game Raytrace\u6280\u8853\u6642\u5c31\u6709\u63d0\u5230\uff0c\u5c07\u6703\u6709GPU\u786c\u9ad4\u7684\u914d\u5408\uff0c\u5c0dRaytrace\u505a\u7279\u5225\u7684\u52a0\u901f\u3002\u7576\u6642DXR\u7a31\u672a\u4f86\u652f\u63f4\u4f7f\u7528Nvidia Volta\u67b6\u69cb\u9084\u5ba3\u7a31\u4f7f\u7528Hardware + Software (RTX)\u7684\u6a21\u5f0f\u3002\u9019\u6b21\u5247\u662f\u5c07RTX\u76f8\u95dc\u898f\u683c\u66f4\u660e\u78ba\u5316\uff0c\u76f4\u63a5\u5c07\u4f7f\u7528Hardware\u7684RT\u6838\u5fc3\u4f86\u8ca0\u8cacRaytrace\u7684\u904b\u7b97\u3002\u64daNvidia\u8868\u793a\u65b0\u7684\u8655\u7406\u67b6\u69cb\u53ef\u4ee5\u52a0\u901f\u5149\u7dda\u8207\u4e09\u89d2\u9762\u7684\u4ea4\u96c6(ray-triangle intersection)\u6aa2\u67e5\u548c\u5728\u73fe\u4eca\u5e38\u898b\u7684\u7528\u4f86\u5132\u5b58Raytracing\u7684\u8cc7\u6599\u7d50\u69cbbounding volume hierarchy (BVH)\u7684\u64cd\u4f5c\u3002<br \/>\n<div id=\"attachment_2412\" style=\"width: 2410px\" class=\"wp-caption aligncenter\"><img aria-describedby=\"caption-attachment-2412\" decoding=\"async\" loading=\"lazy\" src=\"https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/gdc_2018_pre_brief_deck.png\" alt=\"\" width=\"2400\" height=\"1350\" class=\"size-full wp-image-2412\" srcset=\"https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/gdc_2018_pre_brief_deck.png 2400w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/gdc_2018_pre_brief_deck-300x169.png 300w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/gdc_2018_pre_brief_deck-768x432.png 768w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/gdc_2018_pre_brief_deck-1024x576.png 1024w, https:\/\/www.animator.idv.tw\/wp-content\/uploads\/2018\/08\/gdc_2018_pre_brief_deck-150x84.png 150w\" sizes=\"(max-width: 2400px) 100vw, 2400px\" \/><p id=\"caption-attachment-2412\" class=\"wp-caption-text\">GDC 2018\u6642\u767c\u8868\u7684RTX\u6280\u8853\u5c55\u793a<\/p><\/div><br \/>\nTuring\u67b6\u69cb\u4e5f\u5167\u5efa\u7d50\u5408\u4e86Volta\u7684Tensor\u6838\u5fc3\u3002Tensor\u6838\u5fc3\u662f\u8a31\u591aNVIDIA\u8a08\u5283\u7684\u91cd\u8981\u65b9\u9762\uff0c\u9664\u4e86\u52a0\u901f\u5149\u8de1\u8ffd\u8e2a\u672c\u8eab\u4e4b\u5916\uff0cNVIDIA\u7684\u5176\u4ed6\u5de5\u5177\u9084\u5305\u62ec\u901a\u904e\u4f7f\u7528AI Denoiser\u4f86\u6e05\u7406\u5716\u50cf\u4ee5\u6e1b\u5c11\u5834\u666f\u4e2d\u6240\u9700\u7684\u5149\u7dda\u91cf\uff0c\u9019\u662fTensor\u6838\u5fc3\u6240\u64c5\u9577\u7684\u3002\u6240\u4ee5\u96d6\u7136\u4e0d\u662fSiggraph\u4eba\u7fa4\u7684\u4e3b\u8981\u95dc\u6ce8\u9ede\uff0c\u4f46\u662f\u8b49\u5be6\u4e86NVIDIA\u6700\u5f37\u5927\u7684\u795e\u7d93\u7db2\u7d61\u786c\u9ad4\u5373\u5c07\u9032\u5165\u5230GPU\u66f4\u5ee3\u6cdb\u7684\u61c9\u7528\u3002<\/p>\n<p>Turing\u67b6\u69cb\u63d0\u4f9b\u4e86\u66f4\u5ee3\u6cdb\u7684\u7cbe\u5ea6\u63a7\u5236\u3002Tensor\