{"id": 959817, "name": "GPU computational performance per dollar", "unit": "FLOP/s/$", "createdAt": "2024-07-29T11:40:03.000Z", "updatedAt": "2025-09-25T19:31:14.000Z", "coverage": "", "timespan": "", "datasetId": 6640, "columnOrder": 0, "shortName": "comp_performance_per_dollar", "catalogPath": "grapher/artificial_intelligence/2024-07-11/epoch_gpus/epoch_gpus#comp_performance_per_dollar", "descriptionShort": "Graphics processing units (GPUs) are the dominant computing hardware for artificial intelligence systems. GPU performance is shown in [floating-point operations](#dod:flop) operations/second (FLOP/s) per US dollar, adjusted for inflation.", "descriptionProcessing": "- Reporting a time series of AI investments in nominal prices (i.e., without adjusting for inflation) means it makes little sense to compare observations across time; it is therefore not very useful. To make comparisons across time possible, one has to take into account that prices change (e.g., there is inflation).\n- It is not obvious how to adjust this time series for inflation, and we debated it at some length within our team.\n- It would be straightforward to adjust the time series for price changes if we knew the prices of the specific goods and services that these investments purchased. This would make it possible to calculate a volume measure of AI investments, and it would tell us how much these investments bought. But such a metric is not available. While a comprehensive price index is not available, we know that the cost for some crucial AI technology has fallen rapidly in price.\n- In the absence of a comprehensive price index that captures the price of AI-specific goods and services, one has to rely on one of the available metrics for the price of a bundle of goods and services. In the end we decided to use the US Consumer Price Index (CPI).\n- The US CPI does not provide us with a volume measure of AI goods and services, but it does capture the opportunity costs of these investments. The inflation adjustment of this time series of AI investments therefore lets us understand the size of these investments relative to whatever else these sums of money could have purchased.", "type": "int", "grapherConfigIdETL": "0191c16a-923e-7a88-9773-8dcc949c8ebd", "dataChecksum": "11677805239544766958", "metadataChecksum": "-1415833582406621854", "datasetName": "Trends in Machine Learning Hardware", "updatePeriodDays": 365, "datasetVersion": "2024-07-11", "nonRedistributable": false, "display": {"unit": "FLOP/s/$", "zeroDay": "2000-01-01", "yearIsDay": true, "numDecimalPlaces": 0}, "schemaVersion": 2, "processingLevel": "major", "presentation": {"topicTagsLinks": ["Artificial Intelligence"]}, "dimensions": {"years": {"values": [{"id": 3089}, {"id": 3965}, {"id": 4699}, {"id": 4952}, {"id": 5074}, {"id": 5162}, {"id": 5434}, {"id": 5554}, {"id": 5792}, {"id": 5939}, {"id": 6100}, {"id": 6118}, {"id": 6246}, {"id": 6269}, {"id": 6381}, {"id": 6544}, {"id": 6799}, {"id": 6837}, {"id": 6891}, {"id": 7549}, {"id": 7583}, {"id": 7625}, {"id": 7772}, {"id": 8123}, {"id": 8298}, {"id": 8342}]}, "entities": {"values": [{"id": 309878, "name": "NVIDIA GeForce GTX 280", "code": null}, {"id": 309821, "name": "NVIDIA GeForce GTX 580", "code": null}, {"id": 309986, "name": "NVIDIA Tesla K20c", "code": null}, {"id": 309844, "name": "NVIDIA Quadro K6000", "code": null}, {"id": 370015, "name": "NVIDIA Tesla K40s", "code": null}, {"id": 370016, "name": "NVIDIA GTX Titan Black", "code": null}, {"id": 370020, "name": "NVIDIA Tesla K80", "code": null}, {"id": 370039, "name": "NVIDIA GTX Titan X", "code": null}, {"id": 370023, "name": "NVIDIA M40", "code": null}, {"id": 370018, "name": "NVIDIA P100", "code": null}, {"id": 370034, "name": "NVIDIA P40", "code": null}, {"id": 309694, "name": "NVIDIA Quadro P5000", "code": null}, {"id": 309712, "name": "NVIDIA Quadro P6000", "code": null}, {"id": 309783, "name": "NVIDIA Quadro P4000", "code": null}, {"id": 370042, "name": "NVIDIA Geforce GTX 1080 Ti", "code": null}, {"id": 370036, "name": "NVIDIA Tesla V100 SMX2", "code": null}, {"id": 370025, "name": "NVIDIA Titan V", "code": null}, {"id": 309724, "name": "NVIDIA Quadro RTX 5000", "code": null}, {"id": 309748, "name": "NVIDIA Quadro RTX 6000", "code": null}, {"id": 309734, "name": "NVIDIA GeForce RTX 2080 Ti", "code": null}, {"id": 309695, "name": "NVIDIA Quadro RTX 4000", "code": null}, {"id": 309738, "name": "NVIDIA GeForce RTX 3080", "code": null}, {"id": 309752, "name": "NVIDIA GeForce RTX 3090", "code": null}, {"id": 309754, "name": "NVIDIA RTX A6000", "code": null}, {"id": 370028, "name": "NVIDIA A100", "code": null}, {"id": 370037, "name": "NVIDIA RTX A4000", "code": null}, {"id": 370029, "name": "NVIDIA RTX A5000", "code": null}, {"id": 370041, "name": "NVIDIA GeForce RTX 3090 Ti", "code": null}, {"id": 370038, "name": "NVIDIA GeForce RTX 4080", "code": null}, {"id": 370014, "name": "NVIDIA GeForce RTX 4090", "code": null}, {"id": 370027, "name": "AMD Radeon RX 7900 XTX", "code": null}]}}, "origins": [{"id": 8747, "title": "Trends in Machine Learning Hardware", "description": "This dataset contains performance data for 53 GPUs and AI chips used in machine learning experiments from 2008 to 2023. It includes details on computational performance, memory capacities and bandwidths, and interconnect bandwidths.", "producer": "Epoch AI", "citationFull": "Epoch AI, 'Data on ML GPUs'. Forthcoming.", "urlMain": "https://airtable.com/appDFXXgaG1xLtXGL/shr5STcUv1HmzUIyw/tblNMdPBHaKqJJbW6", "dateAccessed": "2024-07-26", "datePublished": "2024-07-26", "license": {"url": "https://creativecommons.org/licenses/by/4.0/", "name": "CC BY 4.0"}}]}