{ "docId": "019de517-740e-746d-9716-4a0728bc32aa", "docSlug": "64e4d5ca4b0f77867c88fcf9cc9649c0", "documentTitle": "NVIDIA | Investor Presentation Deck | 65 slides", "authorId": "nvidia", "authorName": "NVIDIA", "documentKindSlug": "conference-presentation", "documentKindLabel": "Conference presentation", "sourceTypeSlug": "investor_relations", "sourceTypeLabel": "Investor relations", "presentationDate": "2023-11-01 00:00:00", "orientation": "landscape", "aspectRatio": 1.7777778, "pageNumber": 14, "pageCount": 65, "prevPage": 13, "nextPage": 15, "slideType": "propose_solution", "function": "propose_solution", "density": "overcrowded", "nDataPoints": 8, "notes": null, "elementsJson": null, "metadataConfidence": 0.9, "imagePath": null, "slideHref": "/slides/019de517-740e-746d-9716-4a0728bc32aa/14", "deckHref": "/decks/019de517-740e-746d-9716-4a0728bc32aa", "deckJsonHref": "/decks/019de517-740e-746d-9716-4a0728bc32aa.json", "deckAnchorHref": "/decks/019de517-740e-746d-9716-4a0728bc32aa#slide-14", "components": [ { "bbox": { "h": 0.42, "w": 0.205, "x": 0.55, "y": 0.29 }, "kind": "chart", "text": "8X Increase in GPT-J 6B Inference Performance\n8X\n7X\n6X\n5X\n4X\n3X\n2X\n1X\nOX\nA100\nH100 August H100 TensorRT-LLM", "attrs": { "data_values": [ { "A100": 1 }, { "H100 August": 4 }, { "H100 TensorRT-LLM": 8 } ], "series_names": [ "Performance" ], "x_axis_labels": [ "A100", "H100 August", "H100 TensorRT-LLM" ], "y_axis_labels": [ "0X", "1X", "2X", "3X", "4X", "5X", "6X", "7X", "8X" ] }, "subkind": "bar-vertical", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "b992dab5-39b6-4d1a-9769-b53c51d2caea", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.42, "w": 0.205, "x": 0.77, "y": 0.29 }, "kind": "chart", "text": "4.6X Higher Llama2 Inference Performance\n5X\n4X\n3X\n2X\n1X\nOX\nA100\nH100 August H100 TensorRT-LLM", "attrs": { "data_values": [ { "A100": 1 }, { "H100 August": 2.6 }, { "H100 TensorRT-LLM": 4.6 } ], "series_names": [ "Performance" ], "x_axis_labels": [ "A100", "H100 August", "H100 TensorRT-LLM" ], "y_axis_labels": [ "0X", "1X", "2X", "3X", "4X", "5X" ] }, "subkind": "bar-vertical", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "c12c376b-2a4c-4620-82f2-b774c2d35068", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.02, "w": 0.37, "x": 0.096, "y": 0.519 }, "kind": "paragraph", "text": "5.3X reduction in TCO and 5.6X reduction in energy costs", "attrs": null, "subkind": "bullet", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "39ca0f29-33d5-4348-906a-07419864f5d4", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.07, "w": 0.408, "x": 0.084, "y": 0.339 }, "kind": "paragraph", "text": "NVIDIA developed TensorRT-LLM, an open-source software library that enables customers to more than double the inference performance of their GPUs", "attrs": null, "subkind": "bullet", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "433746c5-bdd4-4e2c-acfb-84ead20ccdea", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.043, "w": 0.408, "x": 0.084, "y": 0.69 }, "kind": "paragraph", "text": "TensorRT-LLM for data centers now publicly available; TensorRT-LLM for Windows in beta", "attrs": null, "subkind": "bullet", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "49cca828-7af2-4940-98b4-e5477bf82e0c", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.07, "w": 0.408, "x": 0.084, "y": 0.44 }, "kind": "paragraph", "text": "TensorRT-LLM on H100 GPUs provides up to an 8X performance speedup compared to prior generation A100 GPUs running GPT-J 6B without the software", "attrs": null, "subkind": "bullet", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "7674514e-3f79-4ec5-8a50-507ad64e6684", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.086, "w": 0.408, "x": 0.084, "y": 0.57 }, "kind": "paragraph", "text": "With TensorRT-LLM for Windows, LLMs and generative AI applications can run up to 4x faster locally on PCs and Workstations powered by NVIDIA GeForce RTX and NVIDIA RTX GPUs", "attrs": null, "subkind": "bullet", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "9b69adf4-3eee-48bb-9a61-a669b0346a65", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.028, "w": 0.39, "x": 0.58, "y": 0.764 }, "kind": "source-note", "text": "Text summarization, variable input/output length, CNN / DailyMail dataset | A100 FP 16 PyTorch eager mode / H100 FP8 | H100 FP8, TensorRT-LLM, in-flight batching", "attrs": null, "subkind": null, "toolName": null, "toolSlug": null, "confidence": null, "componentId": "bf9d52b9-5a9d-467c-80fb-9d538df63b7b", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.045, "w": 0.38, "x": 0.579, "y": 0.209 }, "kind": "title", "text": "TensorRT-LLM Supercharges Hopper Performance\nSoftware optimizations double leading performance", "attrs": null, "subkind": "action-title", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "f0bd5944-e971-424b-965c-c4bcea3f71fe", "frameworkName": null, "frameworkSlug": null }, { "bbox": { "h": 0.139, "w": 0.408, "x": 0.072, "y": 0.08 }, "kind": "title", "text": "New TensorRT-LLM Software More Than Doubles Inference Performance", "attrs": null, "subkind": "headline", "toolName": null, "toolSlug": null, "confidence": null, "componentId": "0e1c2017-e15d-4bad-9ae3-c07c5d2f74d5", "frameworkName": null, "frameworkSlug": null } ], "metrics": [], "tools": [ { "name": "Chartjunk Elimination", "slug": "chartjunk-elimination", "agent": "Designer", "layer": "slide", "matchId": "eb722228-63e1-45eb-8a38-7526cfe802e3", "evidence": "chart/bar-vertical: 8X Increase in GPT-J 6B Inference Performance", "confidence": 0.7 } ], "frameworks": [], "arcBeats": [ { "to": 63, "from": 6, "beatId": "1b4d46fc-6d6e-448a-a6a6-a1dd2c5e67c9", "arcName": "The Sequoia Pitch", "arcSlug": "sequoia-pitch", "beatName": "Financials", "beatSlug": "sequoia-pitch-financials", "evidence": "NVIDIA's financial performance and growth strategy", "position": 3, "confidence": 0.8, "parentBeatName": "Evidence", "parentBeatSlug": "evidence" } ], "loops": [], "imagePathAlt": null, "thumbSrc": null, "thumbSrcAlt": null, "locked": true }