u6838\u5fc3\u66f4\u652f\u63f4INT8\u751a\u81f3INT4\u7cbe\u5ea6\u3002\u5b83\u5011\u5206\u5225\u6bd4FP16\u5feb2\u500d\u548c4\u500d\uff0c\u53ef\u4ee5\u5728\u4e0d\u9700\u8981\u9ad8\u7cbe\u5ea6\u7684\u5de5\u4f5c\u8ca0\u8f09\u4e2d\u5be6\u73fe\u986f\u8457\u7684\u52a0\u901f\uff0c\u53ef\u4ee5\u66f4\u6709\u6548\u7684\u5c07\u6f14\u7b97\u6700\u4f73\u5316\u4ee5\u63d0\u5347\u8a08\u7b97\u6548\u7387\u3002<\/p>\n<p>\u70ba\u4e86\u66f4\u597d\u5730\u5229\u7528\u5149\u8de1\u8ffd\u8e2a\u548c\u5c08\u696d\u6df1\u5ea6\u5b78\u7fd2\u8edf\u4ef6\u4e4b\u5916\u7684Tensor\u6838\u5fc3\uff0cNVIDIA\u5c07\u63a8\u51fa\u4e00\u500bSDK\uff0cNVIDIA NGX\uff0c\u5c07\u795e\u7d93\u7db2\u7d61\u96c6\u6210\u5230\u5716\u50cf\u8655\u7406\u4e2d\u3002NVIDIA\u6b63\u5728\u8a2d\u60f3\u4f7f\u7528\u795e\u7d93\u7db2\u7d61\u548cTensor\u6838\u5fc3\u9032\u884c\u984d\u5916\u7684\u5716\u50cf\u548c\u8996\u8a0a\u8655\u7406\u3002\u4e5f\u5373\u5c07\u63a8\u51fa\u4f7f\u7528\u8a72\u6280\u8853\u6240\u767c\u5c55\u51fa\u7684\u6df1\u5ea6\u5b78\u7fd2\u6297\u92f8\u9f52(Deep Learning Anti-Aliasing DLAA)\u7b49\u65b9\u6cd5\u3002<\/p>\n<h5>\u8655\u7406\u67b6\u69cb\u7684\u65b0\u6280\u8853<\/h5>\n<p>\u65b0\u7684Turing\u4f7f\u7528\u4e86Volta\u67b6\u69cb\u7684\u4e00\u500b\u65b0\u7684\u65b9\u6cd5\uff0c\u5c08\u7528\u7684INT\u6574\u6578\u6838\u5fc3(INT core)\u3002\u5b83\u5c07\u6574\u6578\u6838\u5fc3\u5206\u96e2\u6210\u81ea\u5df1\u7684\u5340\u584a\uff0c\u800c\u4e0d\u662f\u6d6e\u9edeCUDA\u6838\u5fc3\u7684\u4e00\u90e8\u5206\u3002\u9019\u65b9\u9762\u7684\u512a\u52e2\uff0c\u81f3\u5c11\u8207\u6211\u5011\u5728Volta\u4e2d\u770b\u5230\u7684\u4e00\u6a23\uff0c\u662f\u5b83\u52a0\u901f\u4e86\u4f4d\u5740\u7684\u751f\u6210(address generation)\u548c\u878d\u5408\u4e58\u6cd5\u52a0\u6cd5(Fused Multiply Add FMA)\u6027\u80fd\u3002<\/p>\n<p>\u5feb\u901fFP16\uff0c\u5feb\u901f\u6253\u5305\u6578\u5b78\u4ee5\u53ca\u5c07\u591a\u500b\u8f03\u5c0f\u7684\u64cd\u4f5c\u6253\u5305\u5230\u4e00\u500b\u8f03\u5927\u7684\u64cd\u4f5c\u4e2d\u7684\u5176\u4ed6\u65b9\u6cd5\uff0c\u90fd\u662f\u5728\u6469\u723e\u5b9a\u5f8b\u6e1b\u901f\u6642\uff0c\u63d0\u9ad8GPU\u6027\u80fd\u7684\u95dc\u9375\u7d44\u6210\u90e8\u5206\u3002\u901a\u904e\u50c5\u6839\u64da\u9700\u8981\u4f7f\u7528\u5927\uff08\u7cbe\u78ba\uff09\u7684\u6578\u64da\u985e\u578b\uff0c\u53ef\u4ee5\u5c07\u5b83\u5011\u6253\u5305\u5728\u4e00\u8d77\u4ee5\u5728\u76f8\u540c\u7684\u6642\u9593\u6bb5\u5167\u5b8c\u6210\u66f4\u591a\u7684\u5de5\u4f5c\u3002\u9019\u53cd\u904e\u4f86\u5c0d\u65bc\u795e\u7d93\u7db2\u7d61\u63a8\u7406\u5c24\u5176\u91cd\u8981\uff0c\u4f46\u5728\u904a\u6232\u958b\u767c\u4e2d\u4e5f\u8d8a\u4f86\u8d8a\u91cd\u8981\uff0c\u56e0\u70ba\u4e26\u975e\u6240\u6709\u8457\u8272\u5668\u7a0b\u5e8f\u90fd\u9700\u8981FP32\u7cbe\u5ea6\uff0c\u4e26\u4e14\u964d\u4f4e\u7cbe\u5ea6\u53ef\u4ee5\u63d0\u9ad8\u6027\u80fd\uff0c\u4e26\u6e1b\u5c11\u5bf6\u8cb4\u7684\u8a18\u61b6\u9ad4\u983b\u5bec\u548c\u5bc4\u5b58\u5668\u6587\u4ef6(register)\u7684\u4f7f\u7528\u91cf\u3002<\/p>\n<p>\u5728Volta\u4e2d\uff0c\u9019\u8868\u73fe\u70baFP16\u64cd\u4f5c\u901f\u5ea6\u662fFP32\u901f\u7387\u76842\u500d\uff0cINT8\u64cd\u4f5c\u901f\u5ea6\u662fINT32\u901f\u7387\u76844\u500d\u3002\u96d6\u7136Turing\u5728\u65b0\u805e\u7a3f\u6c92\u63d0\u53ca\uff0c\u4f46\u662f\u56e0\u70baTensor\u6838\u5fc3\u5df2\u7d93\u652f\u6301\u9019\u500b\u6982\u5ff5\uff0c\u4e5f\u8a31\u5728CUDA\u4e5f\u652f\u6301\u9019\u6a23\u7684\u64cd\u4f5c\u3002<\/p>\n<p>\u65b0\u805e\u7a3f\u4e2d\uff0c\u63d0\u5230\u4e86\u5c0d\u53ef\u8b8a\u901f\u7387\u8457\u8272(Variable Rate Shading)\u7684\u652f\u6301\u3002\u9019\u662f\u4e00\u7a2e\u76f8\u5c0d\u5e74\u8f15\u4e14\u5373\u5c07\u63a8\u51fa\u7684\u5716\u5f62\u6e32\u67d3\u6280\u8853\uff0c\u5176\u76f8\u95dc\u8cc7\u8a0a\u6709\u9650(\u7279\u5225\u662f\u95dc\u65bcNVIDIA\u5982\u4f55\u5be6\u73fe\u5b83)\u3002\u4f46\u5728\u975e\u5e38\u9ad8\u7684\u5c64\u9762\u4e0a\uff0c\u5b83\u807d\u8d77\u4f86\u50cf\u662f\u4e0b\u4e00\u4ee3NVIDIA\u7684\u591a\u5206\u8fa8\u7387\u8457\u8272(multi-res shading)\u6280\u8853\uff0c\u5b83\u5141\u8a31\u958b\u767c\u4eba\u54e1\u4ee5\u5404\u7a2e\u6709\u6548\u5206\u8fa8\u7387\u6e32\u67d3\u5c4f\u5e55\u7684\u4e0d\u540c\u5340\u57df\uff0c\u4ee5\u4fbf\u5c07\u54c1\u8cea(\u548c\u6e32\u67d3\u6642\u9593)\u96c6\u4e2d\u5230\u6700\u6709\u6548\u7684\u5340\u57df\u3002<\/p>\n<h5>\u652f\u63f4DDR6<\/h5>\n<p>\u65b0\u7684Turing\u652f\u63f4DDR6\u8a18\u61b6\u9ad4\u3002\u7b2c\u4e00\u4ee3GDDR6\u6bcf\u500b\u8173\u4f4d\u8a18\u61b6\u9ad4\u983b\u5bec\u9ad8\u905416Gbps\uff0c\u662fNVIDIA\u65b0\u4e00\u4ee3GDDR5\u986f\u5361\u76842\u500d\uff0c\u6bd4NVIDIA\u6700\u65b0\u7684GDDR5X\u5361\u5feb40%\u3002NVIDIA\u5df2\u7d93\u8b49\u5be6\uff0c\u7b2c\u4e00\u6b3eTuring Quadro\u986f\u5361\u5c07\u4ee514Gbps\u7684\u901f\u5ea6\u904b\u884cGDDR6\uff0c\u5c0d\u65bc\u5178\u578b\u7684256\u4f4dGPU\u4f86\u8aaa\uff0cNVIDIA\u53ef\u4ee5\u4f7f\u7528\u6a19\u6e96\u76848\u500b\u6a21\u7d44\uff0c\u4e26\u7372\u5f9716GB\u7684\u7e3d\u5bb9\u91cf\uff0c\u5982\u679c\u4f7f\u7528clamshell mode\uff0c\u751a\u81f3\u53ef\u4ee5\u9054\u523032GB\u3002<\/p>\n<table background=\"#F00\" cellspacing=\"1\" cellpadding=\"1\" width=\"100%\">\n<tr>\n<td colspan=7 align=\"center\" style=\"background:#444\">Quadro RTX Specification Comparison<\/td>\n<\/tr>\n<tr>\n<td style=\"background:#000;vertical-align:middle;text-align:center;padding-left:0px;padding-right:0px;\">GPU<\/td>\n<td style=\"background:#000;vertical-align:middle;text-align:center;padding-left:0px;padding-right:0px;\">Memory<\/td>\n<td style=\"background:#000;vertical-align:middle;text-align:center;padding-left:0px;padding-right:0px;\">Memory(NVLink)<\/td>\n<td style=\"background:#000;vertical-align:middle;text-align:center;padding-left:0px;padding-right:0px;\">Raytracing bandwidth<\/td>\n<td style=\"background:#000;vertical-align:middle;text-align:center;padding-left:0px;padding-right:0px;\">CUDA core<\/td>\n<td style=\"background:#000;vertical-align:middle;text-align:center;padding-left:0px;padding-right:0px;\">Tensor core<\/td>\n<td style=\"background:#000;vertical-align:middle;text-align:center;padding-left:0px;padding-right:0px;\">Price<\/td>\n<\/tr>\n<tr>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">RTX 8000<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">48GB<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">96GB (NVLink x 2)<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">10 Gagarays\/sec<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">4608<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">576<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">$10000<\/td>\n<\/tr>\n<tr>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">RTX 6000<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">24GB<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">48GB (NVLink x 2)<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">10 Gagarays\/sec<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">4608<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">576<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">$6300<\/td>\n<\/tr>\n<tr>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">RTX 5000<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">16GB<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">32GB (NVLink x 2)<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">6 Gagarays\/sec<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">3072<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">384<\/td>\n<td style=\"background:#333;vertical-align:middle;text-align:center;padding:0px\">$2300<\/td>\n<\/tr>\n<\/table>\n<table background=\"#F00\" cellspacing=\"1\" cellpadding=\"1\">\n<tr>\n<td colspan=5 align=\"center\" style=\"background:#444\">NVIDIA Quadro Specification Comparison<\/td>\n<\/tr>\n<tr>\n<td align=\"center\" style=\"background:#000\"><\/td>\n<td align=\"center\" style=\"background:#000\">RTX 8000<\/td>\n<td align=\"center\" style=\"background:#000\">GV100<\/td>\n<td align=\"center\" style=\"background:#000\">P6000<\/td>\n<td align=\"center\" style=\"background:#000\">M6000<\/td>\n<tr>\n<td align=\"center\" style=\"background:#333\" width=\"150px\">CUDA Cores<\/td>\n<td align=\"center\" style=\"background:#033\">4608<\/td>\n<td align=\"center\" style=\"background:#033\">5120<\/td>\n<td align=\"center\" style=\"background:#033\">3840<\/td>\n<td align=\"center\" style=\"background:#033\">3072<\/td>\n<tr>\n<td align=\"center\" style=\"background:#333\">Tensor Cores<\/td>\n<td align=\"center\" style=\"background:#033\">576<\/td>\n<td align=\"center\" style=\"background:#033\">640\t<\/td>\n<td align=\"center\" style=\"background:#033\">N\/A<\/td>\n<td align=\"center\" style=\"background:#033\">N\/A<\/td>\n<tr>\n<td align=\"center\" style=\"background:#333\">Memory Clock<\/td>\n<td align=\"center\" style=\"background:#033\">14Gbps GDDR6<\/td>\n<td align=\"center\" style=\"background:#033\">1.7Gbps HBM2<\/td>\n<td align=\"center\" style=\"background:#033\">9Gbps GDDR5X<\/td>\n<td align=\"center\" style=\"background:#033\">6.6Gbps GDDR5<\/td>\n<tr>\n<td align=\"center\" style=\"background:#333\">Memory Bus Width<\/td>\n<td align=\"center\" style=\"background:#033\">384-bit<\/td>\n<td align=\"center\" style=\"background:#033\">4096-bit<\/td>\n<td align=\"center\" style=\"background:#033\">384-bit<\/td>\n<td align=\"center\" style=\"background:#033\">384-bit<\/td>\n<tr>\n<td align=\"center\" style=\"background:#333\">VRAM<\/td>\n<td align=\"center\" style=\"background:#033\">48GB<\/td>\n<td align=\"center\" style=\"background:#033\">32GB<\/td>\n<td align=\"center\" style=\"background:#033\">24GB<\/td>\n<td align=\"center\" style=\"background:#033\">24GB<\/td>\n<tr>\n<td align=\"center\" style=\"background:#333\">Architecture<\/td>\n<td align=\"center\" style=\"background:#033\">Turing<\/td>\n<td align=\"center\" style=\"background:#033\">Volta<\/td>\n<td align=\"center\" style=\"background:#033\">Pascal<\/td>\n<td align=\"center\" style=\"background:#033\">Maxwell 2<\/td>\n<tr>\n<td align=\"center\" style=\"background:#333\">Manufacturing Process<\/td>\n<td align=\"center\" style=\"background:#033\">\u53f0\u7a4d\u96fb12nm FFN<\/td>\n<td align=\"center\" style=\"background:#033\">\u53f0\u7a4d\u96fb12nm FFN<\/td>\n<td align=\"center\" style=\"background:#033\">\u53f0\u7a4d\u96fb16nm FFN<\/td>\n<td align=\"center\" style=\"background:#033\">\u53f0\u7a4d\u96fb28nm FFN<\/td>\n<\/tr>\n<\/table>\n<h5>\u652f\u63f4\u7684\u5ee0\u5546\u8207\u8edf\u9ad4<\/h5>\n<p>Nvidia\u8868\u793a\u6709\u8d85\u904e\u4e8c\u5341\u5bb6\u4e3b\u8981\u8edf\u4ef6\u516c\u53f8\u652f\u6301\u65b0\u7684Turing\u67b6\u69cb\uff0c<a href=\"https:\/\/blogs.nvidia.com\/blog\/2018\/08\/13\/turing-industry-support\/\">\u8a73\u898b<\/a>\u3002<\/p>\n<p>Chaos Group\u4e5f\u540c\u6642\u5728Siggraph 2018\u4e0a\u767c\u5e03\u4e86Project Lavina\u7684\u6280\u8853\u5c55\u793a\u3002\u5c55\u793a\u4e86\u5728\u5373\u6642\u5149\u8de1\u8ffd\u8e2a(Real time Ray Tracing)\uff0c\u55ae\u500bQuadro RTX 6000\u4e0a\u4ee525fps\u6e32\u67d33000\u5104\u500b\u4e09\u89d2\u5f62\u7684\u5834\u666f\u3002<\/p>\n<p><iframe loading=\"lazy\" width=\"800\" height=\"450\" src=\"https:\/\/www.youtube.com\/embed\/-NCf85ilN58?rel=0\" frameborder=\"0\" allow=\"autoplay; encrypted-media\" allowfullscreen><\/iframe><\/p>\n<h5>CUDA 10<\/h5>\n<p>CUDA10 \u65b0\u7279\u5fb4\u5305\u62ec\uff1a<\/p>\n<ul>\n<li>\u652f\u63f4\u5716\u9748(Turing)\u67b6\u69cb\u3002\n<li>\u65b0\u589e\u6548\u80fd\u6700\u4f73\u5316\u7a0b\u5f0f\u5eab\uff0c\u5728 16-GPU \u7cfb\u7d71\u9054\u6210\u5f37\u5927\u7684 FFT \u6548\u80fd\u5ef6\u4f38\u3002\n<li>nvJPEG\uff1a\u7528\u65bc\u6df7\u5408 JPEG \u8655\u7406\u7684\u65b0\u7a0b\u5f0f\u5eab\uff0c\u55ae\u500b\u548c\u6279\u91cf\u5f71\u50cf\u89e3\u78bc\u901f\u5ea6\u662f\u539f\u4f86 2 \u500d\u3002\n<li>CUDA \/ Graphics Interop\uff1aCUDA \u548c\u5716\u5f62 API \u4e4b\u9593\u7684\u65b0\u4e92\u4f5c\u696d\u6027\u3002\n<li>\u65b0\u7684\u958b\u767c\u4eba\u54e1\u5de5\u5177\uff1a\u7528\u65bc\u8ffd\u8e64\uff0c\u5206\u6790\u548c\u9664\u932f CUDA \u61c9\u7528\u7a0b\u5f0f\u7684\u65b0 Nsight \u7522\u54c1\u7cfb\u5217\u5de5\u5177\u3002\n<\/ul>\n<h5>Siggraph 2018 Nvidia RTX \u767c\u8868<\/h5>\n<p><iframe loading=\"lazy\" width=\"800\" height=\"405\" src=\"https:\/\/www.youtube.com\/embed\/jY28N0kv7Pk?rel=0\" frameborder=\"0\" allow=\"autoplay; encrypted-media\" allowfullscreen><\/iframe><\/p>\n<p>\u76f8\u95dc\u65b0\u805e<br \/>\n<a href=\"https:\/\/nvidianews.nvidia.com\/news\/nvidia-unveils-quadro-rtx-worlds-first-ray-tracing-gpu\">NVIDIA Unveils Quadro RTX, World\u2019s First Ray-Tracing GPU<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Nvidia\u5728\u4eca\u5e74Siggraph\u5ba3\u5e03\u63a8\u51fa\u65b0\u67b6\u69cb\u7684Quadro\u7cfb\u5217\u5c08\u696d\u5de5\u4f5c\u7ad9\u986f\u5361\u7684\u4e09\u6b3e\u65b0\u7522\u54c1\uff1aQuadro RTX 5000\uff0cQuadro RTX 6000\u548cQuadro RTX 8000\u3002 Quad\u2026<\/p>\n<p class=\"continue-reading-button\"> <a class=\"continue-reading-link\" href=\"https:\/\/www.animator.idv.tw\/?p=2338\">\u7e7c\u7e8c\u95b1\u8b80<i class=\"crycon-right-dir\"><\/i><\/a><\/p>\n","protected":false},"author":1,"featured_media":2420,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[18,2],"tags":[47,113,112,111],"_links":{"self":[{"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=\/wp\/v2\/posts\/2338"}],"collection":[{"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=2338"}],"version-history":[{"count":78,"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=\/wp\/v2\/posts\/2338\/revisions"}],"predecessor-version":[{"id":2425,"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=\/wp\/v2\/posts\/2338\/revisions\/2425"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=\/wp\/v2\/media\/2420"}],"wp:attachment":[{"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2338"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2338"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.animator.idv.tw\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2338"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}